AI skills needs and gaps for MSMEs in the retail sector in Cyprus, Germany, Italy, Poland, and Romania

AI skills needs and gaps for MSMEs in the retail sector in Cyprus, Germany, Italy, Poland, and Romania AI skills needs and gaps for MSMEs in the retail sector in Cyprus, Germany, Italy, Poland, and Romania Report prepared by researchers of DELab, University of Warsaw: Weronika Łebkowska: preparation and analysis of statistical data Michał Paliński: systematic literature review and text mining Katarzyna Śledziewska: general research design and overview, analysis of statistical data, preparation of recommendations Bartosz Ślosarski: research design and overview, qualitative literature review, thematic analysis, text preparation and modification Karol Teodorowicz: preparation of use cases, qualitative literature review, overall research assistance, text preparation Renata Włoch: general research design and overview, structure of the report, text preparation and modification, preparation of recommendations Research support: Agata Komendant-Brodowska: design and carrying out of co-creation workshops Research organization: Michał Szostek: overall research support This publication received funding from the European Union'’s Horizon Europe Research and Innovation programme - Grant Agreement No. 101133847. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them. The reuse of this document is authorised under the Creative Commons Attribution-Noncommercial-Sharealike 4.0 International (CC BY-NC-SA 4.0) licence. This license enables reusers to distribute, remix, adapt, and build upon the material in any medium or format for noncommercial purposes only, and only so long as attribution is given to the creator. If reusers remix, adapt, or build upon the material, they must license the modified material under identical terms. How to cite this report: Włoch, R., Ślosarski, B., Paliński, M., Śledziewska, K., Teodorowicz, K, Łebkowska, W., (2025). AI Skills Needs and Gaps in the Retail Sector in Cyprus, Germany, Italy, Poland, and Romania. Version 2. Zenodo. DOI: 10.5281/zenodo.12793437 Version 2.0 (October 2025). Included executive summaries tailored to specific retail audiences and refined synthesis connecting research findings to AI learning pathways. 1 | 249 TABLE OF CONTENTS How to Read This Report – and What to Take from It 5 Executive Summary for Frontline Retail Workers 7 Executive Summary for Retail Managers 10 Executive Summary for Retail Stakeholders 14 Executive Summary for Vocational Teachers and Students in the Retail Sector 18 Introduction 21 Aim of the Report 21 Main results 22 Overview of the Report 23 Research methodology 26 1.The Potential of AI technologies use in retail 31 Key takeaways 31 1.1. AI uses in retail based on Systematic Literature Review 31 1.2. AI uses in retail based on qualitative literature review 39 1.3. AI functions in retail according to the retail sector stakeholders 59 2. The state of AI adoption in retail sector 69 Key takeaways 69 2.1. The main characteristics of the retail sector in the selected countries 69 2.2. The level of digitilization in the retail sector 75 2.3. Use of advanced digital technologies 80 2.4. The level of digital skills 93 2.5. Characteristics of selected states 103 3. Real-life AI uses in retail companies: use cases and best practices 105 Key takeaways 105 3.1. ALOHAS 106 3.2. Feed. 108 3.3. Grünewald 113 3.4. Kerrigans 115 3.5. Kuchyne Valent 118 3.6. L Cosmetics 121 3.7. Pet Media Group 124 3.8. Procosmet 127 3.9. Purelei 130 3.10. Repeat 133 3.11. Sinnerup 137 3.12. Sortmund 140 3.13. Velasca 144 3.14. Best practices - Innovative Retail Laboratory 147 2 | 249 3.15. Best practices - Footprints AI & Danubius 150 3.16. Best practices - Dare Media 151 3.17. Towards classification of models of AI adoption and integration in retail companies 153 4. AI Skills: Definitions and Classifications 158 Key takeaways 158 4.1. Importance of AI Skills 158 4.2. AI Skills Defined 159 4.3. AI Skills Classifications 161 4.4. AI Skills in Retail: Systematic Literature Review Approach 165 5.AI skills needs: perspective of employers 173 Key takeaways 173 5.1. Data collection 173 5.2. Text mining analysis 174 5.3. Main insights 176 6. AI skills needs and gaps: perspective of the retail sector stakeholders 193 Key takeaways 193 6.1. AI Skills in Retail Companies 193 6.2. AI Skills for Managers 196 6.3. AI Skills for Workers 200 7. Challenges to AI Adoption in Retail 205 Key takeaways 205 7.1. Challenges identified in the academic literature 205 7.2. Challenges identified by retail sector stakeholders 209 8. Recommendations and guidelines for designing an AI Core curriculum for MSMEs in Retail 217 8.1. Practical guidelines for course design 217 8.1. Development of transversal competencies 218 8.2. Development of technical skills 220 8.3. From Research Insights to Learning Pathways for AI Readiness 223 References 227 Annex 234 Appendix 1. Snapshots from job portals included in the job advertisements analysis (section 5) 234 Appendix 2. Co-creation workshop scenario 237 Appendix 3. Interviews and questionnaires scenario 240 Appendix 4. Suggested levels of skills and competences development through INAiR open education resources as presented in the project application 242 Appendix 5. The scope of the Work Package 3 244 Appendix 6. Indicators of the goal realization in WP3 246 Appendix 7. The list of the stakeholders who contributed to the co-creation of insights presented in the report 248 3 | 249 HOW TO READ THIS REPORT – AND WHAT TO TAKE FROM IT The retail sector is complex and diverse. It brings together company owners and managers, trade organisations and unions, frontline employees, and vocational teachers and students. All of them are part of the same transformation, yet each experiences it differently. Artificial intelligence is entering retail not as a single technological revolution, but as a gradual, uneven process that influences strategy, daily work, and learning across the sector. This report captures that complexity. It is based on extensive research conducted in Cyprus, Germany, Italy, Poland, and Romania, combining literature review, text mining, job market analysis, stakeholder workshops, and case studies. Together, these data provide a multi-layered view of how AI is perceived, implemented, and negotiated in the retail sector, highlighting both opportunities and tensions that accompany technological change. The report can be read in multiple ways. It offers evidence, interpretation, and orientation for different groups within the retail ecosystem. Its purpose is not to prescribe solutions but to provide a coherent, evidence-based understanding of the sector’s current position – knowledge that has so far been dispersed across separate studies, industries, and national contexts. For readers from different backgrounds, the same material may offer different insights: ● Managers may focus on how AI affects business operations and decision-making, and on what competences their teams will need to manage technology effectively. ● Retail stakeholders – such as chambers, associations, and trade unions – may find in the data a systemic view of the sector and evidence to guide policy coordination, training initiatives, or collective learning. ● Frontline workers may recognise everyday realities reflected in the findings – the changes in tasks, tools, and expectations that accompany automation. ● Vocational teachers and students can draw from the report an understanding of how retail work is evolving and how education can prepare for that future. These four perspectives are also reflected in the four executive summaries that accompany this report. Each highlights a different way of reading the findings: from business strategy and policy design to workplace experience and education. They are not separate interpretations but complementary lenses, emphasising that digital transformation is a shared process in which every actor has a distinct but interconnected role. At a deeper level, the report identifies recurring areas of competence that are crucial for navigating AI adoption across the sector. These can be understood as three interlinked levels of learning and readiness: ● Understanding and awareness – building basic knowledge of what AI is, how it functions 4 | 249 in retail, and what tasks or decisions it supports. ● Application and integration – learning how to use AI and data-driven tools in practice, interpret their results, and combine them with human experience and judgement. ● Strategy and reflection – linking AI adoption with broader organisational goals, innovation, ethics, and sustainability. Taken together, these insights form a structured knowledge base that provides a foundation for future learning pathways. The report does not yet describe these learning processes in detail, but it points clearly to where they are most needed and what they should encompass. In this sense, the report fills an important gap: it brings together a comprehensive and comparative understanding of how AI transforms retail work, knowledge, and organisation across multiple European contexts. This understanding serves as the groundwork for subsequent phases of the project – where research findings evolve into practical, accessible, and inclusive educational materials designed to help the sector learn, adapt, and grow responsibly in the age of AI. 5 | 249 EXECUTIVE SUMMARY FOR FRONTLINE RETAIL WORKERS Sales staff, customer service, warehouse, logistics, and operations workers. WHY THIS REPORT MATTERS FOR YOU The world of retail is changing. New technologies – especially artificial intelligence (AI) – are becoming part of everyday work: from chatbots and recommendation systems to smart shelves, warehouse scanners, and online customer support. For many workers, these changes can seem worrying. Will AI replace people? Will machines decide what we do? Our research shows something different: AI in retail is not replacing people – it is changing how people work. The goal of the INAiR project is to make sure this change is fair, inclusive, and beneficial for everyone. We studied what’s really happening in retail companies across Cyprus, Germany, Italy, Poland, and Romania – how AI is used, what skills are needed, and how workers experience the transformation. WHAT WE LEARNED AI is already here – and it’s growing fast AI is not only for big tech companies. In retail, it’s used to: ● answer customer questions online through chatbots, ● recommend products based on customer preferences, ● predict which goods will sell best, ● help plan stock levels and deliveries, ● automate simple data or reporting tasks. These tools make work faster and reduce repetitive tasks. But they also change how roles are organized: people spend less time on routine activities and more time on communication, teamwork, and decision-making. Workers are essential to AI success Even the best AI tool needs human knowledge. It’s people who understand customers, products, and daily operations. Our research shows that AI works best when employees know how to use it and can combine human experience with digital tools. In interviews and workshops, many workers said that learning about new technologies helps them feel more confident and valued. They want to understand how these systems work and how to use them to make their jobs easier, not harder. As one participant said: “Talking to a chatbot can be frustrating – but when I know how it works, I can help customers faster instead of fighting the system.” 6 | 249 WHAT SKILLS MATTER NOW Retail companies across Europe are beginning to look for workers who are comfortable with technology and open to learning new tools. You don’t have to be an IT specialist – but some basic skills are becoming essential: ● Digital literacy: being comfortable with computers, tablets, and mobile apps at work. ● Understanding how data is used: knowing that every barcode, purchase, and customer interaction creates information that helps the business run better. ● Communication skills: working with both people and digital systems, such as POS tools, chatbots, or CRM systems. ● Problem-solving: noticing when something goes wrong with an AI tool and reporting it in a constructive way. ● Teamwork and openness to learning: sharing experiences, asking questions, and helping each other adapt to change. We call this combination of attitudes a “digital mindset” – the ability to stay curious, adaptable, and positive about technology, even when it feels new or uncertain. WHAT ARE THE RISKS AND HOW TO MANAGE THEM Change can be difficult. Our research found that many workers worry about losing their jobs, being replaced by machines, or being judged by algorithms. These fears are understandable – and they must be addressed openly. The good news is that most companies introducing AI do not reduce staff. Instead, they reorganize roles so people can focus on higher-value tasks: customer care, problem-solving, personalization, and coordination. Still, AI brings challenges that need attention: ● Some tools are not well designed and make work harder. ● Not all employees receive proper training. ● Communication between managers and teams about digital changes is sometimes weak. That’s why the INAiR project calls for inclusive transformation – where workers are informed, trained, and invited to participate in shaping how technology is used. WHAT YOU CAN DO AS A RETAIL WORKER ● Stay curious. Try to understand what new tools do and how they work. Ask questions. Explore how they can make your job easier. ● Build your digital confidence. Learn basic computer skills, how to use data tools, and how to interact with digital systems. Small steps make a big difference. ● Ask for training. Every company should offer learning opportunities. If you’re not sure how to use a new system, ask your manager for support or suggest taking part in one of the INAiR learning modules. ● Share your experience. Workers know best where technology helps and where it fails. Your feedback is essential for improving AI tools. 7 | 249 ● Protect what matters. Be aware of how data is collected and used. Make sure that automation does not reduce the quality of customer service or the dignity of work. HOW INAIR SUPPORTS WORKERS The INAiR project is developing free, practical learning materials for the retail sector – available online in several languages. These Open Educational Resources (OERs) are designed to help workers like you: ● understand what AI is and how it’s used in retail, ● learn the basics of data and digital tools, ● strengthen problem-solving, communication, and teamwork, ● gain confidence to work with technology instead of against it. You can use them individually or as part of company training. The goal is to make sure that AI supports people, not the other way around. THE MESSAGE FOR ALL RETAIL WORKERS AI is part of the future of retail – but that future is not fixed. It depends on how people and organizations choose to use technology. If workers are included, trained, and respected, AI can make retail jobs more interesting, less repetitive, and better paid. But that requires awareness, communication, and trust between employees, managers, and organizations. You are not just adapting to the digital transformation – you are part of it. Your experience, creativity, and empathy are what make retail human. AI can help you do your job better – but it cannot replace what only people can bring: understanding, connection, and care. 8 | 249 training if benefits are clear. ● Managers struggle to connect AI to business models, rather than treating it as a “tech add-on.” ● Cross-sector cooperation and support structures are missing. The workshops confirmed that the retail sector’s AI transition requires a new social contract – one that balances innovation with inclusion, and competitiveness with fairness. CHALLENGES THAT CALL FOR SYSTEMIC RESPONSE The following systemic barriers were repeatedly identified across research methods: ● Fragmented policy landscape: Different support schemes exist but are not aligned or easily accessible to SMEs. ● Unequal access to digital infrastructure: small retailers often lack affordable data tools and connectivity. ● Skill shortages and training gaps: both managers and workers lack structured pathways to build AI-relevant skills. ● Low trust and transparency: ethical and social concerns remain unresolved, discouraging adoption. ● Dominance of global digital platforms: European retailers risk becoming dependent on non-EU technologies and data ecosystems. These challenges cannot be solved by individual companies alone. They require collective strategies that combine education, policy alignment, and industrial cooperation. RECOMMENDATIONS FOR SECTORAL ORGANISATIONS AND POLICYMAKERS 1. Create shared training infrastructures. Support the roll-out of sector-specific learning programmes like the INAiR AI Core Curriculum and OERs. Chambers and associations can help local retailers access these free resources and tailor them to their needs. 2. Promote skills visibility. Encourage companies to define and communicate the AI-related skills they require. This helps design better curricula and align training supply with market demand. 3. Build trusted knowledge networks. Establish hubs for sharing best practices, case studies, and peer learning. National chambers and trade associations can act as intermediaries between technology providers and small retailers. 4. Support social dialogue and worker participation. Include unions and employee representatives in discussions about automation, retraining, and task redesign. Shared responsibility fosters trust and reduces resistance to change. 5. Ensure fair competition and data sovereignty. Advocate for European solutions and interoperability standards that protect small retailers from over-dependence on dominant global platforms. 6. Integrate sustainability and ethics. Make sure AI adoption aligns with green and social objectives, promoting responsible innovation and decent work. 15 | 249 THE INAIR CONTRIBUTION: TOOLS FOR A COORDINATED ECOSYSTEM The INAiR project offers ready-to-use instruments that can serve as a foundation for collaborative initiatives: ● AI Core Curriculum – a structured learning framework for MSMEs in retail, focusing on transversal, technical, and managerial competences. ● Open Educational Resources (OERs) – multilingual, accessible online modules co-created with retailers and experts. ● Case studies and best practices – 13 examples of AI implementation in European retail MSMEs, illustrating diverse paths to innovation. ● Guidelines for curriculum development – adaptable by sectoral organisations, VET providers, or chambers of commerce. These resources are open-access and can be integrated into national or regional programmes to support capacity building and policy alignment. THE WAY FORWARD AI in retail is not just a technological shift – it is a social and organisational transformation that will shape the future of European commerce. For this transformation to be inclusive, competitive, and human-centred, it must be supported by a strong ecosystem: ● Companies willing to learn and experiment, ● Workers empowered to adapt and grow, ● Institutions and associations ready to coordinate and lead. The INAiR project provides the analytical basis and the practical tools for such cooperation. What is now needed is collective leadership — where trade organisations, chambers, unions, and educational partners work hand in hand to ensure that no retailer, worker, or community is left behind in the age of AI. 16 | 249 EXECUTIVE SUMMARY FOR VOCATIONAL TEACHERS AND STUDENTS IN THE RETAIL SECTOR VET schools, training centres, and learners preparing for retail and customer service careers WHY AI MATTERS FOR VOCATIONAL EDUCATION The retail and service industries are changing faster than ever before. Online sales, self-checkout systems, smart shelves, data-driven marketing, and customer chatbots are no longer the future – they are part of everyday business. Behind most of these tools stands artificial intelligence (AI): technologies that help companies analyze data, predict demand, and improve customer service. AI changes how stores are managed, how teams are organized, and what skills are needed from employees. For vocational schools and training centres, this transformation creates a clear mission: to prepare learners for jobs that are increasingly digital, data-driven, and collaborative. The INAiR project, funded by the European Union, supports this goal by studying what skills the retail sector really needs and creating Open Educational Resources (OERs) that help teachers and students learn about AI in a practical, accessible way. WHAT THE RESEARCH SHOWS The INAiR team, led by researchers from the University of Warsaw, studied AI adoption in five countries: Cyprus, Germany, Italy, Poland, and Romania. The results highlight a major transition that is already underway. ● AI adoption is still low, especially in small retail businesses, but it’s growing. ● Companies use AI mainly to improve customer experience, manage stock, forecast sales, and automate repetitive tasks. ● The biggest barrier is not technology, but skills and confidence – many employees and managers still lack understanding of what AI can do. ● Digital literacy and transversal skills – such as problem-solving, communication, and adaptability – are becoming as important as traditional technical skills. For vocational education, this means that training must combine practical digital tools with critical and creative thinking – helping future workers understand both how AI works and how to work with it responsibly. WHAT SKILLS ARE NEEDED IN THE NEW RETAIL LANDSCAPE The report identifies four main groups of competences that are crucial for the next generation of retail professionals: 17 | 249 ● Digital and AI literacy – using digital devices, apps, and data tools confidently; understanding how AI helps in sales, logistics, and customer support. ● Communication and teamwork – collaborating effectively with colleagues and technology; solving problems that machines can’t. ● Adaptability and learning mindset – being ready to update skills as technology changes. ● Ethical and social awareness – understanding how AI affects customers, jobs, and society, and using it in a responsible way. Teachers can help learners develop these competences by integrating AI-related examples into everyday lessons: from analyzing online customer feedback to experimenting with chatbots or using basic data dashboards to track sales trends. HOW TEACHERS AND TRAINERS CAN USE THE INAIR RESULTS The INAiR report and educational materials provide a ready-to-use foundation for modernizing vocational curricula in retail and related fields. Here’s how they can support you: ● Understand industry needs. The report explains what skills employers look for, which tools they use, and where the biggest knowledge gaps are. ● Bring real-life examples to class. INAiR collected 13 case studies of companies that successfully implemented AI – from family-run stores to online platforms. These stories show that AI is not only for big tech companies but also for small, creative businesses. ● Use open educational resources (OERs). The project is developing online, multilingual learning modules that help teachers introduce topics such as AI basics, digital transformation, customer engagement, and data ethics. ● Encourage co-learning. Students and teachers can learn together – testing new tools, discussing real problems, and reflecting on how technology changes work and skills. This approach helps schools stay connected to the real needs of the labour market and ensures that learning remains practical, engaging, and future-oriented. WHAT THIS MEANS FOR STUDENTS For students preparing to work in retail, hospitality, or customer service, the message is clear: technology will be part of your job, whatever role you choose. But technology will not replace you. It needs your creativity, empathy, and judgment to work well. To be ready, you can: ● learn how digital systems work and what data they use, ● build your confidence with tools like spreadsheets, chatbots, or analytics dashboards, ● stay curious about new applications and ask how they help customers and coworkers, ● think critically about fairness, privacy, and sustainability in technology use. Employers value workers who combine practical skills with curiosity and responsibility. AI tools may change tasks – but people who know how to learn and adapt will always be in demand. 18 | 249 THE ROLE OF VOCATIONAL EDUCATION IN A CHANGING EUROPE The European Union is investing heavily in digital upskilling and AI literacy. Vocational education plays a central role in this agenda, connecting schools, employers, and local communities. By integrating AI-related learning outcomes into their programmes, VET institutions can: ● strengthen their relevance to the labour market, ● build partnerships with local businesses, ● help students access quality jobs in growing digital sectors, ● contribute to a fair, human-centred digital transition. The INAiR project contributes to this effort by creating resources that can be freely used and adapted by teachers, trainers, and educational policymakers across Europe. THE TAKEAWAY: LEARN, ADAPT, AND LEAD AI is reshaping retail, but it is also creating new learning opportunities. Vocational schools can be the bridge between technology and people – giving young professionals the skills and confidence to shape the future of work. Teachers, trainers, and students have a shared mission: to make sure that technology supports human creativity, not replaces it. Learning about AI today is the best way to build secure, meaningful, and innovative careers tomorrow. 19 | 249 INTRODUCTION Aim of the Report The report represents the findings of research conducted under the auspices of the Work Package 3 titled “Analysis of AI Skills Needs and Gaps in Retail” of the “InAIR: Increasing the Uptake of AI in Retail” project, Coordination and Support Action funded by the European Union'’s Horizon Europe Research and Innovation programme under Grant Agreement No. 101133847. The aim of this report is to fulfill the first specific objective of the project, which is to identify the AI skill needs and gaps for MSMEs in the retail sector in Cyprus, Germany, Italy, Poland and Romania. This research aims to guide retail sector companies and their employees through the digital revolution. Our motto is well reflected in the words of one of the retail sector stakeholders who shared their knowledge with us. “ It seems we are always afraid of the unknown, akin to hesitating at the edge of a dark forest. Yet, every breakthrough and revolution involves stepping into that darkness, getting to know it, and then moving beyond. Int_Exp_Poland_9 20 | 249 Main results AI adoption in retail is under-researched Research on the impact of AI on the retail sector's operations is still in its early stages. The existing literature lacks comprehensive studies on the role of AI skills in small and medium-sized enterprises within the retail sector, especially concerning sustainable development and green skills (see Section 1 and Section 4). Retail companies exhibit a low level of AI adoption The retail sector faces distinct challenges in its digital transformation journey, notably with AI integration, due to sectoral fragmentation (97% of companies are micro, small, and medium-sized enterprises), low levels of digitalization (especially in Poland and Romania), a slow rate of AI adoption (particularly in Poland), and insufficient digital skills among employees (see Section 2 and Section 7). Employers do not seek AI skills (yet) The limited progress in AI integration within retail companies is evident from employers' minimal interest in hiring AI-skilled workers (see Section 5). Retail companies adopt AI technologies in diverse ways Companies demonstrate diverse approaches to using and integrating AI technologies (see Section 3 and Section 6). Consequently, the AI Core Curriculum and open educational resources should account for these varied models. Similarly, educational strategies should cater to specific subsets of transversal skills required by retail company managers, such as change management competencies, leadership skills, the ability to formulate growth strategies, and comprehensive knowledge of AI applications and AI project management (see Section 8). Digital mindset is a new transversal skill Digital transformation necessitates a new set of transversal skills and competencies, referred to as a "digital mindset" characterized by openness to innovation and organizational change, collaboration with new technologies and AI, and fundamental technical skills (see Section 8). 21 | 249 Overview of the Report Section 1 of this report examines the potential of AI technology in the retail sector. We conducted a systematic and qualitative literature review, complemented by insights from experts and stakeholders gathered through co-creation workshops and questionnaires. The academic discourse, led by STEM researchers, primarily focuses on technical matters such as decision support systems, natural language processing, and the Internet of Things. However, there is a lack of in-depth studies on AI skills in MSMEs and sustainability. This examination of AI in retail considers both technical and socio-cultural perspectives, focusing on computational advancements and ethical considerations. AI has the potential to transform the retail sector by enhancing decision-making, automating processes, personalizing customer interactions, and optimizing workforce management. Despite these advantages, AI also raises ethical and social concerns. One concern is the potential for AI to be perceived as having mental capacities, which could complicate the roles of humans and machines. Additionally, AI-driven automation poses a risk of job displacement. Therefore, it is crucial to adopt a proactive approach to managing the socio-economic impacts of digitalization. Section 2 analyzes the state of AI adoption in the retail sector, focusing on selected European Union countries: Cyprus, Germany, Italy, Poland and Romania. Over the past decade, the wholesale and retail trade sector has notably expanded, with Romania and Poland showing the highest turnover, while Germany and Italy have lagged behind the EU average, indicating disparities in growth rates across Europe. The sector is dominated by micro, small and medium-sized enterprises (MSMEs), especially in Romania, which hinders competitiveness and operational efficiency, thus slowing digitalization. The retail sector lags behind industries such as information and communication technology (ICT) and professional services in terms of digitalization. German and Cypriot enterprises perform above the EU average, while Polish firms exhibit low levels of digitalization. The variation in e-commerce adoption indicates a disparity in digital integration levels. AI adoption in the retail sector is the lowest among the analyzed sectors, with Germany exhibiting the lowest rate at 10%, and Poland having the highest percentage of firms not using AI. AI applications are commonly used in robotic process automation, speech recognition, and text mining. Cloud technology adoption is most prevalent in Cyprus and Italy, while Romania has the lowest rates. Similarly, the implementation of the Internet of Things (IoT) varies considerably across the region. The adoption of AI is hindered by a lack of specialized knowledge, high costs, and legal uncertainties. Low digital skills, particularly in Poland and Romania, further impede the adoption of technology in the retail sector and MSMEs. In Section 3, we present a series of case studies and best practices illustrating the real-world applications of AI in the retail sector. Outsourced AI tools, frequently marketed as requiring no technical expertise or maintenance, in fact necessitate a fundamental comprehension of a company's operational data and the AI tool's functionality and advantages. The implementation of these tools necessitates the exercise of critical thinking skills to evaluate their performance. In most cases, AI is employed to enhance customer support, underscoring the necessity 22 | 249 for organizations to possess a foundational understanding of AI and its applications. Section 4 examines the definitions and classifications of AI skills. Currently, there is a lack of comprehensive studies on AI skills in MSMEs, particularly in the retail sector, and on sustainability and green skills. Most AI research in the retail sector is conducted by STEM professionals, focusing on technical aspects such as decision support systems, natural language processing, and the Internet of Things. Recent initiatives emphasize literacy, social, and technical competencies. AI applications in the retail sector include business decision support, automation of inventory and supply chain processes, enhancement of customer relationships, monitoring consumer behavior, and automating recruitment procedures. Potential challenges associated with AI in retail include job losses, a lack of suitable skills, trust issues, technical difficulties, and the need for sustainable practices. The absence of a systematic definition of the relationship between AI skills and digital transformation in the retail sector highlights the need for empirical research. The OECD Employment Outlook report proposes a framework for AI skills required for the development, maintenance, and interaction of AI systems. Addressing these gaps could enhance our understanding of the interconnections between AI skills, digital transformation, and sustainability in the retail sector. A research agenda integrating qualitative and quantitative methods and multiple stakeholder perspectives is essential. Section 5 presents an analysis of the future AI skills that employers will require. The key findings include a comparative analysis of job advertisements across selected countries, which revealed several commonalities. These include a pronounced emphasis on proficiency in office software such as Microsoft Office and enterprise resource planning (ERP) systems such as SAP, along with a robust demand for customer relationship management (CRM) skills and expertise in digital marketing (e.g., Google Ads) and e-commerce platforms. Furthermore, the report indicates that basic computer skills and overall digital literacy are considered essential. Nevertheless, it is evident that there are notable differences between countries. Cyprus places a premium on agile methodologies and design tools such as Figma and Adobe Design. Germany, on the other hand, places a strong emphasis on cloud technologies (AWS, Docker), advanced IT infrastructure, and creative skills (Adobe Photoshop, Illustrator). Italy, meanwhile, values data business intelligence, social media management, and hardware installation and support. Poland, finally, places a significant focus on advanced Excel skills and specific SAP modules. Romania, on the other hand, places a strong emphasis on emerging technologies such as VR glasses and electronic processing software. Moreover, Italy and Romania place a strong emphasis on customer interaction and retail front-end competencies, while Cyprus and Germany tend to prioritize technical and backend skills. Section 6 examines the AI skills requirements and deficiencies from the perspective of stakeholders in the retail sector. This section examines the challenges currently facing the industry and considers how artificial intelligence can offer potential solutions. In a series of co-creation workshops held across selected EU countries, participants identified a set of essential skills and competencies that micro, small and medium-sized enterprises (MSMEs) should possess to leverage AI effectively. Furthermore, they identified skills 23 | 249 that are essential for managers and employees. This section presents an overview of the skill sets that were collectively identified during the workshops. The subsequent section delineates the requisite skills and characteristics that MSMEs must possess to successfully implement artificial intelligence (AI). This is followed by a discussion on the critical AI skills that are necessary for both business leaders and employees. Section 7 examines the obstacles impeding the implementation of AI in the retail sector, with a particular focus on the challenges faced by MSMEs. Previous studies have identified numerous barriers to the adoption of AI in the retail sector. These include concerns over job displacement, a scarcity of AI expertise, and issues surrounding trust both internally and with customers. Furthermore, technical obstacles, concerns regarding sustainability, and the absence of industry standards present significant challenges. The co-creation workshop identified several key obstacles faced by MSMEs in the retail sector. These include maintaining efficient supply chains and managing inventory, navigating digital transformation, adapting to AI-driven technological shifts, and effectively addressing communication challenges in customer relationships spanning marketing and service. Participants emphasized the competitive pressures exerted by digital giants from the United States and China, which have a profound impact on the sector's landscape. Furthermore, the workshop highlighted the difficulty that MSMEs encounter in retaining and nurturing a skilled workforce capable of handling AI-driven advancements. The objective of Section 8 is to present recommendations and guidelines for the development of an AI core curriculum that is specifically tailored for MSMEs operating within the retail sector. The discussions held during the co-creation workshops, which were conducted in various EU countries, focused on the planned training initiatives that were designed to enhance the AI skills within the retail sector. This section is based on the insights shared by participants and proposes a structured approach that prioritizes practical advice. The report outlines the specific needs identified during the workshops in order to ensure that the curriculum effectively addresses the challenges and requirements that are unique to MSMEs in the retail sector. 24 | 249 helped to identify the main topics addressed within the field, characterize the leading academic journals, and outline the primary research areas discussing AI applications in retail. 1.1.1. Data Collection In the data collection phase, we used Scopus – a leading citation database renowned for its comprehensive repository of high-quality, peer-reviewed papers (Zhu and Liu, 2020). In the retrieval process, we used the Scopus API to systematically extract the relevant publications and their metadata. The keywords combinations we used were: TITLE-ABS-KEY ( X AND Y ) AND PUBYEAR > 2017 AND PUBYEAR < 2025 Where: X=['sales', 'retail', 'e-commerce'], Y=['AI', 'artificial intelligence']. These keywords were scrutinized within titles, abstracts, author keywords, and ‘topics’ as delineated by the platform. To ensure the reliability of our analysis, we refined our focus exclusively to peer-reviewed articles, specifically limiting our scope to the categories of articles and reviews. For reproducibility purposes we present the exact search query: (TITLE-ABS-KEY ( "AI" ) OR TITLE-ABS-KEY ( "artificial intelligence" ) ) AND ( TITLE-ABS-KEY ( "sales" ) OR TITLE-ABS-KEY ( "retail" ) OR TITLE-ABS-KEY ( "e-commerce" ) ) AND ( DOCTYPE ( ar ) OR DOCTYPE ( re ) ) AND PUBYEAR > 2017 AND PUBYEAR < 2025 This search yielded a total of 1699 publications (after duplicates removal). In the analysis we used full texts of 645 articles for which we had permissions to retrieve pdfs and metadata and abstracts of the full sample. 1.1.2. Results The final dataset includes a total of 1,604 research articles and 95 reviews (scholarly papers summarizing the current state of research). These articles cover research from 105 different countries, as indicated by authors affiliations, demonstrating a wide geographical distribution. The total number of authors is 5532 and they represent 273 distinct institutions, and their work is published in a total of 809 different journals. Average collaboration index (number of authors per article) is 3.42 (single author papers constitute only 13% of total) meaning that research in this area is often done in collaborations. Average citation per document is 19 which is a good score considering that we analyze recent research. In Table 2 we present the most common keywords provided by authors to classify their research. We used top 50 keywords from which we removed the trivial ones (e.g. AI, e-commerce). 31 | 249 Table 2 Top author keywords deep learning covid-19 big data internet of things natural language processing industry 4.0 chatbot social media data mining innovation blockchain digital marketing decision support systems supply chain management sentiment analysis recommender systems decision support system purchase intention digitalization forecasting customer experience automation trust prediction iot personalization explainable ai chatgpt marketing service robots customer service internet of things (iot) classification data analytics neural networks customer satisfaction sustainability customer relationship management supply chain computer vision digital transformation Source: own elaboration In Table 3 we present the subject areas (broad disciplines) to categorize the research on AI and retail. Table 4 shows main journals in which authors publish this research. Table 3 Top subject areas Subject area Number of publications Computer Science 960 Engineering 569 Business, Management and Accounting 504 Social Sciences 312 Decision Sciences 192 32 | 249 Mathematics 185 Economics, Econometrics and Finance 139 Materials Science 111 Psychology 89 Arts and Humanities 83 Environmental Science 81 Source: own elaboration Table 4 Top journals Journal Name Number of Publications Sustainability 34 IEEE Access 31 Technological Forecasting and Social Change 22 Computers and Industrial Engineering 18 Expert Systems with Applications 17 Applied Sciences 17 International Journal of Information Management 16 Mobile Information Systems 15 Electronics (Switzerland) 15 Decision Support Systems 14 Source: own elaboration The selected scientific articles are dominated by those from the fields listed in Table 3, i.e. computer science and engineering, followed by those from management and social sciences. Within the total pool of articles, those with characteristics similar to those of the STEM fields dominate, with fewer articles from the social sciences and humanities. It is noteworthy that the articles listed include those written in the field of environmental sciences. In terms of journals, the most important journals listed in Table 4 are Sustainability, IEEE Access and Technological Forecasting and Social Change. As in the fields, the journals are dominated by journals with a technical focus, while journals in the social sciences and humanities play a minor role or, as in the case of Sustainability, are of low quality. We also detailed the most common words and phrases in the article abstracts as part of our systematic analysis of the existing literature on AI in retail, resulting in Figure 2 and Table 5. Figure 2 Wordcloud of most common terms appearing in articles' abstracts 33 | 249 Source: own elaboration Table 5 Top 3-grams appearing in articles’ abstracts 3-gram Number of occurrences decision support systems 114 natural language processing 58 internet of things 56 ant colony optimization 53 supply chain management 45 artificial intelligence decision 45 artificial intelligence technologies 44 language processing systems 36 classification of information 35 decision making decision 31 learning algorithms learning 31 algorithms learning systems 30 intelligence decision making 28 sales artificial intelligence 27 making decision support 26 support vector machines 26 algorithms machine learning 25 article artificial intelligence 25 artificial intelligence data 23 deep neural networks 23 34 | 249 learning algorithms machine 23 quality of service 22 language processing natural 22 optimization artificial intelligence 21 artificial intelligence commerce 21 artificial intelligence behavioral 20 customer relationship management 19 colony optimization artificial 19 machine learning algorithms 19 intelligence behavioral research 19 particle swarm optimization 19 artificial intelligence consumption 18 intelligence consumption behavior 18 artificial intelligence big 18 intelligence big data 18 human computer interaction 17 sales sentiment analysis 17 long short term 17 short term memory 17 convolutional neural network 17 public relations sales 16 5g mobile communication 16 artificial intelligence customer 16 deep learning learning 16 swarm optimization pso 16 information and communication 16 artificial intelligence electronic 16 design methodology approach 15 algorithm artificial intelligence 15 convolutional neural networks 15 Source: own elaboration Among the top 3 grams listed in Table 5, the top 10 issues worth noting are decision support systems, natural language processing, the Internet of Things, ant colony optimization, supply chain management, artificial intelligence decision, artificial intelligence technologies, language processing systems, classification of information, and decision-making decision. The analysis shows that the literature to date on the application of AI in retail has mainly focused on the technical aspects of applying this technology, its material considerations related to infrastructure and cloud computing, as well as the management aspects of using this type of 35 | 249 technology. Figure 2 presents the most common single words and Table 5 combinations of three words (so called 3-grams) used in articles’ abstracts. The main interest has been limited to picking up the thread of models applied to customer areas, as the cloud with the most frequently used words might suggest. Moreover, we employed Latent Dirichlet Allocation (LDA) using BERTopic to identify topics within the abstracts of articles focused on AI in retail. LDA is a generative statistical model that allows sets of observations to be explained by unobserved groups, effectively uncovering the hidden thematic structure in a collection of documents (Jelodar et al., 2019). BERTopic, on the other hand, leverages BERT embeddings for document representation and clustering, enhancing the interpretability and coherence of the topics discovered (Grootendorst, 2022). By applying LDA with BERTopic, we were able to systematically organize the research articles into distinct topics, facilitating easier navigation and analysis of the literature. The resulting model has been saved in our project repository on Zenodo. This file allows researchers with basic Python skills for an organized review of research and enables browsing articles by specific topics. In Table 6 we present the topics along with the most characteristic keywords and number of articles for which the given topic is a main one. Names of topics were created using ChatGPT4o based on full lists of characteristic keywords. 36 | 249 Table 6 Topic modeling results Topic Name Keywords Article Count AI Chatbots and Consumer Interaction [ai, chatbots, chatbot, consumers] 215 Product Reviews and Sentiment Analysis [reviews, sentiment, product, analysis] 62 IoT and Cloud Computing [iot, cloud, data, smart] 60 Recommendation Systems [recommendation, recommender, recommendations] 55 Digital Trade and Economic Transformation [digital, trade, countries, transformation] 49 Customer Data Mining and Behavior [data, customer, mining, behavior] 43 Fashion Design and Image Processing [fashion, design, image, clothing] 43 Supply Chain Decision Making [decision, supply, chain, inventory] 39 Demand Forecasting and Sales Models [forecasting, demand, models, sales] 37 Agricultural Practices and Poultry Farming [poultry, farmers, agricultural, rural] 35 Service Robots and Customer Acceptance [robots, service, robot, acceptance] 26 Customer Loyalty and Metaverse [customer, metaverse, loyalty, tools] 25 Omnichannel Retail and AI [retail, ai, retailers, omnichannel] 23 Industrial Supply Chain Management [chain, supply, manufacturing, industrial] 22 Customer Churn Prediction [churn, customer, models, prediction] 21 Smart Retail Sector [retail, retailing, smart, sector] 19 Logistics and E-commerce Distribution [logistics, distribution, commerce, cross] 19 Business Marketing and E-commerce Research [commerce, marketing, business, research] 18 Financial Credit Risk and Scoring [credit, risk, financial, scoring] 18 AI in Sales and Selling Techniques [sales, salespeople, ai, selling] 16 Fraud Detection in Transactions [fraud, detection, transactions, fraudulent] 15 User Models in Mobile Commerce [user, model, trained, mobile] 10 Source: own elaboration 37 | 249 Among the themes detailed in the systematic literature review in Table 6, the issue of consumer interaction with chatbots based on artificial intelligence technology emerges as a key area of focus. This includes how consumers engage with AI-powered chatbots and the implications for customer service and satisfaction. In addition, consumer relationships extend to several other critical aspects. Firstly, product reviews and marketing sentiment analysis are highlighted as key issues. This involves the use of AI to analyze consumer feedback and sentiment expressed in product reviews, helping retailers to understand consumer opinions and improve their products and services. Secondly, recommendation systems are identified as another important area. AI-powered recommendation systems enhance the consumer shopping experience by suggesting products based on past purchase behavior and preferences. Consumer data mining and the analysis of consumer shopping behavior also appear as central themes in the literature. These themes explore how AI technologies can process vast amounts of consumer data to identify patterns and trends, which can then be used to tailor marketing strategies and optimize inventory management. Structural issues related to data infrastructure and cloud computing also receive significant attention. These themes delve into the technological backbone required to support AI applications in retail, highlighting the importance of a robust data infrastructure and the role of cloud computing. Finally, supply chain management is another key theme in the literature. The role of AI in optimizing supply chain operations, from inventory management to logistics and distribution, is explored, showing how AI can improve efficiency and reduce costs in the retail sector. 1.2. AI uses in retail based on qualitative literature review In this section, we discuss the challenges of defining artificial intelligence in retail as presented in various literature on the subject. We also highlight how AI is defined in terms of its functions and applications in the retail industry. 1.2.1. Definitions of AI Artificial intelligence is broadly described as the ability of a digital computer or computer-controlled robot to perform tasks typically associated with intelligent beings. The term often refers to efforts to create systems that possess human-like intellectual abilities, such as reasoning, understanding meaning, generalizing, and learning from past experience 38 | 249 (Gordon, 2023: 16). However, as Feher and Katona (2021) point out, artificial intelligence is understood in two general ways: in a technical way and in a socio-cultural way. The dominant current in defining AI is of a technical nature. Technical understanding of AI refers to a category of computational technologies that drive AI advancements, emphasizing the STEM fields – science, technology, engineering, and mathematics – that underpin innovations like robotic automation and machine learning (Feher and Katona, 2021: 1). This approach focuses on using AI technology across various industries, with the development and application of AI serving as the ultimate goal to achieve a competitive advantage (Feher and Katona, 2021: 1). Nevertheless, as the authors point out, a second strand is also emerging in defining AI, focusing on the socio-cultural understanding of AI. The second proposed term focuses on the use of technology as a means rather than an end in itself. This category emphasizes the incorporation of AI technology into social and cultural contexts, considering factors such as human adaptation and ethical issues. The socio-cultural AI perspective is supported by increasing attention to public and corporate policy, cultural norms, and AI ethics, highlighting the broader implications of socio-cultural technology (Feher and Katona, 2021: 2). To account for the social uses of AI in different contexts, including retailing contexts, we should focus on the socio-cultural strand of AI definition. When it comes to defining AI by its social-alike characteristics, Caluori (2023: 4-6) has formulated a set of definitional criteria: ● Learning Ability: learning ability is a key criterion for AI, particularly because machine learning is frequently used as a synonym for AI; ● Human Likeness: the resemblance to human beings is a major factor that makes AI intriguing, both in scientific and popular contexts. This criterion is the most frequently used in AI definitions, although its application varies; ● State of Mind: the evolution of AI can also be seen in the “state of mind” dimension understood as giving technology an intellectual state; ● Solving Hard Problems: the complexity of the problem which AI is able to solve is the least contentious criterion, as most definitions do not consider it essential; ● Successful Solutions of Problems: the criterion of successfulness shows a distribution like the "state of mind" dimension, indicating it is also contentious. The above set of criteria for defining artificial intelligence has implications for how AI technologies are viewed in different social, cultural and economic contexts. The field of retail and studies of the application of AI in retail reflect all of the above criteria. Consider the following definition of AI in retail provided by Giroux et al. (2022: 1028): One technology that companies (..) have started to embrace is AI, which refers to “programs, algorithms, systems and machines that demonstrate intelligence” (Shankar, 2018, p. vi) and is “manifested by machines that exhibit aspects of human 39 | 249 intelligence” (Huang & Rust, 2018, p. 155). With its ability to accurately perform tasks and goals based on external inputs, AI is revolutionizing how companies and organizations create content, make recommendations, and interact within the store (de Ruyter et al., 2018; Haenlein & Kaplan, 2019; Weber & Schütte, 2019). Robots are also promising avenues for frontline services. For the moment, robots are mainly implemented in the manufacturing and delivery of products and services (Ivanov, 2020), but we can expect their growing presence in frontline services. In the definition above, AI is defined as computer programs in the broadest sense that exhibit intelligence and in this respect resemble human intelligence. AI technologies are also defined in terms of their ability to solve problems and achieve goals, i.e. to successfully present solutions to specific problems. In the quote above, the authors also outline the potential applications of AI in retail, with a focus on content production and in-store recommendations. Moreover, various types of chatbots and virtual assistants are being used in the retail sector, capitalizing on their ability to mimic human interactions. Designers can strategically incorporate elements such as small talk, greetings, and conversational transitions to build user trust in the interface and encourage specific behaviors, including self-disclosure and persuasion (Schanke et al., 2021: 6). However, the use of chatbots in customer interactions brings with it additional dilemmas – which are discussed more in the section on co-creation workshops. As described by Giroux et al. (2022: 1030), the research highlights that the attribution of mental capacities to technologies significantly influences moral judgements and notions of responsibility, but they caution against direct comparisons with human-to-human interactions, ultimately emphasizing the peculiarities of the human mind and the challenges of applying human norms and moral responses to machines. It is important to note that we are not dealing with a single artificial intelligence, but many different ones that manifest themselves in specific customer interaction tools. Pantano and Scarpi (2022: 586) presented the following typology of forms of artificial intelligence used in customer interaction: ● Mechanical or operational intelligence: the ability to learn and perform basic and repetitive tasks. ● Thinking: the ability to perform analytical and intuitive tasks (reasoning-based intelligence). ● Emotional or affective: the ability to recognize human emotions and adapt behavior accordingly. ● Self-organizing cooperation: the ability to coordinate with other AI to create a self-managed, autonomous, collaborative network (distributed intelligence). ● Social cognition: the ability to process, store, and apply information about others and behave accordingly. 40 | 249 B. Customer engagement When interacting with a company's sales or communication channels, customers may be willing to benefit from optimized experience, but at the same time distanced from the idea of being guided by fully automated services. Augmented intelligence offers help through problem-solving and personalized recommendations without depriving a client of the sense of control over decision-making. It can contextualize a person's intentions and balance between the sufficiency of chatbots and the need for human support. Whenever suitable and desired, augmented intelligence performs complex analytical as well as the unengaging repetitive tasks, freeing up human labor on both sides of the shopping experience. C. Inventory management Augmented intelligence support can help organize an efficient supply chain. It enables the employees to easily access the detailed information on predicted demand and how the inventory maintenance, as well as other supply-related processes, can be adjusted to meet it in the optimal way. It is able to oversee the complex data and provide context-specific reports without independently interfering with the company’s operations. D. Operations optimisation With augmented intelligence systems aid, a company's productivity is not as limited by its labor force capabilities. It may be easier for the enterprise to scale up its operations, when some of the activities are handled through intelligent solutions. Lower-ranked employees can become responsible for performing more advanced tasks without the need of extensive supervision, which in turn may contribute to increased trust inside the institution. 1.2.3.4. Insight engines Insight engines “apply relevancy methods to describe, discover, organize and analyze data. This allows existing or synthesized information to be delivered proactively or interactively, and in the context of digital workers, customers or constituents at timely business moments. Products in this market use connectors to crawl and index content from multiple sources. They index the full range of enterprise content, from unstructured content such as word processor and video files through to structured content, such as spreadsheet files and database records.” (Gartner, 2024c). Insight engines, backed by algorithms and natural language processing, are able to 47 | 249 process more nuanced and diverse data than traditional search engines, identifying relevant information, patterns and relationships for a real-time decision making. A. Knowledge and insight management Comprising knowledge based on data of different types and origins, insight engines provide relevant real-time information to respond to current trends, gaps and challenges. They can search through the information to check their regulatory compliance and detect possible anomalies. Insights from scientific, technical, legal and market data may efficiently advance a company's R&D (research and development). When all the relevant data is organized and appropriately distributed throughout the company, the decision-makers are able to collaborate and develop strategies based on the shared knowledge. On the other hand, insight engines’ security measures include controlling access and encrypting sensitive supply chain information. B. Customer engagement Insight engines can help companies meet the needs and possibilities of their customers in a proactive manner. Analyzing extensive data on customers’ characteristics and behavior patterns, engines can provide real-time solutions to delivering personalized content at the right moment. Tools powered by insight engines may improve the experience of customers, allowing for efficient self-service and faster order fulfillment. C. Inventory management Apart from providing data security, insight engines can process diverse information in real time and detect disruptions at different stages of the supply chain. Together with insights on the market fluctuations and customers’ activity patterns, this can help the company to update its strategies, optimize stock organization, minimize lead time and quickly adapt to encountered difficulties. D. Operations optimisation Combined data of a company’s operations, its consumers, competitors and the market may become a foundation for major strategies as well as specific problem-solving. Inefficiencies and opportunities can be detected and analyzed to create solutions for streamlined processes. Optimizing resource use not only contributes to cost reduction and overall agility, but also helps the company improve the sustainability of its practices. 48 | 249 1.2.3.5. Natural language processing Natural language processing (NLP) “enables computers to understand, interpret, generate, and transform text. Current state-of-the-art models, such as OpenAI’s GPT-4 and Google’s Gemini, are able to generate fluent and coherent prose and display high levels of language understanding ability” (Maslej et al., 2024, p. 85). NLP allows for the communication between machines and humans, with AI systems interpreting natural language input and generating data accessible to people. Its analytical capabilities include classification of symbols, words and phrases, but also identification of underlying context, intent or emotion of the text’s author. A. Knowledge and insight management NLP enables human agents to easily access knowledge with commands and queries formulated in natural language. With the help of speech recognition and optical character recognition tools (OCR), NLP helps analyze any kind of verbal data. It sorts unstructured bits of information into organized frameworks and can provide visualizations or summaries of complex datasets. Backed by ML or deep learning algorithms, NLP tools are able to identify relevant information in the data and the relationships between them. For example, named entity recognition (NER) is a technology used for finding specific elements in the text, such as targeted topics, proper names or values. B. Customer engagement Through natural language processing, it becomes easier to identify actual and potential customers’ needs. It can analyze not only the lexical construction of the search phrases, reviews or messages, but also the meaning of the words and phrases in context. Due to this, NLP helps recognise the intentions of users, which results in a more accurate depiction of the products or services that may be of their interest. Further processing of the semantic content of texts allows for sentiment analysis. With ML-based classification, the statements are categorized by emotional features and attributed positive, negative or neutral value. That, in turn, provides more comprehensive feedback information and contributes to dynamic and personalized interaction with customers. Well-conducted natural language analysis also allows the automated support system to decide which queries or calls are suitable for chatbot service and which demand a human-consultant’s intervention. C. Operations optimisation Semantic search is also beneficial for company’s employees. With NLP technology it becomes easier to navigate through documentation or website’s content. Along with summary features this can speed up analytical tasks. When it comes to marketing, 49 | 249 natural language processing tools prove to be beneficial in reviewing content and providing suggestions for optimized readability, tone, grammar as well as SEO-compliance. Finally, NLP enables content generation. Large language models contain enough data for the generative AI to produce texts that in many cases match the quality and effectiveness of human writing. Enhancing abilities of human agents in creating and reviewing content to meet company’s policies and SEO strategies reduces the need for a costly labor force and can significantly improve marketing performance. 1.2.3.6. Computer vision Computer vision “allows machines to understand images and videos and create realistic visuals from textual prompts or other inputs” (Maslej et al., 2024, p. 96). Backed by ML algorithms and deep neural networks, computer vision is used for tasks such as facial and object recognition, image classification, tagging, semantic segmentation and even image generation. A. Knowledge and insight management Through this technology, visual representations of a company's resources, including documentation and real-life processes, can be captured, attributed meaning and analyzed for improved, standardized performance. Connected to sensors in stores or warehouses, computer vision detects patterns and anomalies, providing a comprehensive view of a company's operations, highlighting gaps in efficiency or security. B. Customer engagement Computer vision enhances customer experience in new ways, allowing for closer and better informed contact with the virtual representations of products. A person can take or copy a picture of a desired object and upload it to a visual search tool. Image-based queries can be efficient when not enough textual information is initially provided, either at a store or in ordinary life situations. When combined with augmented reality applications or extensions, computer vision makes it possible to visualize a product in a real-life environment, e.g. a person’s room, or perform a virtual try-on of a piece of clothing/equipment. C. Inventory management Computer vision enables constantly updated monitoring of the inventory. It analyzes stock levels and provides information on their compliance with predicted demand or 50 | 249 delivery schedules. When applied to store shelves, this technology can help optimize the placement of products, so that they are exposed in the most efficient way. D. Operations optimisation In-store operations can be tracked to improve the shopping experience and generate more sales. Heat maps and traffic patterns show the areas where customers' presence and attention may be increased, which helps with planning an optimal organization of stores. Computer vision is also used to track suspicious behavior or fraud, both online and offline. Security appliances range from detection of incorrect scanning, counterfeit products, to identifying fake reviews. Object recognition and image classification can become effective tools for a website’s content moderation. 1.2.3.7. Speech recognition Speech recognition tools “interpret human speech and translate it into text or commands. Primary applications are self-service and call routing for contact center applications; converting speech to text for desktop text entry, form filling or voice mail transcription; and user interface control and content navigation for use on mobile devices, PCs and in-car systems. Control of consumer appliances (such as TVs) and toys is also commercially available but not widely used” (Gartner, 2024e). A. Knowledge and insight management Speech recognition, when implemented into any communication system within the company, can collect and analyze metadata about interactions. Whether it is a conversation involving an employee or a customer, the system is able to distinguish patterns of activity (pauses, response-time) as well as emotional tones of different statements. An advanced type of this technology - voice recognition - allows for identification of a person taking part in the talk. B. Customer engagement Insights into clients’ behavior and emotions may prove beneficial for optimizing their experience. A customer can dictate his/her demands or queries and engage in natural-language conversations with minimized loss of time and attention. Data gathered through these interactions enables better personalisation of an offer and automated, easy-to-scale-up support system. In parallel with traditional search engine query processes, companies can improve their answer engine optimisation (AEO) and provide an alternative software for shopping activities. Speech recognition tools reduce 51 | 249 the need for a client’s technical abilities and efforts to search, order and reorder desired products. C. Inventory management Voice picking lets the employees control the supply without time-consuming manual inputs. Through a headset, a warehouse worker can give commands and receive guidance to easily navigate through the stock and be aware of its levels and precise locations. D. Operations optimisation Voice commands provide an additional channel of communication, as well as an easy access to company’s operational data. Identified gaps and slowdowns in interactions may suggest solutions for speeding up the workflow. Since any recorded conversation can be transcribed in real time, the creation of up-to-date documentation becomes automated and full of detail. 1.2.3.8. Chatbots Chatbots are an example of “a domain-specific conversational interface that uses an app, messaging platform, social network or chat solution for its conversations. Chatbots vary in sophistication, from simple, decision-tree-based marketing stunts, to implementations built on feature-rich platforms. They are always narrow in scope. A chatbot can be textor voice-based, or a combination of both” (Gartner, 2024b). A. Knowledge and insight management Engaging in interactions with customers, chatbots may seamlessly introduce surveys to collect feedback. Along with gathered metadata, this provides real-time information on issues and preferences, which makes it easier to develop relevant sales and support strategies. B. Customer engagement The main advantage of chatbots lies in their availability. They offer non-stop customer assistance, managing multiple interactions simultaneously. Because automated support outruns the capabilities of employees working within limited hours, it enables easier scaling-up of the operations and, what is more, improves customer experience. Real-time analysis of clients’ input and behavior may result in more personalized and 52 | 249 contextual offers. Chatbots are able to suggest instant follow-ups directed at a potential buyer’s preferences and needs. C. Operations optimisation Successful implementation of automated services and collection of extensive customer data could contribute to increased lead generation, clients’ attachment to the brand and, as a result, higher conversion rates. Chatbots release the human employees from performing repetitive tasks related to customer support and may lead to optimized order processes. As for the tasks still performed manually by a company’s team, bots can be just as useful to assist workers in resolving issues and searching through the data. 1.2.3.9. Edge artificial intelligence Edge artificial intelligence (edge AI) is “the implementation of artificial intelligence in an edge computing environment, which allows computations to be done close to where data is actually collected, rather than at a centralized cloud computing facility or an offsite data center. Edge AI lets devices make smarter decisions faster, without connecting to the cloud or offsite data centers” (Red Hat, 2024). A. Knowledge and insight management Edge AI speeds up the conversion of knowledge into decision-making. Smart devices process updated information on the spot without the need of sending them to data centers, which minimizes latency of resulting operations. Keeping the data within the equipment or a local server, edge AI ensures privacy of sensitive information and allows for immediate responses to critical issues. B. Customer engagement Fast processing of edge devices improves the efficiency of textual, visual and voice searches. Sensors gather information on customers’ behavior and enable real-time personalisation of offers. C. Inventory management Edge AI keeps the inventory information up to date and in accordance with supply chain optimisation analytics. Applied in a warehouse or a store, smart shelves monitor the 53 | 249 products for their quantity, need to restock and abnormal activities. This ensures the right level of supply and proactive prevention of unnecessary loss. D. Operations optimisation Edge computation distributes the impact of extensive processing throughout smart devices. This not only reduces latency but also cuts the costs of transferring all the data to the cloud. It may become easier for a company to scale up, when additional computing tasks are performed within the installed devices. With the help of edge AI, in-store operations are also better secured. Radio-frequency identification (RFID) security tags prevent the products from being secretly taken through the security gates, whereas automated points of sale (POS) report on any mis-scans or ticket-switching almost in real-time. 1.2.3.10. Robotic process automation Robotic process automation (RPA) can be understood as “the software to automate tasks within business and IT processes via software scripts that emulate human interaction with the application user interface. RPA enables a manual task to be recorded or programmed into a software script, which users can develop by programming, or by using the RPA platform’s low-code and no-code graphical user interfaces. This script can then be deployed and executed into different runtimes. The runtime executable of the deployed script is referred to as a bot, or robot” (Gartner, 2024d). A. Knowledge and insight management RPA enables automating complex data-related tasks, including their collection from various sources, cleansing, categorisation and comparisons. The technology can be set to perform these actions regularly and provide updated reports for quick problem solving, strategy planning and decision making. Automated analysis involves real-time data and predictions to generate solutions for optimized pricing, marketing campaigns, new products, customer service or operations’ workflow. B. Customer engagement Implemented into bots, RPA enables constant interaction with clients and generating immediate responses to simple queries. Automating order tracking provides customers with updates on every step that leads from the purchase to completed delivery. The feedback can be also collected automatically and return procedures - carried out without the delay. Gathered data on clients’ experiences, past activities and behavior is 54 | 249 easily transferred between CRM and marketing systems to relevantly segment the target groups and deliver personalized content for repeated purchases. C. Inventory management With extensive data on supply chain actual and predicted dynamics, RPA automates planning and streamlines communication between different members of inventory and fulfillment systems. Regular tracking and reporting keeps the operating teams always up to date with stock information and ready for immediate response to encountered problems and opportunities. D. Operations optimisation RPA can handle mundane manual operations with better accuracy and in shorter time. It can be assigned with various accounting, recordkeeping and invoice processing tasks and ensure their organization, compliance and security. RPA systems are used to keep information up to date, harmonize operations of a company's sectors and develop dynamic solutions for pricing, order fulfillment and marketing strategies. Along with extensive customer data, RPA handles information on employees and automatically controls workforce distribution with scheduling and payroll insights. 1.2.3.11. Generative AI Generative AI is a kind of artificial intelligence technology based on machine learning. It differs from other kinds of AI in its ability to create new data objects based on its training dataset rather than making predictions concerning correlations between data in its training dataset. This technology leverages neural networks and deep learning to generate content, such as text, images, and even complex data sets, making it highly versatile and innovative. Introduced in 2022, companies are still experimenting with its potential. A more specific analysis of the impact of generative AI on the retail sector is lacking in the literature we reviewed. However, during interviews, workshops, and data collection for use cases, we confirmed that some companies are already using it to their advantage (see case 3.12). Current applications include personalized marketing campaigns, dynamic pricing strategies, and automated content creation for websites and social media. Generative AI, such as ChatGPT, has the potential to enhance four main areas of company operations: A. Knowledge and Insight Management: Generative AI can process and synthesize vast amounts of information, providing valuable insights and facilitating knowledge sharing 55 | 249 and creation. For example, it can summarize large datasets and generate comprehensive reports that help businesses make informed decisions. B. Customer Engagement: Generative AI can enhance customer relationships by offering personalized responses, handling inquiries, and providing 24/7 support. It can simulate human-like conversations, improving customer satisfaction and loyalty through more interactive and responsive communication channels. C. Inventory Management: Generative AI can analyze sales data and trends to predict demand, assisting in balancing supply and optimizing stock levels. This helps retailers maintain appropriate inventory levels, reducing waste and improving availability of products. D. Operations Optimization: Generative AI can automate routine tasks, generate reports, and provide decision support, helping to minimize costs and maximize operational efficiency. It can streamline workflows, optimize resource allocation, and enhance overall productivity. Despite its promising potential, implementing generative AI comes with challenges and ethical considerations. These include data privacy concerns, the need for significant computational resources, and potential biases in AI-generated content. Addressing these challenges is crucial for leveraging the full benefits of generative AI while ensuring ethical and responsible use. 1.2.3.12. Summary Table 8 summarizes main uses of each of the described AI technologies in relation to the value chain in retail companies. Table 8 Main uses of AI technologies in 4 retail areas Technology Main uses in retail value chain (Knowledge and insight management, Customer engagement, Inventory management, Operations optimisation) Machine Learning Data classification, analysis and predictions, data security Customer segmentation and personalisation Matching stock levels with demand forecasts Resource allocation, market trends, fraud detection Deep learning Context-specific data patterns, real-time analysis Dynamic marketing, hyper-personalisation Augmented intelligence Data monitoring and exploration, problem-solving Self-service Inventory reports and suggestions Employee support 56 | 249 workforce. Workshop participants generally expressed a positive and calm perspective on the swift digital transformations driven by the rise of AI. Participants from Romania highlighted that AI does not aim to replace or automate tasks to displace workers from the labor market. Instead, AI enhances their existing activities: During our research with some managers from stores like Metro, Cash and Carry, Selgross, etc. we realized that AI can be an extension of the employee, for the employees, so AI will not replace them, it will be an extension and also we can help with the training that I was talking about a few minutes ago. Because the AI can also be used by the employees, not only by the customer. So the customer comes, you know, to a store and interacts with an AI, let's say an in-store AI assistant, to get answers like price description, where it is placed in the store, the employee can also do the same thing. And that way they can learn and they can increase their knowledge base about the products in the store. (Workshop_Romania_1) During a workshop, a participant shared insights derived from research conducted in Romania, highlighting the dual role of artificial intelligence in customer interactions and employee enhancement. AI, exemplified by virtual or desktop chatbots, automates customer interactions, thereby streamlining processes. Furthermore, AI enhances the analytical capabilities and knowledge of employees across all experience levels. This multifaceted application of AI garnered a positive reception among workshop participants, fostering an optimistic outlook towards automation's potential benefits. The theme of automation emerged from the analysis of survey questionnaires conducted in Cyprus. Respondents believe that AI can replace company employees in the case of repetitive and time-consuming tasks. One in three states that technologies such as chatbots can take over customer service and support tasks, and one in four believes that AI technologies can relieve employees of tasks such as data analysis and text generation (Figure 3). Moreover, AI can enhance employees' tasks by freeing them from repetitive duties. Additionally, it can create recommendations for customers, make decisions, and manage inventory. 63 | 249 Figure 3 In what aspects do AI technologies replace tasks performed by employees? (% of responses) Source: own elaboration 64 | 249 Figure 4 In what aspects do AI technologies enhance the way tasks are performed by employees? (% of responses) Source: own elaboration Automation, according to workshop and survey participants, boils down to automating basic and repetitive activities, particularly in the areas of interaction and customer support, as well as more advanced work in the supply chain. Significantly, the latter area is more difficult to automate, as it requires adapting the data infrastructure or business structure for analytical capabilities and having large data sets on which to train differentiated AI models. A positive attitude towards change prevailed among the workshop participants, who pointed in particular to the positive role of automation in filling the gaps caused by the unavailability of certain types of employees, as well as strengthening the workforce in terms of knowledge and data analysis. 1.3.4. Good Service The workshop participants explored not only the technical aspects of prediction, optimization, and automation but also the qualitative dimensions of artificial intelligence and its application in the retail sector. They viewed this as an opportunity to enhance service quality, which involves effective communication with customers and appropriate customer care throughout the sales process. These aspects are intertwined with the challenge of fostering trust in customer interactions and cultivating a supportive environment for MSME customers. 65 | 249 As presented by a workshop participant from Italy, the implementation of AI tools in retail has to take into account the issue of trust in customer relationships: I would like to draw attention to trust: to introduce AI into the retail world, we need to build a relationship of trust. There is - and this is measured by numerous studies - a certain repulsion to innovations that imply a radical change in habits. Think of a customer where the presence of a human being is replaced by AI. The relationship with the customer changes. We need to know how to introduce a new relationship of trust into this workflow. This is achieved by choosing systems that can be tailored or fine-tuned to the context in which they operate. I believe it is essential to address how the customer experience will change, how their attitude may be altered, and how their propensity to trust may be modified. This is a critical element that developers and management adopting AI must address. (Workshop_Italy_3) According to one workshop participant, successfully implementing artificial intelligence involves integrating this technology not just with data infrastructures and organizational frameworks, but also with a robust service system. The cornerstone of this system is trust between small to medium-sized enterprises and their customers. Upholding trust necessitates the careful and skillful deployment of AI, which should be customized to suit the context and unique characteristics of customer groups throughout the sales process. This poses a significant challenge in terms of developing company-wide expertise and managing interactions with external stakeholders. It is therefore important to consider customers' perspectives and attitudes towards automated customer service when designing changes related to the implementation of AI. On the other hand, once contextual factors such as the identification and comprehensive understanding of customer groups have been considered, the participant points to the conditions of opportunity that allow for the valuable use of AI technology. Based on trust in customer relationships and the thoughtful use of AI technology, effective customer care can be integrated into the sales process. Yeah, I think that's where at least this market segmentation comes in. I mean, it's with chatbots, that's for sure, it's kind of basic, but it's precisely because of the aftersales care, somehow generating at least some messages that go out now of purchase or just: "Your product is still in the basket, maybe you still want to buy it". This is what many online shops do now. For example, I have an account with many platforms that sell digital books. I get these reminder emails all the time: "There is still a product in your shopping basket, maybe you would like to buy it". This is not done by anyone, not by some tired Mr Smith, it is done automatically. Sending newsletters, that's all .... should be doing by now. (Workshop_Poland_2) Among the key factors for providing excellent customer service are the judicious use of chatbots or virtual assistants, and the automation of marketing and customer care. However, as workshop participants pointed out, delivering good service necessitates high-quality technologies capable of responding to customer demands and establishing an appropriate care system for them. Customer care is understood here by maintaining constant contact at all 66 | 249 stages of sales and after-sales process. Therefore, the role of skills in enabling the appropriate use of AI in retail is important (as discussed in the next section). Table 9 AI functions needed in relation to challenges in the retailing sector AI functions needed In relation to challenges in retail Predictive functions • Predicting demand in supply chains • Forecasting sales trends Optimization functions • Optimizing inventory levels • Sustainable use of resources and products Automation functions • Automating marketing and customer relationship • Managing customer feedback • Automating routine tasks or complementing workforce “Good service delivery” • Used skilfully, it helps build a relationship of trust with customers • Provides opportunities for holistic customer care during the sales process and afterwards Source: own elaboration Looking at this comprehensively, participants in co-creation workshops across selected EU countries highlighted four key functions of AI in retail. Firstly, AI's predictive capabilities enable better management of supply chains and sales. Secondly, it optimizes related processes. Thirdly, participants emphasized the automation of customer service and specific tasks, addressing workforce gaps or enhancing existing staff capabilities. Fourthly, AI's potential contribution to enhancing customer service, particularly in customer relations, was noted. Participants stressed the importance of applying AI within the context of MSMEs, considering their understanding of their customers to avoid undermining trust in these relationships. Only through this approach can automated and technology-enabled customer care, such as virtual assistants, be effectively implemented. 67 | 249 2. THE STATE OF AI ADOPTION IN RETAIL SECTOR Key takeaways ■ Over the last decade, the wholesale and retail trade sector has grown significantly. ■ The retail sector is dominated by a high number of micro, small and medium-sized enterprises (MSMEs), especially in Romania. This structure can hinder competitiveness in international markets due to lower operational efficiency and challenges in achieving economies of scale, significantly impacting the digitalization process. ■ The retail sector's digitalization lags behind other industries such as ICT and professional services. ■ E-commerce sales in the retail sector vary significantly between countries, indicating different levels of digital technology adoption. ■ Most MSMEs in EU countries do not use AI technologies, with Poland having the highest percentage of firms not using AI. ■ In the retail sector, AI adoption is the lowest among the analyzed sectors. The most common AI applications in MSMEs are robotic process automation (RPA), speech recognition, and text mining. ■ The main barriers to AI adoption are insufficient specialist knowledge, high implementation costs, and legal uncertainties, limiting the implementation of these technologies in MSMEs and the retail sector. Insufficient specialist knowledge and difficulties in hiring skilled employees are significant barriers to adopting digital technologies in the retail sector and MSMEs. ■ While digital skills are increasingly necessary for implementing digital innovations, the level of these skills among employees remains low, particularly in Poland and Romania, which hampers digital technology adoption. 2.1. The main characteristics of the retail sector in the selected countries In the last decade, the wholesale and retail trade sector has experienced growth, except during the COVID-19 pandemic. This growth is particularly noticeable in Romania and Poland, which lead in turnover, while Germany and Italy have fallen below the EU average, highlighting increasing disparities in growth rates across Europe. A key feature of this sector is the high number of small and medium-sized enterprises (SMEs), especially in Romania, and to a lesser extent in Germany. This structure can limit competitiveness in international markets due to lower operational efficiency and difficulties in 68 | 249 achieving economies of scale. Additionally, it plays a crucial role in the digitization process. Data on turnover and sales volume in the wholesale and retail trade in selected countries show interesting trends. Romania and Poland have the highest turnover among the analyzed countries (Figure 5). The chart presents the annual index of turnover and sales volume in wholesale and retail trade in 2023, using 2021 data as the reference point (index = 100). Bulgaria leads among EU27 countries and shows an increase compared to the previous year. Germany and Italy have results below the EU average, with Germany also experiencing a decline compared to 2022. Figure 6 shows the long-term trend in turnover and sales volume in wholesale and retail trade from 2014 to 2023 for the analyzed countries. The data reveal significant growth in this sector, particularly in Romania, which had an index of 55.10 in 2014 and reached 128.40 in 2023, indicating a dynamic market development. In contrast, Italy, which started from a higher index level of 87.70 in 2014, now occupies the last place. The data suggest a stable and growing development of the retail and wholesale market. Notably, since 2020, all analyzed countries have shown accelerated growth, likely due to economic recovery following the COVID-19 pandemic. A notable characteristic is the very high number of small and medium-sized enterprises (SMEs) in the wholesale and retail trade sector (Figures 7, 8, 9). In the analyzed countries, over 97% of businesses in this sector are SMEs. On average, one in five SMEs operates in the retail trade sector, except in Germany, where the share is lower at 16.8%, and in Romania, where it is higher at 26%. This structure leads to significant market fragmentation, which can affect competitiveness and growth dynamics in these countries. This fragmentation can result in lower operational efficiency and difficulties in achieving economies of scale, limiting international competitiveness. Additionally, a large number of SMEs can lead to greater market volatility and instability, complicating long-term strategic planning. Furthermore, the high number of SMEs may face challenges in the digitalization process, as smaller businesses often have limited financial and technological resources, making it difficult to implement advanced digital technologies. This can impact their ability to adapt to changing market conditions and their competitiveness in the digital era. Indicators of digitalization levels among enterprises in the EU support this. The digitalization of SMEs in the analyzed countries (Poland, Cyprus, Romania, Germany, Italy) does not significantly differ from the EU average (24%) for small and medium-sized enterprises, except for Romania (9%) (Figure 9). However, these figures for SMEs are considerably lower than those for large enterprises. Among the analyzed group, Germany and Cyprus achieve the highest results, with about one in three SMEs characterized by advanced use of digital technologies. Romania stands out negatively in this regard, both among the analyzed countries and within the entire EU. 69 | 249 Figure 5. Turnover and Sales Volume in Wholesale and Retail Trade – Annual Data (2023) Source: Eurostat. Index of Turnover – Total, Unit: Index, 2021=100, Calendar Adjusted Data, Circles – Result from 2022. Figure 6 Turnover and Sales Volume in Wholesale and Retail Trade – Annual Data (2014-2023) 70 | 249 Source: Eurostat. Index of Turnover – Total, Unit: Index, 2021=100, Calendar Adjusted Data Figure 7 Number of Small and Medium-Sized Enterprises in the Retail Sector (2022) 71 | 249 Source: Eurostat. Circles – Result from 2021 Figure 8 Share of SMEs in All Retail Enterprises (2022) 72 | 249 2.3. Use of advanced digital technologies In 2023, the vast majority of SMEs did not use any AI technology. Among small and medium-sized enterprises, the highest percentages of firms not using AI were found in Finland (92%), Sweden (86%), and Poland (84%) (Figure 14). In other EU27 countries, these values ranged between 70% and 80%. Regardless of company size, Poland is among the countries with the highest percentage of enterprises not using AI. Larger enterprises show higher levels of AI adoption, likely due to greater financial and organizational resources enabling investments in modern technologies. The retail sector ranks lowest in terms of AI usage (Figure 15). AI adoption in retail is significantly lower than in the ICT sector, where the EU27 average is around 30%. Among the analyzed countries, Germany has the highest AI usage in retail (about 10%), where one in ten retail enterprises uses at least one AI technology. Other countries fall below the EU average, with few firms employing such technologies. Furthermore, all the analyzed countries show low growth in AI usage compared to 2021 (Figure 16). Over the past two years, AI implementation has increased most rapidly in Cyprus (by about 2 percentage points). Retail enterprises in Romania have minimal AI usage. When AI is used (both in SMEs and retail companies), it is most commonly for robotic process automation (RPA) based on AI. RPA automates routine, repetitive tasks and, when supported by AI, can also make decisions based on data analysis, pattern recognition, or natural language processing. This makes business processes more efficient, accurate, and faster (Figures 17 and 18). Other common AI applications in SMEs include speech recognition and text mining. Speech recognition converts spoken words into text, improving processes such as customer service and meeting transcription. This increases communication efficiency, documentation accuracy, and saves time and human resources. Text mining involves analyzing large text datasets to extract valuable information such as patterns, trends, or sentiments. In SMEs, text mining is used to analyze customer opinions and monitor the market. In the retail sector, machine learning is also frequently used, enhancing various business aspects. For example, analyzing customer behavior data allows for personalized product recommendations. The main barriers to AI adoption in SMEs and the retail sector are insufficient specialist knowledge (Figure 19). Another significant barrier is the perceived high implementation costs of these technologies. Less frequently reported issues include incompatibility with existing systems and unclear legal consequences. On average, across all EU countries, the biggest barrier is insufficient specialist knowledge (Figure 20). Legal uncertainties and ethical considerations are also significant. Similar differences are seen in the use of other technologies, such as cloud computing services (Figures 22 and 23). Poland and Germany have moderate cloud usage levels, both at 44.5%, while Cyprus and Italy have higher values, at 52.4% and 58.4%, respectively. Romania is at the bottom with the lowest level (13.8%), indicating significant digitalization delays. In terms of 79 | 249 purchasing office software, security, and file storage, Italy leads, while Romania scores the lowest, highlighting the differences in digital technology adoption in the retail sector among these countries. The countries also differ significantly in integrating IoT technologies (Figures 23 and 24). The highest IoT adoption in the retail sector is in Cyprus (40%), while Poland (11%) and Romania (13%) are far behind the EU average (26%). Cyprus excels in building security (39.1%) and energy management (13.9%). Germany is close to the EU average in building security (26.9%), while Poland and Romania show lower engagement in most analyzed IoT applications. Figure 14 Enterprises not using AI systems (2023) Source: Eurostat. Artificial Intelligence by Class Size of Enterprise (Enterprises do not use any AI technologies). The breakdown includes small and medium-sized enterprises (from 10 to 249 employees), all (from 10 employees), and large (250 employees and more). 80 | 249 Figure 15 Enterprises from different sectors that use at least one AI system (2023). Source: Eurostat. Artificial Intelligence by NACE Rev. 2 activity (Enterprises use at least one of the AI technologies). 81 | 249 Figure 16 Retail sector enterprises using at least one AI system (2023) Source: Eurostat. Artificial Intelligence by NACE Rev. 2 activity (Enterprises use at least one of the AI technologies). Circles - result in 2021, bars - result in 2023. 82 | 249 Figure 17 AI applications in small and medium-sized enterprises (2023) Source: Eurostat. Artificial Intelligence by Class Size of Enterprise. 83 | 249 Figure 18 AI applications in retail trade enterprises (2023) Source: Eurostat. Artificial Intelligence by NACE Rev. 2 activity. 84 | 249 Figure 19 Reasons for not using AI in SMEs and retail trade enterprises (2023) Source: Eurostat. Artificial Intelligence by NACE Rev. 2 activity and Artificial Intelligence by Class Size of Enterprise. 85 | 249 Figure 20 Reasons for not using AI in retail trade enterprises (2023) Source: Eurostat. Artificial Intelligence by NACE Rev. 2 activity 86 | 249 Figure 21 Sources of data analysis for retail trade enterprises (2023) Source: Eurostat. Data analytics by NACE Rev. 2 activity 87 | 249 Figure 22 Use of cloud computing in retail trade enterprises (2023) Source: Eurostat. Cloud computing services by NACE Rev. 2 activity Circles: result from 2021 88 | 249 Figure 28 Use of AI and advanced digital skills of employees in retail trade enterprises (2023) Source: Eurostat Digital Skills & Enterprises using at least one AI technology. Figure 29 Use of AI and programming skills of employees (2023) Source: Eurostat Computer skills (Individuals who have written code in a programming language) & Enterprises using at least one AI technology. 95 | 249 Figure 30 Companies where digital skills are becoming increasingly important - SMEs and retail trade enterprises (2023). Source: Flash Eurobarometer 529 To what extent are the following skills becoming more or less important for your company? "Digital skills" (e.g., skills required for implementing or using digital technologies) 96 | 249 Figure 31 Employees with advanced digital skills in the retail and wholesale trade sector, compared to all employees (2023) Source: Eurostat. (Digital Skills) 97 | 249 Figure 32 Employees in the retail and wholesale trade sector with advanced digital skills (2023). Source: Eurostat. (Digital Skills). Circles - Skills in 2021, bars - Skills in 2023. 98 | 249 Figure 33 Share of employees with advanced skills in specific groups of digital competencies (2023) Source: Eurostat. Individuals' level of digital skills 99 | 249 Figure 34 Companies facing difficulties in hiring employees with the right competencies - SMEs and retail trade enterprises (2023). Source: Flash Eurobarometer 529 How difficult are the following tasks for your company? Finding employees with the right competencies 100 | 249 Figure 35 Companies facing difficulties in hiring specialists, junior specialists, and technicians - SMEs and retail trade enterprises (2023). Source: Flash Eurobarometer 529 Does your company have difficulties recruiting employees for the following positions? Specialists, junior specialists, and technicians 101 | 249 Figure 36 Companies where the lack of digital skills among employees hinders the implementation of digital technologies (2023). Source: Flash Eurobarometer 529 Lack of skills hinders the implementation or use of digital technologies in your company 2.5. Characteristics of selected states 2.5.1. Cyprus Cyprus stands out for its high level of digital technology adoption, particularly in cloud computing (52.4%) and IoT (40%). While digital skills are highly valued, the actual level of these skills among employees remains low. The Cypriot retail sector uses advanced technologies for building security and energy management, demonstrating a proactive approach to integrating new technologies. 2.5.2. Germany Germany has a moderate level of cloud computing (44.5%) and IoT (26.9%) usage but the highest share of e-commerce in the retail sector (20.9%). German firms show better advancement in 102 | 249 digital technology use but still face issues with the shortage of employees with above-basic digital skills. Despite this, Germany maintains a strong position in the digitalization of the retail sector compared to other countries. 2.5.3. Italy Italy features the highest level of cloud computing usage (58.4%) in the retail sector and a high level of e-commerce adoption. However, Italian SMEs struggle with a lack of specialized digital knowledge, hindering the implementation of new technologies. Italy holds average positions in AI and IoT usage, indicating the need for further digitalization investments to enhance market competitiveness in Europe. 2.5.4. Poland In Poland, the wholesale and retail trade sector reports high turnover but is characterized by a low level of digitalization, especially among small and medium-sized enterprises (SMEs). In 2023, only 7.7% of retail companies used e-commerce channels, and 84% of SMEs did not use AI technologies, indicating a need for intensified digitalization efforts. Additionally, Polish firms struggle to hire employees with the necessary digital skills, which hinders the implementation of new technologies and lowers market competitiveness. 2.5.5. Romania Romania records the highest growth in the wholesale and retail trade sector but also has the lowest level of digitalization among the analyzed countries. Only 13.4% of retail companies use e-commerce, and 13% use IoT, indicating significant delays in adopting digital technologies. Additionally, Romanian firms face significant challenges in hiring employees with the necessary digital skills, limiting their adaptability and market competitiveness. 103 | 249 3. REAL-LIFE AI USES IN RETAIL COMPANIES: USE CASES AND BEST PRACTICES In this section, we present how selected retail companies incorporate artificial intelligence technologies into their everyday operations. We compiled these use cases and best practices through a review of academic and gray literature, consultations with industry experts, and participation in industry conferences. The descriptions are based on the information sourced through each company’s website and, in several cases, through interviews with the company’s representatives. Key takeaways ■ In the majority of cases, AI was specifically used for optimizing customer support, mostly through interactions with chatbots and automated personalization of offers. ■ Other uses focused on automated SEO processes, supply chain management, content generation and moderation or general data organization and analytics. ■ Outsourced AI tools are often offered as a service, therefore marketed as requiring no technical skills or maintenance efforts. ■ Nevertheless, several cases suggest a need for basic understanding of the company’s operational data, as well as the knowledge on AI tool’s functionality and the benefits of it. ■ Deeper insight into the implementation process revealed a need for critical thinking skills for evaluating tools’ performance and communication skills for sharing the collected insights. ■ The analyzed cases tend to portray automation of manual processes as a positive change, focusing either on human-machine cooperation or the change of employees’ focus from repetitive to more complex tasks. To be noted, while we refer to “use cases” to highlight the diversity of applications of different AI technologies, the research process revealed that in business practice the term is overused for marketing purposes. For technology providers, “use case” usually means a general description of any possible use of an AI tool. This, however, begs the question of how applicable, in reality, a given solution would be for an MSME with limited resources. That is why we focused on the data provided in more specific cases. Frequently named “case studies” or “customer stories”, these accounts described successful AI applications in actual companies with precise information on the implementation’s results. This could occur in the form of KPIs (e.g. conversion rate, customer satisfaction score) or an explanation by a company representative of the impact on the operations/employees. While using these measures may indicate a well-grounded success of the described AI uses, it should be remembered that the metrics could have been defined with the involvement of the technology providers, presumably in favor of their product’s efficiency. The companies in which AI was implemented were all categorized as MSMEs, using the number 104 | 249 ⓘ Shipup is an order monitoring and communication tool that allows for tracking of shipping processes and providing real-time alerts to contact customers before they experience inconveniences of delayed order. Through Shipup, Feed. could automatically inform clients about order issues. This, in turn, reduced the number of tickets received by the support team and enabled quick solving of encountered difficulties. Additional use of Magento10 and Diduenjoy tools11 helped the company collect more valuable information on their customers. The first software was used to obtain order details, while the other one became a source of feedback and customer experience data. ⓘ Magento Order Management System (OMS) is a solution for monitoring and managing inventory, ensuring its visibility on different stages of supply chain and responsive customer service. ⓘ Diduenjoy offers solutions for optimized customer experience through actionable feedback. It enables survey creation and multichannel distribution, as well as organized feedback collection from different sources. Extensive customer experience data proved to be helpful not only for the streamlined support system. Effective communication within the company resulted in transferring valuable insights that influenced future marketing strategies, newsletter writing and even product design. 3.2.4. Results The implemented solutions led to a significant increase in Feed.’s sales. In less than a year, the live chat conversion rate reached 6%, which resulted in €180,000 revenue. The ticket management was optimized, releasing the employees from excessive workload. Tickets that required multiple touchpoints (employee’s actions), once adding up to 80% of all tickets, were now reduced to 10% and to a great extent replaced by one-touch tickets. Optimized customer support eventually proved to realize the idea of genuine “royalty experience” with Customer Satisfaction (CSAT) scores jumping from 75 to 94%. 3.2.5. Employee skills needed to adopt the solutions Despite limited resources, Feed. was able to easily adopt new technologies and make use of them through the team of 6. No developer skills or additional support were required to benefit from Zendesk’s system. Analytics tools were equipped with data visualization and reporting features and the company only needed a proper communication system to share the acquired knowledge for informed decision making. 11 https://www.diduenjoy.com/en/integration/zendesk 10 https://www.zendesk.com/marketplace/apps/support/321722/magento-2-by-zenplate 111 | 249 3.3. Grünewald 3.3.1. Description of the company Source: https://foodforecast.com/en/references/herr-gruenewald-von-der-baeckerei-gruenewald/ Name Country Size Grünewald Germany <250 employees Description: Grünewald is a Waldlaubersheim-based bakery that has existed in the German market for over 125 years. Today, with 29 branches in the region, the company preserves tradition while embracing changes of the modern world. It provides an example how new technologies can work in favor of sustainability, keeping the business safe for the future generations. Goal: Optimizing ordering processes to better match the demand for specific items. Strategy: Automated ordering system based on regular forecasts and centrally controlled. Technology provider: foodforecast Tools: MLand DL-powered software for order automation. 3.3.2. The need As a company with long traditions, Grünewald made considerable efforts to keep the business local. In spite of scaling up, the bakery procures raw materials from suppliers within the region. Caring for the environment and financial sustainability of the business, Grünewald wanted to use their resources prudently, while maintaining the established quality. With the energy and raw materials prices growing, optimal supply chain management was indeed a matter of survival. 112 | 249 Having developed a sizable chain of stores, however, the company’s management realized that some organizational practices were not fit for the new business model. To date, individual sellers fulfilled orders by themselves based on their intuition and experience. Even if there were instructions from above as to which products in what quantities should be ordered, the decision was eventually determined by a worker’s, some better some worse, personal judgment. With the number of branches increasing, new sellers had to be hired. This made the ordering process even more prone to errors. As a result, the inventory preparedness varied in time and place. On the one hand, unexpectedly high returns of unsold products occurred, while in different cases some products were out of stock too soon. Lastly, manual order generation engaged employees’ time and blocked the cash registers used in the process. In this particular case, it was the technology provider, foodforecast12, that reached the potential customer: ⓘ Foodforecast (formerly werksta.tt) is a German company that “uses artificial intelligence (AI) to reduce food waste” through their Software as a Service (SaaS) model. It combines a given company’s historical data with external factors to predict future demand and automatically optimize production, ordering and store planning. The main goal of Grünewald was to refine their ordering system with automation, reducing the consequences of human errors and overcoming demand uncertainty. 3.3.3. AI-based solution Convincing automation Grünewald’s POS data was analyzed with foodforecast’s software in order to implement the Order optimisation solution13: ⓘ AI-based Order optimization is one of the tools built on foodforecast’s algorithm that provides daily forecasts at order item level. The predictions of sales volumes are automatically translated into order proposals, ready to be triggered or manually adjusted by employees. 13 https://foodforecast.com/en/unsere-loesungen/ 12 https://foodforecast.com/en/ 113 | 249 After successful testing in one of the Grünewald’s stores, the solution was introduced in other locations. Each day, the system generated orders based on the forecast of the following day’s sales of different products. Initially, the more experienced employees were skeptical and modified the automatic orders. As the time passed, the predictions made by AI software turned out to be so accurate that the change was uniformly embraced. Eventually, the entire ordering process was automated. The employees could stop bothering with manual calculations and their uncertain predictions and fully concentrate on selling the products. More control with less effort All of the data became organized and easily available for the management to overlook the operations and set strategic targets, e.g. concerning the desired balance between returns and stockouts, although as the system proved successful and consistent, the top-down adjustments could also become less frequent. Transparency of information enabled unprecedented insight into discrepancies and became a reliable basis for theft prevention. 3.3.4. Results The implemented solutions led Grünewald to better financial outcomes. The returns became smaller in size, and demanded products were sold in larger quantities, in some cases reaching 80% increase in sales. With reduced loss of resources, the company had strong evidence to claim sustainability in front of its clients. 3.3.5. Employee skills needed to adopt the solutions Since the success of the AI system was based on lower levels of employee intervention, it may suggest little to no need for additional skills. Grünewald’s manager also claimed that in case of troubles, foodforecast’s support was always available. However, he then stated that the technology provider could “reduce the complexity even further”. Convincing skeptical employees towards the system was said to be enabled mostly due to technology’s efficiency but the mere occurrence of the problem may point to the need of adaptation and change management skills. 3.4. Kerrigans 114 | 249 3.4.1. Description of the company Source: https://www.shopbox.ai/casestudies/shopbox-ai-transforms-the-shopping-experience-for-cr aft-butcher-kerrigans-boosting-conversions-by-250 Name Country Size Kerrigans Ireland <50 employees Description: Kerrigans is a family-run craft butcher. Along with 5 locations in northern Dublin, the company conducts its sales online, delivering products all over Ireland. The offer ranges from traditional cuts to ready-to-eat low-fat meals, along with barbecue and cooking condiments. Owners take pride in the quality of the meats, assuring a fully transparent farm-to-fork supply chain. Goal: Engaging and profitable e-commerce with limited resources. Strategy: Personalized customer experience through automation tools. Technology provider: Shopbox AI Tools: AI Curated Homepage, AI Shop Assistant 3.4.2. The need Brendan Kerrigan valued locality and direct contact with the customers since he founded his family business in 1973. In the present, the company operated by his two sons expanded its operations to the ecommerce model. Keeping up with modern requirements of the retail industry could not, however, be limited to launching an online store within the Shopify platform. The owners wanted to preserve the quality of service provided by their small team in the physical locations. This traditional approach was also motivated by the company’s limited resources. Kerrigans entered into partnerships with alliance groups for national retailers that, to a point, helped them become fair competition to large companies. Yet, for the new ecommerce channel to serve its purpose in an efficient way, some change in operations management was necessary. Kerrigans needed more online presence and conversions without overly exploiting its staff and finances. 115 | 249 The solution to this challenge emerged when the company came across Shopbox AI14 services. ⓘ Shopbox AI is a sales engine for optimizing traditional customer experience in ecommerce environments. Fully automated and maintenance free, the software generates conversions through personalisation tools, intelligent merchandising and unobtrusive assistance starting from the user’s first click within the website. The main goal of Kerrigans was to maintain the in-store quality of its customer service, while engaging more online customers to purchase the products. 3.4.3. AI-based solution Personal website After only 3 days of preparations, Shopbox AI implemented their first technology into Kerrigans’ Shopify store - Curated Homepage15. ⓘ Curated Homepage is a feature that performs dynamic personalisation based on a user’s every interaction with a website. Through targeted layout and product offers it provides a unique shopping experience for each customer. Curated Homepage displayed recommendations, promotions and recently viewed or trending items. As a result, new clients could become engaged from the very first visit to the online store, whereas previous users gained quick access to the products they were interested in. Recommended items shown in product display page (PDP) carousels not only followed customers’ preferences but were also linked to Kerrigans’ inventory and marketing data. Stock levels, repricing and promotions were automatically integrated with the actual offer. This way, the process required no intervention on the part of the company’s small team. Maintaining excellent service As Kerrigans’ offer is designed to outmatch the regular supermarket supplies’ quality, it must also come with the proper customer support. In physical stores, employees made sure that the clients were served with care for their individual needs. To match the level of online shopping experience with traditional in-store encounters, Shopbox AI introduced its AI Shop Assistant16. 16 https://www.shopbox.ai/blog/how-ai-shop-assistants-are-changing-the-retail-industry 15 https://www.shopbox.ai/shopbox-demo 14 https://www.shopbox.ai/ 116 | 249 ⓘ AI Shop Assistant is a customer service tool providing personalized follow-ups for similar products, up-selling and cross-selling. Through tailored offers it helps increase the average order value (AOV) and reduce the bounce rate. AI Assistant enabled Kerrigans to present its wide range of products without bringing the clients to one of their locations. The fully automated process not only reduced the demand for employees’ customer support but also did not require any technical maintenance. The distinctive quality of this approach was that it allowed for advanced customer experience without the need for substantial amounts of historical data that are usually used for personalisation. 3.4.4. Results The implemented solutions led the small company to reach growth. Shopbox tools, while still engaging only 39% of the online customers, now mediate 59% of transactions. Spending twice as much time on Kerrigans’ website and users view three times more products. This translated to a 250% increase in conversions and an average order value higher by 9%. 3.4.5. Employee skills needed to adopt the solutions Since the solutions offered by Shopbox AI are based on full automation, the technology provider claims that no IT support or maintenance is required to benefit from its services. 3.5. Kuchyne Valent 3.5.1. Description of the company Source: https://pl.semrush.com/company/stories/kuchynevalent/ Name Country Size Kuchyne Valent Slovakia <50 employees 117 | 249 Description: Kuchyne Valent is a family-run kitchen and furniture manufacturing company based in Bratislava. KV offers their products to individual customers through the company's website https://www.kuchynevalent.sk/ which combines an online store and a blog. Goal: “Creating exceptional blog content and ensuring its widespread distribution.” Strategy: SEO, image optimization Technology provider: Semrush Tools: Site Audit, Keyword Magic Tool, Keyword Gap, Keyword Overview 3.5.2. The need After expanding their sales channels to the digital environment, Kuchyne Valent faced new challenges of reaching the online customers. With limited resources, and little SEO experience, they set on a task to develop a successful marketing strategy that would make the company visible and its offer - matching potential clients’ searches. The company’s website serves two purposes. Its online store provides a sales channel positioning it in an e-commerce model. The blog, on the other hand, acts as a marketing tool helping Kuchyne Valent showcase its solutions and finished products. It also became the main source of shared content contributing to visibility of the company's offer. Initially, the team running KV made independent efforts to promote their company using different SEO techniques. The results, however, proved to be unsatisfactory and Kuchyne Valent reached out for analytic tools offered by Semrush17: ⓘ Semrush is the online visibility management and content marketing SaaS platform. It provides services such as website auditing, keyword research, backlink optimization, competitive analysis, content creation and ranking monitoring. The main goal of the company was to improve their blog content so that it generated more leads which, in effect, would increase sales and company’s revenue. 17 https://www.semrush.com/ 118 | 249 3.5.3. AI-based solution Understanding the problem First step taken by the Semrush team was to understand the problems that Kuchyne Valent was facing and find a solution tailored for their specific needs. Using the Semrush Site Audit18, the team identified specific influence of various SEO techniques on the company's organic traffic. The audit proved that the former strategy, which favored copywriting, could be improved by an approach that put more emphasis on image optimization. ⓘ Site Audit - SEO analysis tool, crawling and checking every page on a given website. It produces regular reports on website’s issues, their levels of priority and solutions to fix them. Optimizing content with keywords Well-selected keywords helped Kuchyne Valent organize their web content in accordance with SEO demands. With the help of Semrush Keyword Magic Tool19, they were able to find the most efficient keywords for image titles. All the keywords were categorized, creating clusters with relevant subcategories and related searches. That helped define the structure of the whole blog content and distribute images across different categories and subcategories. ⓘ Keyword Magic Tool - topic-based keywords suggestions for optimized and structured content organization. Shows possible links between more general topics, subgroups and related searches, including long-tail keywords. Understanding competition Using the Semrush Keyword Gap20, KV could analyze their competitors’ best-performing keywords that were not as efficiently used by the company. Identified gaps and strengths acted as a basis for target keyword masterlist. ⓘ Keyword Gap - enables a comparison of keyword profiles between competing companies. Lists all common and unique keywords (including organic, paid and PLA) a given firm ranks for with values and cross-company differences and intersections. Understanding trends Keyword Overview21 helped KV identify trending keywords which resulted in developing a specific product line that met the demand, increased blog’s SERP score and let the company achieve expert status in the specific area. 21 https://www.semrush.com/analytics/keywordoverview/ 20 https://www.semrush.com/analytics/keywordgap/ 19 https://www.semrush.com/analytics/keywordmagic/ 18 https://www.semrush.com/siteaudit 119 | 249 ⓘ Keyword Overview - keyword research database with general information on a given keyword, including search volume, difficulty, intent and CPC value. 3.5.4. Results The implemented solutions helped Kuchyne Valent achieve 250% revenue growth. Over the next 5 years, the website's organic traffic remained the primary source of lead generation. Later on, the company expanded its intake of marketing technologies and started PPC advertising using Semrush PPC insights and Google Transparency Center. 3.5.5. Employee skills needed to adopt the solutions Overall, main gaps in KV’s resources were: a) lack of data, b) lack of understanding of off-page SEO (search engine rankings influenced by processes conducted outside the website). 3.6. L Cosmetics 3.6.1. Description of the company Sources: https://www.leafio.ai/case-studies/l-cosmetics/, an interview with Leafio representative Name Country Size L Cosmetics Estonia <250 employees Description: L Cosmetics is a drugstore retail chain selling hygiene, health and personal care products for women, men and children. Their offer includes over 90 brands from global and domestic producers. Sales take place through over 25 stores located all over Estonia and an e-commerce website. Goal: Creating an optimized inventory management system. Strategy: Automation of order processing and stock management. 120 | 249 dependency on human support agents, the company was unprepared to meet the periodically rising demand without the delay of service. Dissatisfied with former providers and modes of operation, Procosmet decided to try out the offer of Tidio26, a platform providing AI-based tools and solutions for optimized customer service. ⓘ Tidio - “Our suite of live chat, chatbots, helpdesk tools, and AI solutions helps brands of all sizes drive their businesses forward by offering quick and qualitative support and creating real connections with their customers.” The main goal of the company was to implement a single app that would handle all the customer service tasks and be accessible to the employees and clients alike. As a result, new contacts would be generated to create “a healthy community” of interested and satisfied users. 3.8.3. AI-based solution One app for all To adjust Tidio’s technology to their specific needs, the Procosmet’s team decided to choose Tidio+ plan27 with customized analytics and automation solutions. ⓘ Tidio+ provides a set of tailored customer services, including dedicated support team, live chat, chatbots and ticketing system. It allows the companies to improve the rates of their leads, conversions, sales and customer satisfaction with quick return on investment and easily scalable solutions. With Tidio+, the company was able to carry out all the necessary tasks within one environment. Order management, data collection and customer support could now be conducted in an efficient and consistent manner. Consolidating communication through live chat Live chat feature28 enabled employees to gain easy access to order information. Any modifications, cancellations and refunds can be quickly processed through the conversation panel. Streamlining communication through one tool sped up the reaction time and unified conversational data for further analyses. 28 https://www.tidio.com/live-chat/ 27 https://www.tidio.com/tidio-plus/ 26 https://www.tidio.com/ 127 | 249 ⓘ Tidio’s live chat, tailored for e-commerce and small businesses, is a customer support tool for managing communication channels. It provides access to multiple interaction platforms from any workplace, which leads to quicker responding, increased customer engagement and constant tracking of users’ activity for better personalisation and higher sales. Automated interaction One of the major changes that occurred through implementation of Tidio’s system was the introduction of chatbots29. Connected to the company’s operations they proved useful in carrying out numerous tasks with unprecedented efficiency and speed. ⓘ Tidio’s chatbots automate sales support and customer service through engaging in interactions with clients and collecting data generated in the process. Through personalized assistance, they can simplify lead generation, increase average order value and reduce cart abandonment. Real-time multi-language conversations allow for effective communication further improved by advanced problem-solving and targeted recommendations. In Procosmet’s case, there are five chatbots serving a diverse range of functions, such as FAQ answering, newsletter management, data collection and customer service. They are able to perform their tasks outside regular working hours and assist many clients at the same time. Through friendly interactions, chatbots greet users, offer help and create an inviting atmosphere to keep the customers engaged and inclined to finalize the transactions. Getting the data Chat interactions allow the potential clients to get to know the offer that might suit their personal needs, resolve issues, as well as subscribe to the newsletter and give their contact information. The efficiency of these processes is both enabled by and further beneficial for analytical tools offered by Tidio30. ⓘ Tidio’s analytics tools collect and process data to monitor performance and provide actionable insights. They keep track of the chat flows, activity patterns and emerging issues to optimize operations and improve customer engagement. Detailed analytics allow for constant monitoring of sales data and details of customers’ behavior and preferences. Information from previous transactions results in deeper understanding of obstacles and opportunities that could be used for quick problem-solving and up-to-date workflow optimisation. Real-time improvements and data-based support were key to increased satisfaction of the customers. 30 https://www.tidio.com/analytics/ 29 https://www.tidio.com/chatbot-ai/ 128 | 249 3.8.4. Results The implemented solutions led to improved Procosmet’s KPIs in many areas. They were able to launch a marketing campaign that resulted in over €1000 return on investment. The overall conversion rate was not only stabilized but systematically increased by 27% in a year. Along with that the company achieved a 23% growth in sales. Monthly lead generation changed from 10-30 to over 100. Email opening rate reached an average of 18-22%. According to Procosmet’s team, Tidio’s solution helped generate over ⅓ of their current ecommerce revenue. The changes were also appreciated by the clients. Average ratings of their reviews grew from 3.8 to 4.75 out of 5 stars. 3.8.5. Employee skills needed to adopt the solutions After having issues with the complexity of previous operating systems, Procosmet’s team seeked “an easy-to-teach app” to serve their needs. Tidio promotes their products as accessible, without a need of programming that would require IT specialists to work with. 3.9. Purelei 3.9.1. Description of the company Sources: https://www.ultimate.ai/customer-stories/purelei, https://www.zendesk.de/blog/die-zielgruppe-an-der-richtigen-stelle-abholen/ Name Country Size Purelei Germany <250 employees Description: Founded in 2016, Purelei offers jewelry and lifestyle products inspired by Hawaii. The people behind the company take pride in the durability of materials, collaborative design process and the guiding “Aloha way of life”. Through various sales channels and communication media, Mannheim-based Purelei reaches customers across Europe. Goal: Quality customer experience without overexploiting the support team. Strategy: Personalized self-service and automation of repetitive tasks. Technology providers: Zendesk, Ultimate Tools: Zendesk Explore, Ultimate Zendesk integration, UltimateGPT 129 | 249 3.9.2. The need “Be your own customer” - is the motto of Purelei’s customer service team. With an international group of clients and numerous forms of communication ranging from telephone, e-mail to different social media chats, the company needed an organized system of support. The idea of providing comfort and care to every client resulted in a help center developed through Zendesk platform31: ⓘ Zendesk is a customer service software providing a complete CS solution that is both scalable and easy to use. Its range of tools are expected to improve customer experience, optimize employee productivity and provide valuable insights into companies’ operations. Their service is based on original solutions integrated with external apps and technologies. The help center gathered data from various communication channels and contained a knowledge base of FAQs and helpful information in 4 (now 5) languages. However, aiming for high quality service, Purelei’s team wanted to personalize its communication with every customer. At the same time, they were aware that supporting so many requests with adequate care was laborious and oftentimes involved repetitive work, let alone time. Purelei looked for a way to optimize the execution of some of these tasks. First of all, a deep insight into customers’ behavior was needed. Then, with enough data, the support team planned to automate part of the processes they were involved in. Purelei decided to expand the help obtained from Zendesk by integrating its services with the tools provided by Ultimate32: ⓘ Ultimate is an AI-powered customer support automation platform that offers “the tools, the team, and the tactics that can take you from 0 to +60% automation across digital support channels, including chat, email, messaging, and more”. With the help of large language models (LLMs) and generative AI, Ultimate offers services in up to 109 languages. The main goal of Purelei was to create a proactive support system that would encourage self-service and free-up the human agents, while maintaining quality of personalized customer experience. 3.9.3. AI-based solution Getting the most out of the data Zendesk equipped Purelei's support team with a help center designed for multichannel, multilingual communication. Service chat, now working 24/7, along with other transactional 32 https://www.ultimate.ai/ 31 https://www.zendesk.com/ 130 | 249 data provided knowledge about consumers’ activities, preferences and most frequently encountered problems. To access all these information, Zendesk Explore33 was introduced: ⓘ Zendesk Explore processes and analyses data of customer experience within different communication channels to generate reports of real-time activities, encountered issues and possible interests of a particular client. It reduces response time, which results in more solved requests, and provides more accurate content that is backed by insights from collected data. These insights became a basis for most of Purelei’s customer service decisions. Getting to know the patterns and trends of support requests enabled the employees to adopt a proactive attitude towards the clients. The foreseen fulfillment challenges or potential concerns could now be addressed in advance, providing customers with “the greatest added value with the lowest contact hurdle”. Automating service Purelei was now equipped with a responsive help center and a sizable knowledge base. Their proactive customer care approach, however, not only required much work on the part of the support team, but also involved mostly repetitive tasks. Despite having an exhaustive list of FAQs, the help center was still flooded with simple inquiries. Although Zendesk does include in-house chatbots in their offer, it has also a broad offer of integrations with partner companies. Based on the suggestion by Zendesk’s team itself, Purelei decided to include Ultimate’s tools34 in their work: ⓘ Ultimate Zendesk integration combines extensive CRM knowledge with optimizing capabilities of virtual agents. It allows for automated ticket handling, interactions with customers and workflow monitoring, while not sacrificing advanced personalisation coming from data analysis. Ultimate's conversational AI is developed through the company’s own multilingual GPT model. By choosing the integration-based solution, Purelei was able to launch their chat automation within 3 weeks. Ultimate, as opposed to simpler forms of automation, was not solely dependent on predefined buttons. In the upgraded chat windows, customers were able to input free text, as well as images, that were then processed by the bot. This meant that they were able to engage in much more complex conversations. Due to integration, Ultimate had access to Zendesk’s data, including their CRM and order management system (OMS). Therefore, chat interactions, although automated to some extent, still benefited from detailed customer information to generate personalized messages. As for the simple requests, these were automated completely. The customer experience maintained its quality, all while requiring less repetitive effort from the support team. 34 https://www.ultimate.ai/integrations/zendesk 33 https://www.zendesk.com/service/analytics/ 131 | 249 3.9.4. Results The implemented solutions led to positive outcomes on both sides of Purelei’s transactions. A majority of clients opted for automated self-service, with a deflection rate surpassing 50%. Resolution times dropped. The employees could now save their attention and time for more complex issues, while customer satisfaction was claimed to remain stable over the 93% mark. Further, yet undisclosed benefits were attributed to the accompanying Ultimate Shopify integration. 3.9.5. Employee skills needed to adopt the solutions “No developers, no coding, no worries.” - concludes the promotional video of Ultimate one-click integration into Zendesk. Nevertheless, it is worth noting that in the case of Purelei the in-house team included “2 experts for automation processes and reporting”. One of them even mentioned the benefits of her recognising links between the processed customer data, but did not specify if it was done in cooperation with or in spite of AI tools. As for the process of implementing the solution, the team was “kept in the loop” when it came to the data imported for testing. 3.10. Repeat 3.10.1. Description of the company Source: https://www.semrush.com/company/stories/repeat/ Name Country Size Repeat France <50 employees Description: Repeat is an e-commerce company, based in Paris, selling sustainable menstrual underwear. Motivated by care for the environment and customers’ economic stability, they offer reusable products to replace sanitary towels, tampons and cups. Goal: Launch a new eCommerce brand in paid and organic channels. Strategy: computer vision, image classification 132 | 249 Technology provider: Semrush Tools: Backlink Analytics, PLA Research, Site Audit, Keyword Overview, Keyword Magic Tool, SEO Writing Assistant 3.10.2. The need Menstrual underwear in France is a trending market. When starting out their digital store, the team behind Repeat, with 3 members and limited budget, was facing a challenge of attracting the customers to their product. They needed a marketing strategy that would ensure a long-term reach of the company’s offer. Inexperienced in the field of marketing, Repeat founders tried different approaches and channels to sell the underwear. They started out partnerships with influencers in paid media but without the adequate insight into performances the method proved to be time-consuming and generated too big of a cost without sufficient and timely ROI. The idea to market through paid traffic on Google and Facebook required technological knowledge that the team was lacking. Without efficient keywords, this strategy would not bring the desired immediate results. To go beyond influencers and paid advertising, Repeat wanted to channel their marketing through organic traffic. For this to be effective, however, they needed a structured plan that would ensure long-term sustainability. To analyze the company’s possibilities and develop a strategy that would fit them, the team reached out for help from Semrush35 with their broad range of AI-based tools. ⓘ Semrush is the online visibility management and content marketing SaaS platform. It provides services such as website auditing, keyword research, backlink optimization, competitive analysis, content creation and ranking monitoring. The main goal of Repeat was to develop a long-lasting marketing strategy that would involve multiple channels and bring notable results in the website's traffic performance. 3.10.3. AI-based solution Finding the verified influence To limit the risk of investing in the wrong paid channel of communication, Semrush experts convinced Repeat to search for solutions that already worked for the company’s competitors. 35 https://www.semrush.com/ 133 | 249 Using Semrush’s Backlink Analytics36, the team was able to search through other firm’s backlink portfolios and provide a list of sites and bloggers that efficiently connected users to e-stores. Repeat could then reach out to potential partners with collaboration proposals. Backlink analysis also allowed for successful affiliate programs, in which some partners received a % from sales of the products that were purchased by their audience through the provided link. ⓘ Backlink Analytics - composed of a backlink database and discovery tool, it enables evaluation and comparison of link profiles of any domain. Provides up-to-date insight into backlinks’ type, lifespan, quality and monitors marketing strategies of competing firms. Paid advertising Paid campaigns remained a part of the strategy to provide initial traffic for the website. Although the main channel of Repeat’s advertising was Facebook, Semrush tools helped the company plan and run a successful campaign in Google Shopping. Semrush’s PLA Research37 was used to create a list of valuable keywords based on the metrics and strategies of competitors. The keywords could then be used in copywriting for the implemented paid campaigns. ⓘ PLA Research - product listing ads analysis for optimizing Google Shopping campaigns. Enables researching competitors’ rankings, strategies along with detailed data on ads and keywords. Organic traffic Competition analysis proved that other companies were benefiting from efficient SEO strategies. Domain Overview38 gave insights into channels that generated traffic for Repeat’s rivals. The firm became convinced that improving organic traffic could truly complement their strategy and provide its sustainability. ⓘ Domain Overview - providing a full overview of a domain and its online visibility. Comprises data on representation in specific markets, top keywords and growth trends for different traffic channels. Also includes competition analysis and comparisons by country or market type. Building organic traffic required several refinements of Repeat’s website. Through Site Audit39 Semrush conducted checks of site’s health in search of problems that would influence crawling and Google rankings. 39 https://www.semrush.com/siteaudit/ 38 https://www.semrush.com/analytics/overview/?searchType=domain 37 https://www.semrush.com/analytics/pla/positions/ 36 https://www.semrush.com/analytics/backlinks/ 134 | 249 ⓘ Site Audit - SEO analysis tool, crawling and checking every page on a given website. It produces regular reports on website’s issues, their levels of priority and solutions to fix them. Keyword Overview and Keyword Magic Tool40 helped Repeat find the keywords matching their actual needs. They started with niche keywords (matching searches with a more specific intent) that were easy to be positioned high in rankings, then started targeting more popular search terms. ⓘ Keyword Overview - keyword research database with general information on a given keyword, including search volume, difficulty, intent and CPC value. ⓘ Keyword Magic Tool - topic-based keywords suggestions for optimized and structured content organization. Shows possible links between more general topics, subgroups and related searches, including long-tail keywords. Finally, Repeat was able to create its website’s content in line with the most efficient keywords and users’ comfort. Valuable blog posts could also establish Repeat as experts on female well being and environmental consciousness. With the help of Semrush’s SEO Writing Assistant41, the company could generate articles that were both interesting for their customers and met SEO requirements. The tool provided content briefs for blog posts writers, checked finished articles and returned suggestions for improving them. ⓘ SEO Writing Assistant - a smart writing editor that helps optimize copywriting for engagement and SEO. Assesses the readability of documents, consistency of the tone of voice and checks for plagiarism. Suggesting SEO improvements based on real-time data. 3.10.4. Results The final strategy composed of influencer marketing, paid traffic and organic traffic led Repeat to 3900% traffic growth over 18 months. The organic traffic itself was increased by 45% thanks to just 10 published blog posts. Company’s revenue grew by approx. 300% and the domestic market share of Repeat became substantial. 41 https://www.semrush.com/swa/ 40 https://www.semrush.com/analytics/keywordoverview/, https://www.semrush.com/analytics/keywordmagic/start) 135 | 249 3.10.5. Employee skills needed to adopt the solutions Repeat team members were lacking useful online marketing knowledge. They started building their strategy in the course of trials and errors. Despite recognizing the importance of organic traffic, they were not able to get valuable insights to perform the necessary improvements of their website. 3.11. Sinnerup 3.11.1. Description of the company Source: https://raffle.ai/customers/sinnerup Name Country Size Sinnerup Denmark ~250 employees Description: Sinnerup is a Danish retailer specializing in interior design and lifestyle products. The company’s offer includes popular brands along with their own in-house designed furniture and home accessories. After over 50 years on the market, Sinnerup now runs 14 brick-and-mortar stores in Denmark and Germany and an e-commerce website. Goal: Stable control over customer experience during diverse marketing campaigns. Strategy: Intelligent search functions built into the website. Technology providers: Raffle Tools: Raffle AI Search, Instant Answers, Raffle Insights 3.11.2. The need Sinnerup describes its customers as the “biggest source of knowledge and inspiration” and openly puts their satisfaction at the top of priorities list. While it may not be surprising to hear such bold statements from a company with long traditions, it certainly elevates expectations of shopping experience. 136 | 249 3.13. Velasca 3.13.1. Description of the company Source: https://www.domo.com/customers/velasca Name Country Size Velasca Italy <250 employees Description: Velasca is a footwear and clothing company based in Milan. The artisanal brand is developed with an emphasis on a direct supply chain, ensuring quality, fair value of work and close relationships with customers. The products are sold in multiple locations across Europe, as well as through Velasca’s online store. Goal: Collecting and organizing information from different sources to create data-driven strategies for future development. Strategy: Centralising data, increasing its operational accessibility and automating key tasks. Technology provider: Domo Tools: Business intelligence and analytics, Data integration, Automation apps 3.13.2. The need The mission to deliver products directly from artisans to the client may seem as a clear process that is easily traceable. As Velasca’s operations grew, however, with increasing numbers of customers purchasing shoes and clothing from different locations, getting an insight into the efficiency of the workflow proved difficult. Customer data was indeed collected across numerous online and offline channels, but lacked the proper organization that would provide accessible knowledge for the company’s employees. The widespread distribution of products was faced with diverse demand trends that had to be met with optimal inventory management and production scheduling, let alone accurate marketing campaigns. Making decisions based on the acquired data required time and human resources. Basic customer cohort reports could be done within as much as one week. Still, the results were not satisfying. Employees worked with incomplete data and by the time the analyses were finished, 143 | 249 they could have already been slightly outdated. Lastly, uninformed supply chain management put a threat to the direct-to-customer business model. Aware of these challenges, Velasca seeked a technological solution that would organize the company’s data and make it available on time. This resulted in the cooperation with Domo Data Experience Platform52. ⓘ Domo is “a cloud-based platform that provides simplified, near real-time access to data.” Delivering data management infrastructure, it enables integration and control over information. Among its tools there are also low-code and pro-code automation apps, as well as business intelligence (BI) and analytics features. The main goal of Velasca was to integrate a data hub into its operations for advanced, consistent analytics and informed decision making throughout the organization’s units. 3.13.3. AI-based solution Accessing information To gain access to all of the company’s data, Velasca used the help of Domo’s Data Integration feature53. ⓘ Domo’s Data Integration tools “make disparate data assets accessible and available for business analysis”. They enable connecting large volumes of data from multiple sources and processing them within the cloud infrastructure. Native integrators and connectors simplify and speed up accessing the right information at any moment. Through integration, Velasca was able to break down its data silos and connect their contents to a single environment. The information from every online and offline channel could now be easily synced and shared between different teams. Transactional data, supply chain details and customer insights were all centralized and ready for further analysis. Deeper into data With all the data in the right place, the company was now capable of fully benefiting from the knowledge they provided. For the analyses to be done on time and with proper accuracy, another Domo’s solution was introduced54. ⓘ Domo’s Business Intelligence and Analytics enable comprehensive data exploration. Backed by AI algorithms, these tools perform advanced analyses to generate reports 54 https://www.domo.com/business-intelligence 53 https://www.domo.com/data-integration 52 https://www.domo.com/ 144 | 249 and predictions. With accessible visualizations, the insights provide a basis for near real-time decision making. BI and Analytics resulted in more accurate data that was available at all times and almost instantly. With accessible dashboards and visualizations employees were able to better understand the shared data. Gross and net orders were organized with multiple metrics, such as purchase time, location and product details. Inventory and production processes could now be modified in accordance with any changes in demand. Customer service team gained insights about clients’ preferences, needs and habits related to both offline and online purchases. Top-performing products and major business drivers were recognised. Along with monitoring of current marketing campaigns, this helped optimize the company’s offer and outreach to provide more and better service. A bit of automation The last, yet crucial change to Velasca’s operations was introduced through automation included in Domo’s App Creation Tools55. ⓘ Domo’s Business Apps “are highly curated data experiences that enable anyone to automate business processes and confidently take action using flexible App Creation Tools.” They enable both low-code and pro-code app development using provided frameworks and infrastructure. One of the key features is the integration of automated workflows that perform assigned tasks previously done manually. As any other company, Velasca was forced to carry out repetitive tasks that engaged resources and took a significant amount of time. Although manual processing of data required constant support of engineers it still was not free of human errors. Automation took over part of the responsibility for connecting, combining and visualizing data. Due to that, time and workforce was saved for creative and customer care tasks and more refined decision making. 3.13. 4. Results The implemented solutions led to the measurable outcomes that were accessible thanks to the Domo’s technology itself. Average item value (AIV) became 7.8% higher and average order value (AOV) reached 12.3% growth. Through monitoring of customer lifetime value, Velasca could discover that in the course of 1 year clients spent 16.4% more on their orders. With 23% cost reduction and 18% time saved, the company’s total revenue increased by 61% in over a year. 55 https://www.domo.com/business-apps 145 | 249 3.13.5. Employee skills needed to adopt the solutions Domo’s tools not only organized and processed Velasca’s data for the company’s profit but also enabled its availability for the workers. With personal dashboards and exploration features, employees could understand the information they worked with and improve their data literacy. The need for technical skills, on the other hand, appeared to have been reduced. Low-code automation apps enabled performing tasks with smaller engineering teams and less programming operations. 3.14. Best practices - Innovative Retail Laboratory Sources: interview, information collected during a dedicated workshop organized for the INAiR consortium by the Innovative Retail Laboratory, website https://www.innovative-retail.de/ A special case in the research of AI-powered retail can be found in Saarbrücken, where the Innovative Retail Laboratory is located. A part of the German Research Center for Artificial Intelligence (DFKI), the IRL is one of a kind research organization, specializing in "application-oriented developments for the retail of the future”. As a public-private partnership, they cooperate with government organizations and universities, while maintaining a strong connection to the business. While the IRL conducts the majority of operations in the research centers in Saarbrücken and Berlin the employees also travel through the country, presenting and explaining newly developed solutions. Since 2007 The Laboratory has been contributing to the implementation of new technologies and methods in the retail industry, with artificial intelligence as an important point of focus. Its interdisciplinary team is composed of experts in engineering, programming and psychology, ensuring that any innovation would meet a complex set of demands of the market, society and government. The IRL is very keen to use AI as a research tool. With methods based on advanced analytics, machine learning, deep learning, computer vision and chatbots, the organization aims to resolve complex problems of the retail industry. The multidisciplinary team goes through the entire process of developing a new custom technology. They conduct technology scouting in search of topics trending across retail businesses. First ideas can be materialized in the form of prototypes or MVPs, which are then tested and confronted with the potential users. The mix of theoretical and practical knowledge 146 | 249 resulting from the work of the IRL employees is not only used internally but shared further, during the workshops and training sessions conducted by experts. While the scope of the Innovative Research Laboratory is broad, any part of the described AI development can become a point of focus. An especially interesting element of the Lab’s offer are its Research as a Service (RaaS) practices. Companies that lack knowledge or resources to test new ideas can reach out for the Laboratory’s help. If they ask a specific question, IRL is there to find an answer. The work of particular importance for INAIR is the Mittelstand-Digital Zentrum Handel - a section of the German government’s initiative of digitalising small and medium-sized companies, in this case focusing on the retail sector. The Innovative Retail Laboratory is responsible for the AI-related parts of the research, including workshops, presentations and an educational podcast. With financial support from Mittelstand-Digital, the IRL team also conducts implementation projects for retail SMEs. The Innovative Retail Lab focuses on resolving social and environmental challenges in retail. They develop systems that encourage consumers to shop with their own packaging and monitor their home inventory with smart storage shelves. Other experiments aim to redefine retail stores as “social meeting points,” making them more inclusive for elderly people. Many innovations are designed to familiarize consumers with AI technologies through fun and engaging experiences, such as VR and AR environments or gamified shopping experiences. With practical projects like finger-point-triggered product recognition and intelligent fruit crates, the Innovative Retail Lab aims to stay at the forefront of digital innovations in the retail sector. However, the organization notes that many companies first need simpler solutions. Reluctance towards technological change often means businesses operate without structured enterprise resource planning (ERP) or data awareness. Numerous improvements are needed before artificial intelligence can become a primary focus. The following topics are the primary focus of the IRL: ● Analysis of consumer behavior ● Automated inventory maintenance ● Data protection and privacy ● Influence of the Internet of Things on society 147 | 249 ● Innovative prediction systems ● Innovative marketing for user-oriented services and dialogue marketing ● Intelligent assistance systems ● Machine Learning and sensor fusion ● Navigation and kiosk systems in the supermarket ● Personalized and mobile purchasing support ● Robotics and logistics support ● Smart labels (e.g. RFID and NFC) in retail ● Control of customer flows ● Technologies to enhance the customer experience in department stores ● Fusion of digital and analog worlds Source: Innovative Retail Laboratory presentation. Figure 37 Overview of the Innovative Retail Lab’s activities Source: Innovative Retail Laboratory presentation. 148 | 249 3.15. Best practices - Footprints AI & Danubius Sources: https://www.mist.com/wp-content/uploads/Footprints-AI-Intro.pdf, https://www.romanianbusinessjournal.ro/footprints-ai-romanias-retail-media-market-to-exce ed-10-million-euros-in-2024/, https://danubius.org/en/danubius-si-footprints-ai-lanseaza-reteaua-de-retail-media-cu-ceamai-mare-acoperire-geografica-din-romania/ Another interesting project comes from Romania and is not limited to one organization, but is the result of a fruitful collaboration between two local technology companies. One of them develops advanced software, while the other is focused on supplying retailers with hardware. Together, they managed to include Romanian SMEs in the new environment of B2C business, benefiting from innovations in artificial intelligence. Footprints AI is an AI-based omnichannel retail media platform specializing in dynamic targeting based on consumer behavior. Through data-driven automation, predictive behavioral models and hyperlocal actionable insights, the company enables analysis of anonymous traffic, leveraging the potential of data from digital and physical points of contact alike. It enables bricks-and-mortar retailers to optimize their marketing strategies in a way that was previously only possible in e-commerce. Figure 38 Footprints AI’s customer data generation pattern Source: https://www.mist.com/wp-content/uploads/Footprints-AI-Intro.pdf. 149 | 249 Danubius presents itself as “one of the most important integrators of tax and payment solutions in Romania”. During its 30 years in the market, the company has been offering electronic cash registers that introduced the most recent innovations in fiscal payments (card transactions, bluetooth, OS integrations, portable printers, BlueCash). It does not come as a surprise that at some point, it would cross its path with the development of AI. The solution brought about by Footprints proved successful for the platform’s international partners. The clients, however, were mostly large retailers with adequate resources. Meanwhile, the broad offer of Danubius allowed the ECR provider to reach different kinds of businesses, which resulted in a sizable network of merchants, including the smaller ones. The cash registers turned out to be the key to introducing AI technology to Romanian SMEs. The effect of the collaboration was a new type of smart cash register. Available as a free upgrade to all the clients possessing the newest model of Danubius’ product, this equipment is able to gather transactional data and connect it to Footprints’ system. The AI algorithms enable profiling of the customers based on their activities and generate predictions on future behavior and preferences. The resulting marketing strategies can be deployed within the retail media network, launched together by Footprints and Danubius as an alternative to popular advertising platforms (Google, Meta) and traditional media (radio, TV). Importantly, all the customer data is said to be correctly anonymised, in line with the GDPR policies. As one can see from the scheme above, cash registers are only one of many potential sources of customer data, suggesting that this solution does not yet realize the full potential of Footprints AI’s system. It is, nevertheless, an important step towards democratization of the artificial intelligence technologies, allowing SMEs with physical stores to enter the competition fueled by the digital transformation of retail. Danubius CEO stated that the retailers participating in the first stage of product development can expect to “benefit from an additional income of 25%, subject to certain conditions”. 3.16. Best practices - Dare Media Sources: https://www.daremedia.pl/, interview with Dare Media’s CEO Polish e-marketing company is an example of a different approach towards AI adoption which may be interesting for retail businesses just setting off on their adventure with artificial intelligence technologies. In short, Dare Media specializes in “growth hacking strategies” through data analysis and automation, based on extensive use of popular software. Tools provided by Google Analytics, OpenAI, Semrush or Salesforce are just a few of the vast range of solutions that the firm has to offer. Its areas of interest include customer support, 150 | 249 communication channels, SEO, marketing campaigns and data-driven optimization. It may seem counterintuitive to add a middleman on top of all the financial constraints that stand in the way of MSMEs’ digital transformation. Dare Media’s strength, however, lies in its expertise - the knowledge of where to start and how to do it. The company’s service begins with an audit and a set of recommendations on possible improvements. To minimize the risk of wasted resources, many collaborations operate on a success-fee model, where the technology provider earns a share of the profits that result from improvements. Initial testing and scalability also help smaller businesses that are not yet convinced towards AI solutions. As a marketing-focused company, Dare Media uses a range of tools for content generation. Specifically, transforming one kind of content into another is a process that can be easily accelerated and simplified through automation. Dare Media’s CEO mentions ChatGPT (https://openai.com/chatgpt/), Claude (https://claude.ai/) and Gemini (https://gemini.google.com/) as very handy when it comes to redistributing media. These tools enable generating transcripts of audio files, social media posts and blog articles. Other solutions employed by Dare Media are audiovisual generation software for pre-production (storyboards, animatics), editing (highlights, effects) and even creating whole films or songs from scratch. In these cases, the firm recommends tools such as Luma (https://lumalabs.ai/dream-machine), Midjourney (https://www.midjourney.com/home), Runway (https://runwayml.com/), Storyboarder (https://storyboarder.ai/), Suno AI (https://suno.com/) and Topaz (https://www.topazlabs.com/). In some instances Dare Media uses its subscriptions of the abovementioned services to generate a broad, comprehensive offer for its own clients. Whole systems of automated CRM, SEO, sales or paid advertisement may be chosen, or, as is often the case with MSMEs, smaller improvements remain a lower-budget option. Dare Media claims to be keeping track of the latest innovations and updating their toolsets accordingly with new but verified software available on the market. When it comes to data analysis, however, there is more input from the company itself. Based on ChatGPT technology, Dare Media team creates their own scripts that enable collecting and processing customer data from different channels. Reviews spread all over the social media, forums and e-stores can be easily organized and put together with market analysis, fueling informed decision making as well as sales or marketing strategies. 151 | 249 According to Dare Media’s CEO, even the more technologically advanced firms may need “guidance on integrating AI and digital tools into their operations”. Together with automation fears and concerns about data security, these skill gaps should be addressed through “training and reassurance”. Getting a professional overview and testing of possible AI implementation may be a good first step for MSMEs just entering the new era of retail. After that, it is for them to decide whether to keep outsourcing the whole service or continue the upskilling process for developing in-house solutions. 3.17. Towards classification of models of AI adoption and integration in retail companies Table 10 Main technologies and application areas by use case Company name AI technology applied AI applications in the value chain Benefits of AI application ALOHAS • Chatbot • Augmented intelligence • Customer engagement • Operations optimisation • Good service (greater availability and shorter response time due to chatbot assistance) • Automation of customer service (answering frequent inquiries) Feed. • Chatbot • Insight engine • Speech recognition • Augmented intelligence • Customer engagement • Inventory management • Knowledge and insight management • Operations optimisation • Good service (personalisation, omnichannel connection) • Automation of customer service (order updates, resolving simple issues) • Optimisation (workflow monitoring, chain supply monitoring with real-time reports) Grünewald • ML • Deep learning • RPA • Inventory management • Knowledge and insight management • Operations optimisation • Prediction (forecasting demand and sales trends) • Automation (generating orders matching demand) Kerrigans • Chatbot • Augmented intelligence • Customer engagement • Knowledge and insight management • Automation (recommendations, changes in stock and prices) • Good service (personalisation, 152 | 249