Explainable AI Models for Ensuring Transparency in CPG Markets Pricing and Promotions

Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6858 Samuel Oladapo Taiwo1 et al 6858-6873 Explainable AI Models for Ensuring Transparency in CPG Markets Pricing and Promotions Samuel Oladapo Taiwo1, Oluwatosin Oladayo Aramide2, Oluwabukola Racheal Tiamiyu3 1 Rawls College of Business, Texas Tech University 2 NetApp Ireland Limited, Ireland - Network engineer (Network Layers and Storage) - MTS IV 3 Department of Economics, Georgia State University Abstract The increasing reliance on artificial intelligence (AI) for pricing and promotional decisions in the consumer-packaged goods (CPG) industry has amplified concerns about algorithmic transparency, fairness, and regulatory compliance. This paper explores the role of Explainable Artificial Intelligence (XAI) in enhancing the interpretability and accountability of AI-driven pricing systems. Using a conceptual and analytical approach, it synthesizes current literature on AI-powered pricing, XAI methodologies such as SHAP and LIME, and evolving legal frameworks governing algorithmic decisionmaking. The study compares interpretability techniques, highlights their suitability for CPG applications, and discusses organizational and regulatory implications of adopting transparent AI models. Findings indicate that explainability fosters greater managerial trust, consumer confidence, and compliance readiness, while reducing the risks of bias and reputational harm. The paper concludes with recommendations for integrating XAI from inception, establishing governance protocols, and balancing predictive accuracy with interpretability. Future research directions include causal explainability, real-time transparency, and sustainability of AI-driven promotional systems. Keywords: Explainable AI (XAI); Consumer Packaged Goods (CPG); Dynamic Pricing; Algorithmic Transparency; Machine Learning Interpretability; Fairness and Accountability. 1 Introduction 1.1 Background and Motivation The consumer-packaged goods (CPG) sector relies heavily on dynamic pricing and promotional strategies to maintain competitiveness and optimize revenue. Traditionally, these decisions were driven by human expertise, historical data analysis, and rudimentary statistical models. However, the increasing volume and velocity of market data, coupled with evolving consumer behaviors, have necessitated the adoption of advanced Artificial Intelligence (AI) and Machine Learning (ML) techniques (Cassaigne & Singh, 2001). While AI-driven pricing systems offer enhanced accuracy in forecasting and optimization, their inherent complexity often renders them opaque, creating "black-box" Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6859 Samuel Oladapo Taiwo1 et al 6858-6873 scenarios where the rationale behind specific price recommendations or promotional offers remains obscure. This lack of transparency presents challenges for CPG businesses in justifying decisions to stakeholders, ensuring fairness to consumers, and complying with emerging regulatory expectations (Borgesius, 2020)(Spiridonova & Juchnevicius, 2020). The imperative for explainability stems from the need to build trust in AI systems, enable human oversight, and facilitate responsible deployment, especially in applications with direct consumer impact (Caffo et al., 2022). Despite the proliferation of AI-driven pricing systems, most studies remain focused on accuracy and optimization rather than interpretability. This paper therefore bridges that gap by emphasizing the role of explainability as both a technological and ethical requirement. 1.2 Objectives and Scope This document examines the application of Explainable AI (XAI) models to enhance transparency in CPG pricing and promotions. The primary objective is to articulate how XAI methodologies can demystify the decision-making processes of complex AI algorithms, thereby fostering greater understanding and trust among CPG practitioners, consumers, and regulators. The analysis encompasses a discussion of various XAI techniques, their suitability for different pricing scenarios, and their potential to address concerns regarding fairness, bias, and accountability. The scope includes an assessment of the theoretical underpinnings of XAI in marketing contexts, its practical utility, and the organizational and regulatory considerations involved in its implementation. Figure 1: Conceptual Framework of Explainability in CPG Pricing Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6860 Samuel Oladapo Taiwo1 et al 6858-6873 Figure 1 presents a conceptual framework illustrating how Explainable Artificial Intelligence (XAI) functions as an interpretive layer within the Consumer-Packaged Goods (CPG) pricing ecosystem. Market and consumer data feed into an AI pricing model that produces price or promotion recommendations, which are often opaque to human decision-makers. The framework introduces an explainability layer implemented through model-agnostic tools such as SHAP or LIME that translates the AI model’s internal logic into human-understandable insights. These insights are then communicated to key stakeholders including marketers, regulators, and consumers, thereby enhancing accountability, trust, and ethical oversight in AI-driven pricing decisions 1.3 Significance of Transparency in CPG Pricing and Promotions Transparency in CPG pricing and promotions extends beyond mere regulatory compliance; it forms a cornerstone of consumer trust and brand loyalty (Kim et al., 2020). Consumers increasingly demand clarity regarding how pricing is determined, particularly in personalized or dynamic pricing contexts where different individuals may receive varying offers (Borgesius, 2020). For CPG companies, transparent AI models can facilitate internal decision alignment, allowing marketing and sales teams to comprehend and justify pricing strategies effectively (Caro & de Tejada Cuenca, 2023). Furthermore, the ability to explain AI-driven outcomes is instrumental in identifying and mitigating potential biases that could lead to discriminatory practices or suboptimal business results. Without explainability, CPG firms face increased risks of reputational damage, customer churn, and legal scrutiny, particularly as global regulatory bodies enhance their oversight of algorithmic decision-making (Spiridonova & Juchnevicius, 2020)(Tombal, 2022). Explainable AI, therefore, is not merely a technical add-on but a strategic imperative for responsible and sustainable business operations within the CPG industry. 2 Methodology 2.1 Research Design This investigation employs a conceptual and analytical research design. The approach involves a comprehensive review of existing literature concerning AI in CPG pricing, explainable AI techniques, consumer behavior, and regulatory frameworks. We synthesize insights from these diverse fields to construct a coherent understanding of how XAI can address transparency deficits in CPG pricing and promotions. The methodology prioritizes a qualitative assessment of XAI methods, comparing their characteristics, strengths, and limitations in the context of CPG applications. The analysis also incorporates a discussion of practical implications and challenges for implementation, drawing upon established theoretical models and empirical observations within the broader AI and marketing research communities. No new empirical data is collected; instead, the work builds upon the foundation of prior academic and industry contributions. Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6861 Samuel Oladapo Taiwo1 et al 6858-6873 2.2 Data Sources and Collection The data sources for this conceptual analysis primarily comprise peer-reviewed academic articles, conference proceedings, industry reports, and regulatory guidelines. These documents were systematically identified through searches of prominent scientific databases and academic search engines. Keywords such as "Explainable AI," "XAI," "CPG pricing," "dynamic pricing," "promotions," "transparency," "consumer trust," "algorithmic bias," and "marketing analytics" were used to curate relevant literature. The selection process focused on identifying publications that offered theoretical frameworks, methodological advancements, practical case studies, or regulatory perspectives pertinent to the intersection of AI, explainability, and commercial pricing strategies. Emphasis was placed on recent scholarship to capture the most current developments in XAI and its business applications. Table 1: Summary of Data Sources and Literature Categories Literature Domain Example Sources Focus AI in Pricing Erdmann et al. (2024), Cassaigne & Singh (2001) Predictive pricing, optimization Explainable AI Salih et al. (2024) SHAP/LIME methodologies Consumer Behavior Vorobeva et al. (2023) AI framing, price sensitivity Regulatory Frameworks Tombal (2022), Borgesius (2020) Fairness, compliance Organizational Adoption Caro & de Tejada Cuenca (2023) Managerial trust and analytics use Table 1 categorizes the key domains of literature informing this study, encompassing AI pricing mechanisms, explainability methodologies, consumer behavior models, regulatory frameworks, and organizational adoption. Each domain is represented by seminal works that collectively shape the interdisciplinary foundation of XAI application in pricing. The table underscores how prior scholarship converges on a common theme balancing computational efficiency and interpretability to support responsible business decisions. This mapping clarifies the theoretical scaffolding for the current conceptual analysis and identifies the most influential research areas shaping transparency in algorithmic pricing. 2.3 Model Selection and Evaluation Criteria The selection of XAI models for discussion is based on their prevalence, methodological diversity, and applicability to complex predictive tasks typical in CPG pricing. Prominent Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6862 Samuel Oladapo Taiwo1 et al 6858-6873 techniques such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) are central to this analysis due to their widespread adoption and model-agnostic capabilities (Salih et al., 2024)(Sarvesh Koli Komal Bhat Prajwal Korade, 2024). Additionally, the utility of inherently interpretable models, such as decision trees, is considered for their direct explainability. Evaluation criteria for these models in a CPG pricing context include: 1. Fidelity: How accurately the explanation reflects the underlying model's behavior. 2. Interpretability: The ease with which human users can understand the explanation. 3. Local vs. Global Explainability: The ability to explain individual predictions versus the overall model behavior. 4. Model-agnosticism: The capacity to explain any machine learning model, regardless of its internal structure. 5. Actionability: Whether the explanations provide insights that can lead to concrete business improvements or policy adjustments. 6. Robustness: The stability of explanations across slight perturbations in input data. 7. Computational Efficiency: The resources required to generate explanations, which can be critical for real-time pricing adjustments. These criteria provide a framework for assessing the suitability of XAI techniques for practical implementation in CPG pricing and promotion strategies. 3 Literature Review / Thematic Analysis 3.1 The Evolution of CPG Pricing and Promotion Strategies CPG pricing strategies have undergone substantial transformation, moving from static, cost-plus models to highly dynamic, data-driven approaches. Early strategies often involved fixed pricing, periodic sales, and couponing, primarily informed by production costs, competitive benchmarking, and basic market research. The advent of loyalty programs and scanner data provided initial opportunities for more granular analysis of consumer purchasing patterns. Over time, the integration of advanced analytics enabled CPG firms to segment markets more effectively and personalize offers. The development of intelligent tactical decision support systems, incorporating nonlinear models, optimization, and learning algorithms, has allowed firms to make sophisticated pricing decisions in dynamic competitive environments (Cassaigne & Singh, 2001). Today, AIpowered systems can forecast demand, optimize price points across various channels, and manage complex promotional campaigns by processing vast datasets that include competitor pricing, consumer demographics, weather patterns, and even social media sentiment (Erdmann et al., 2024). Effective trade promotion management, which encompasses designing, executing, and evaluating discounts and rebates, represents a critical area where AI can enhance efficiency and return on investment. This evolution Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6863 Samuel Oladapo Taiwo1 et al 6858-6873 underscores a continuous quest for precision and profitability, which AI systems are designed to deliver. Figure 2: Evolution of CPG Pricing Strategies (2000–2023) Figure 2 illustrates the longitudinal transformation of CPG pricing strategies from 2000 through 2023. The data visualize a clear decline in traditional cost-plus pricing approaches and a concurrent rise in data-driven and elasticity-based models. The growing prevalence of AIand algorithmic-driven methods during the 2020s reflects the industry’s shift toward precision and personalization. The figure also projects the emergence of Explainable AI pricing in 2023 and beyond, marking a new paradigm in which transparency and interpretability become intrinsic components of pricing systems. This trajectory underscores the sector’s broader movement toward responsible, dataempowered decision-making. Table 2: Comparative View of CPG Pricing Models Era Pricing Logic Data Inputs Decision Support Transparency Level 2000s Cost-plus Cost, demand Manual High 2010s Dynamic POS, loyalty data Statistical models Medium 2020s AI-Driven Big data, social, competitors ML/NN Low Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6864 Samuel Oladapo Taiwo1 et al 6858-6873 2023+ Explainable AI Same + interpretability layers SHAP/LIME High Table 2 provides a comparative overview of the evolution of pricing models in the CPG sector, emphasizing the trade-offs between sophistication, data dependency, and transparency. Early cost-plus approaches offered straightforward logic but lacked responsiveness to market signals. Dynamic and statistical pricing models introduced adaptability through data analytics but limited interpretability. AI-driven models represent the peak of analytical power yet introduce opacity and bias risks. The final stage Explainable AI pricing aims to reconcile predictive accuracy with interpretability, enabling transparent and auditable decision support. This progression highlights how the industry’s pricing logic has matured from rule-based systems to autonomous, accountable intelligence. 3.2 Consumer Decision-Making Models and Price Sensitivity Building upon the technological evolution of pricing, it is equally vital to understand how consumers cognitively and emotionally respond to AI-mediated pricing decisions. Consumer decision-making models delineate the cognitive and emotional processes individuals undertake when evaluating and purchasing products. These models frequently involve stages such as need recognition, information search, alternative evaluation, purchase decision, and post-purchase behavior. Price sensitivity, a core component of these models, describes how consumer demand for a product change in response to price fluctuations. Factors influencing price sensitivity are diverse, including perceived value, brand loyalty, availability of substitutes, income levels, and psychological pricing effects. AI models can predict consumer responses to price changes with considerable accuracy by analyzing these multifarious variables. However, the black-box nature of many advanced AI algorithms can obscure the specific drivers behind predicted price sensitivity or promotional effectiveness. For instance, an AI might recommend a certain discount for a product without revealing whether this is due to a predicted competitor action, a seasonal demand shift, or a specific consumer segment's historical behavior. Understanding these underlying reasons is crucial for marketers to refine their strategies and build trust. Furthermore, customer acceptance of AI-based services can be enhanced through appropriate framing, such as presenting AI as an augmentation rather than a substitution of human effort, which can improve enjoyment and perceived ease of use (Vorobeva et al., 2023). This suggests that transparency in how AI influences pricing can positively affect consumer perception and acceptance. 3.3 The Role of Explainable AI in Marketing Decision Models Explainable AI (XAI) addresses the challenge of opacity in complex machine learning models by providing human-understandable explanations of their outputs (Salih et al., 2024)(Sarvesh Koli Komal Bhat Prajwal Korade, 2024). In marketing decision models, XAI serves several critical functions. First, it fosters trust among marketing professionals who rely on AI recommendations. If a model suggests a counterintuitive pricing action, Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6865 Samuel Oladapo Taiwo1 et al 6858-6873 an explanation can provide the necessary justification, preventing managers from deviating from optimal strategies due to a lack of understanding (Caro & de Tejada Cuenca, 2023). Second, XAI assists in model debugging and improvement. By revealing the features that disproportionately influence a prediction, XAI can help identify data biases, feature engineering needs, or model errors (Choi et al., 2023). Third, XAI facilitates compliance with regulatory requirements regarding algorithmic fairness and non-discrimination. Understanding why an AI system offers different prices to different customer segments allows businesses to verify that these differences are based on legitimate commercial factors rather than protected characteristics. Techniques like SHAP and LIME, which explain individual predictions by approximating the complex model locally, are particularly relevant for understanding personalized pricing decisions (Salih et al., 2024)(Sarvesh Koli Komal Bhat Prajwal Korade, 2024). These methods convert the black box into a more digestible form, enhancing transparency and increasing end-user trust (Salih et al., 2024). 3.4 Regulatory Requirements and Transparency in Algorithmic Pricing The increasing use of algorithmic pricing has drawn considerable attention from regulatory bodies globally. Concerns center on potential anti-competitive practices, discriminatory outcomes, and consumer exploitation (Spiridonova & Juchnevicius, 2020). Laws and guidelines, such as the European Union's Digital Services Act (DSA), specifically address transparency requirements for automated systems, including recommender systems that influence consumer choices and potentially pricing (Tombal, 2022). These regulations often demand that platforms provide information on how their algorithms work and how they prioritize information or suggestions. Algorithmic price differentiation, where different prices are offered to different individuals for identical products, can lead to indirect discrimination, particularly if based on proxies for protected characteristics. Non-discrimination law aims to prohibit such outcomes, but its application to AI-driven decisions faces challenges, as algorithmic discrimination can remain hidden from consumers and regulators. Antitrust authorities are also exploring how pricing algorithms might facilitate anti-competitive agreements or coordinated behavior among firms, even without explicit collusion (Spiridonova & Juchnevicius, 2020). Therefore, CPG firms deploying AI for pricing must demonstrate explainability to ensure compliance, avoid penalties, and uphold ethical standards. The demand for transparency extends to understanding the underlying mechanisms of AI models to ensure fairness and prevent systemic vulnerabilities. 3.5 Challenges and Limitations of Data-Driven Decision Making While data-driven decision making offers considerable advantages, it also introduces several challenges and limitations. A primary concern is the "black-box" problem, where complex AI models, particularly deep neural networks, operate without providing easily discernible reasons for their outputs (Caffo et al., 2022). This opacity can hinder efforts to build trust, conduct audits, or debug unexpected behaviors. Another significant challenge arises from data shift, a phenomenon where the distribution of real-world data diverges from the data used for model training (Choi et al., 2023). This can lead to substantial performance degradation in deployed AI models, making explainability Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6866 Samuel Oladapo Taiwo1 et al 6858-6873 techniques crucial for detecting and mitigating such issues (Choi et al., 2023). Furthermore, the quality and representativeness of training data profoundly affect model fairness and accuracy. Biased data can lead to discriminatory outcomes, as seen in algorithmic price differentiation. Over-reliance on historical data without considering evolving market dynamics or consumer preferences can also lead to suboptimal strategies. The interpretability of XAI methods themselves can vary, and even established techniques like SHAP and LIME can be affected by factors such as model dependency and feature collinearity, necessitating careful usage and interpretation (Salih et al., 2024). The computational overhead of generating explanations, especially for large-scale, realtime pricing systems, can also be a practical limitation. Finally, the ethical implications of using AI, particularly concerning privacy and potential manipulation of consumer behavior, demand careful consideration alongside technical challenges. 4 Analysis / Discussion 4.1 Comparative Assessment of Explainable AI Techniques for CPG Pricing The application of various Explainable AI (XAI) techniques in CPG pricing models offers distinct advantages and trade-offs. Inherently interpretable models, such as decision trees, provide clear, rule-based explanations for their predictions. For CPG pricing, a decision tree might illustrate a path like "If competitor A's price is below $5 and product B's stock is low, then set price at $4.99." This directness is highly valuable for business users who require transparent logic. However, decision trees often struggle with the complexity and non-linearity present in real-world CPG data, potentially sacrificing predictive accuracy compared to more complex models like neural networks or support vector machines. For complex "black-box" models, model-agnostic XAI methods like LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations) are frequently employed (Salih et al., 2024)(Sarvesh Koli Komal Bhat Prajwal Korade, 2024). LIME explains individual predictions by creating a local, interpretable model (e.g., linear regression) around the specific instance. For a personalized price offer, LIME could identify that "customer's past purchase frequency" and "current promotional activity" were the most influential factors for that particular price. SHAP, based on cooperative game theory, assigns an importance value to each feature for a prediction, reflecting its average marginal contribution across all possible coalitions of features (Salih et al., 2024)(Sarvesh Koli Komal Bhat Prajwal Korade, 2024). This offers a more robust, globally consistent measure of feature importance compared to LIME's purely local approximation. In a study comparing LIME with decision trees for explaining support vector regression, decision trees demonstrated lower RMSE values in the majority of runs, suggesting better local performance in some contexts. Both SHAP and LIME possess limitations, including sensitivity to the underlying ML model and feature collinearity, which can affect the reliability of explanations (Salih et al., 2024). For CPG pricing, the choice among these techniques depends on the specific requirements for fidelity, interpretability, and computational resources. If high predictive accuracy is paramount and some level of local interpretability is acceptable, LIME or SHAP with a complex black-box model might be suitable. If interpretability is the overriding concern, and a slight reduction in predictive Journal of Computational Analysis and Applications VOL. 33, NO. 8, 2024 6873 Samuel Oladapo Taiwo1 et al 6858-6873 Sarvesh Koli Komal Bhat Prajwal Korade, D. M. A. M. O. K. (2024). Unlocking Machine Learning Model Decisions: A Comparative Analysis of LIME and SHAP for Enhanced Interpretability. In Journal of Electrical Systems (Vol. 20, Issue 2s, pp. 598– 613). Science Research Society. https://doi.org/10.52783/jes.1480 Erdmann, A., Yazdani, M., Mas Iglesias, J. M., & Marin Palacios, C. (2024). Pricing Powered by Artificial Intelligence: An Assessment Model for the Sustainable Implementation of AI Supported Price Functions. In Informatica (pp. 529–556). Vilnius University Press. https://doi.org/10.15388/24-infor559 Vorobeva, D., Costa Pinto, D., António, N., & Mattila, A. S. (2023). The augmentation effect of artificial intelligence: can AI framing shape customer acceptance of AI-based services? In Current Issues in Tourism (Vol. 27, Issue 10, pp. 1551–1571). Informa UK Limited. https://doi.org/10.1080/13683500.2023.2214353 Choi, Y., Yu, W., Nagarajan, M. B., Teng, P., Goldin, J. G., Raman, S. S., Enzmann, D. R., Kim, G. H. J., & Brown, M. S. (2023). Translating AI to Clinical Practice: Overcoming Data Shift with Explainability. In RadioGraphics (Vol. 43, Issue 5). Radiological Society of North America (RSNA). https://doi.org/10.1148/rg.220105