Chapter 24: Comparative Assessment of Retail AI |
Summary |
Retail artificial intelligence has matured from experimental pilots into operational infrastructure across the world's largest retailers. This chapter examines the current state of retail AI through a comparative lens, assessing tools designed for customers against those built for employees, and evaluating the common barriers that limit effectiveness regardless of application. Customer-facing assistants from Amazon, Walmart, Macy's, and others demonstrate remarkable capabilities in personalization and convenience, yet depend entirely on the quality of underlying product data. Employee-facing tools from Lowe's, Walmart, and Home Depot improve service delivery and operational efficiency but face adoption challenges that technology alone cannot solve. Across every application, data fragmentation emerges as the single most significant constraint on retail AI performance. The chapter concludes with a detailed assessment of where retail AI stands, what distinguishes successful implementations from struggling ones, and what the trajectory toward agentic commerce implies for retailers preparing for the next phase of intelligent retail. |
|
1. Introduction: The Retail AI Landscape in 2026 |
Retail has always been a data-intensive industry. Every transaction, every inventory movement, every customer interaction generates information that, in theory, should enable better decisions. Yet for most of retail history, that data sat dormant in disconnected systems, accessible only through retrospective reports that explained what had already happened rather than informing what should happen next. |
Artificial intelligence has changed this calculus. Over the past three years, generative AI and agentic systems have moved from novelty to necessity across the retail sector. Amazon's Rufus assistant has served more than 250 million customers, with users who engage with it during shopping trips being over 60 percent more likely to make a purchase . Walmart has deployed AI tools to more than 1.5 million associates while building Sparky, a customer-facing agentic assistant designed to compress the shopping journey from intent to fulfillment . Target has attributed measurable improvements in on-shelf availability to AI-powered inventory planning, with its most important 5,000 items improving by more than 150 basis points year over year . |
These results are real. They are also unevenly distributed. The same technologies that deliver transformative results for some retailers produce disappointing outcomes for others, and the reasons for this divergence have less to do with the sophistication of AI models than with the condition of the data and systems those models depend upon. |
This chapter provides a comparative assessment of retail AI as it stands in 2026. It examines customer-facing applications, employee-facing tools, and the shared infrastructural challenges that determine whether either category of tool succeeds or fails. The goal is not to catalogue every AI deployment across the retail sector but to identify patterns that distinguish effective implementations from ineffective ones, and to assess what those patterns imply for the industry's trajectory. |
|
2. Customer-Facing Retail AI: Promise and Performance |
2.1 The Rise of Conversational Shopping Assistants |
The most visible retail AI applications are those customers encounter directly: conversational assistants that help with product discovery, comparison, and purchase decisions. Amazon's Rufus, Walmart's Sparky, Macy's Ask Macy's, and similar tools from Home Depot and Ulta Beauty represent a fundamental shift in how retailers mediate the relationship between shoppers and product catalogues. |
Amazon's Rufus, launched in 2024 and significantly upgraded through 2025, exemplifies the current state of the art. Built on Amazon Bedrock and drawing on large language models from Anthropic and Amazon's own Nova family, Rufus handles everything from broad questions that inform a shopping trip to highly specific product queries . The assistant can compare products, answer granular questions about specifications, and make personalized recommendations based on account memory that tracks individual shopping patterns and stated preferences. Amazon reports that Rufus has driven measurable conversion improvements, with users who engage with the assistant during a shopping session being over 60 percent more likely to complete a purchase . |
Walmart's approach with Sparky emphasizes what the company calls 'agentic depth.' Rather than simply answering questions, Sparky is designed to plan and act across multiple steps. A customer can state a goal such as hosting a cookout, and Sparky will work across Walmart's catalogue, inventory, pricing, and fulfillment systems to assemble a complete solution . This represents a meaningful evolution beyond single-turn question answering toward genuine task completion. |
Macy's Ask Macy's, developed in partnership with Google Cloud, was built and scaled with unusual speed. The team went from concept to beta in four weeks, reaching 50 percent of site users within a day of launch and full deployment a week later . The assistant incorporates multimodal capabilities, including virtual try-on that allows customers to upload photos and see how garments might look on them, with configurable backgrounds for different contexts. |
|
2.2 Personalization at Scale |
The defining promise of customer-facing retail AI is personalization: the ability to present each shopper with recommendations and experiences tailored to their individual preferences rather than broad demographic segments. This capability has advanced substantially. |
Amazon's Rufus now maintains account memory that persists across sessions, understanding customers based on their individual shopping activity. If a customer has previously indicated preferences, Rufus incorporates those details into recommendations. A parent who has mentioned children of specific ages receives age-appropriate gift suggestions. A customer who prefers organic produce sees those preferences reflected in grocery recommendations . This goes beyond collaborative filtering to something closer to genuine contextual understanding. |
ASOS has implemented conversational shopping tools built on Microsoft Azure OpenAI that help customers narrow choices and explore styles based on browsing habits and current fashion trends . Etsy has developed what it calls 'algotorial curation,' combining human curation with machine learning to extend and refine product collections in ways that remain visually coherent and relevant to individual tastes . |
The NRF Innovators Showcase for 2026 highlights several startups addressing personalization infrastructure. Vody, for instance, focuses on preparing product data for natural language search and agent-driven shopping, cleaning and enriching catalogues so that AI systems can interpret them accurately . This recognition of data quality as a prerequisite for personalization is significant; it reflects growing industry awareness that sophisticated models cannot compensate for inconsistent product information. |
|
2.3 Limitations of Customer-Facing Tools |
Despite their sophistication, customer-facing AI assistants face persistent limitations that constrain their effectiveness. The most fundamental is dependence on accurate, complete product data. As one industry analysis notes, retailers are discovering that AI initiatives often fail not because of the technology itself but because of fragmented and low-quality data . Gartner has predicted that 60 percent of AI projects without AI-ready data will be abandoned through 2026 . |
This dependency manifests in specific ways. When product attributes are inconsistent or incomplete, AI assistants struggle to make accurate comparisons or recommendations. When inventory data is delayed or inaccurate, assistants may suggest products that are unavailable or misstate delivery timelines. When customer records do not match across systems, personalization becomes unreliable or even counterproductive. |
Edge cases also present challenges. Academic research on fashion e-commerce chatbots has documented consumer frustration with assistants that provide vague or generic responses, push products inappropriately, or fail to handle nuanced queries . One study participant described a chatbot experience where 'the conversation became disengaging and uninterested,' with the system pushing certain products while failing to address the actual query . Another noted that responses felt 'robotic' and command-based, lacking the human touch that builds trust. |
These limitations suggest that while customer-facing AI has advanced remarkably, it has not yet reached the point where it can reliably handle the full range of shopping queries without human backup. The most effective implementations recognize this and design for graceful escalation to human assistance when AI capabilities reach their limits. |
|
3. Employee-Facing Retail AI: Augmenting the Workforce |
3.1 Knowledge Augmentation for Frontline Associates |
While customer-facing AI receives more public attention, employee-facing tools may ultimately prove more consequential for retail operations. These applications aim to augment the capabilities of frontline associates, enabling them to provide expert-level service regardless of their tenure or department familiarity. |
Lowe's Mylow Companion represents the most significant at-scale deployment in this category. Launched in May 2025, it was the first AI assistant deployed across a retail workforce at scale . Built in collaboration with OpenAI on the same foundation as Lowe's customer-facing Mylow tool, Mylow Companion allows associates to ask natural language questions through handheld devices and receive actionable guidance. Whether the query concerns fertilizer selection for a specific grass type or instructions for repairing a leaky faucet, the tool draws on Lowe's expert advice to deliver relevant information quickly . |
The home improvement context makes this application particularly valuable. Lowe's customers routinely ask about complex projects and thousands of different products, and the knowledge required to answer these questions comprehensively has traditionally taken years to develop. Mylow Companion compresses that learning curve, enabling newer associates to handle queries that would previously have required consulting more experienced colleagues . |
Walmart has deployed a suite of AI tools across its workforce of 1.5 million associates. These include an AI-driven task management system that reduced time team leads spend planning shifts from 90 minutes to 30, a real-time translation tool supporting 44 languages, and an upgraded conversational AI platform that handles over 3 million queries per day . The translation tool is particularly notable for its incorporation of Walmart-specific knowledge; it recognizes house brands and maintains proper references even in translation . |
Walmart's Ask Sam and Associate Super Agent tools provide associates with information about shelf locations, prices, policies, scheduling, and store KPIs . The goal, as Walmart's Greg Cathey stated, is to 'eliminate friction, simplify actions and make work more efficient' . |
|
3.2 Operational Efficiency and Task Management |
Beyond knowledge augmentation, employee-facing AI is being applied to operational workflows and task management. Walmart's AI-directed workflow tool, initially deployed for overnight stocking, provides associates with clear guidance on task prioritization based on real-time conditions . This kind of intelligent task routing represents a shift from static procedures to dynamic, context-aware work assignments. |
Target has applied AI to inventory planning with measurable results. The company's machine learning systems forecast, order, and position inventory to optimize flow from suppliers to shelves, with particular attention to high-frequency items that drive customer satisfaction . On-shelf availability for Target's most important items improved by more than 150 basis points year over year in the third quarter of fiscal 2025, with management noting that the pace of improvement accelerated each quarter . |
Walmart's supply chain AI operates at similar scale. The Walmart Fulfillment Engine uses an ensemble of AI agents and real-time decision intelligence to identify optimal fulfillment nodes, balancing speed, cost, and availability . The company's End-to-End Agentic Workflow relies on a multi-agent architecture where specialized digital agents continuously make micro-decisions about routing, driver availability, and timing, recalibrating to shifting conditions like weather and traffic . These systems have contributed to delivery windows that, for some customers, now fall under 30 minutes . |
|
3.3 Adoption Challenges and Training Requirements |
Employee-facing AI tools face a distinct set of challenges from their customer-facing counterparts. The most significant is adoption. Unlike customer tools, which users choose to engage with voluntarily, employee tools are often mandated or strongly encouraged, and their success depends on associates actually using them effectively. |
This creates a training burden. Lowe's Mylow Companion, for all its capabilities, requires associates to learn how to formulate effective queries and interpret responses appropriately. Walmart's deployment across 1.5 million associates necessarily involves substantial change management, ensuring that tools designed to simplify work do not become additional sources of friction . |
There is also the risk of over-reliance. If associates defer to AI recommendations without applying their own judgment, errors in the AI's output can propagate uncorrected. The most effective implementations frame AI as an augmentation tool rather than an oracle, providing information and suggestions while leaving final decisions to human associates who understand the specific context of each interaction. |
Research on retail AI adoption suggests that the organizational dimension of these deployments is often underestimated. As one analysis notes, putting intuitive technology into the hands of millions of associates has transformational potential, but only when paired with the strengths of those people . The technology is necessary but not sufficient. |
|
4. Comparative Analysis: Customer-Facing versus Employee-Facing AI |
4.1 Divergent Success Metrics |
Customer-facing and employee-facing retail AI are evaluated against fundamentally different criteria. Customer-facing tools are judged primarily on engagement and conversion metrics: do customers use them, do they find them helpful, and do they lead to increased purchasesAmazon's report that Rufus users are over 60 percent more likely to make a purchase during a session illustrates the kind of metric that matters for these applications . Walmart's Sparky, by contrast, is designed to compress the journey from intent to fulfillment, reducing the number of steps and decisions required to complete a shopping task . |
Employee-facing tools are evaluated on operational metrics: time savings, task completion rates, service quality improvements, and error reduction. Walmart's task management AI reduced shift planning time from 90 minutes to 30, a concrete and measurable improvement . Target's inventory planning improvements of over 150 basis points in on-shelf availability represent a different kind of metric: not time saved but availability improved . |
These divergent metrics reflect the different value propositions of the two tool categories. Customer-facing AI aims to increase revenue by improving discovery and conversion. Employee-facing AI aims to increase efficiency and service quality by augmenting human capabilities. Both are valuable, but they require different evaluation frameworks and different approaches to implementation. |
|
4.2 Data Dependencies and Infrastructure Requirements |
Both categories of retail AI depend fundamentally on data quality, but the specific dependencies differ in important ways. |
Customer-facing assistants require rich, accurate, and consistently structured product data. They need complete attribute information to make comparisons, reliable inventory data to provide accurate availability information, and coherent pricing data to avoid customer frustration. The rise of agentic commerce, in which AI agents act on behalf of consumers to search and purchase across retailers, raises the stakes further. OpenAI's framework for shopping agents, announced in late 2025, allows agents to search across retailers and complete purchases with participating merchants . For retailers to participate effectively in this ecosystem, their product data must be structured in ways that external agents can interpret accurately. |
Employee-facing tools have somewhat different data requirements. They depend on operational data: inventory levels, task statuses, scheduling information, and policy documentation. The quality of this data determines whether AI recommendations are accurate and actionable. If inventory data is delayed or inaccurate, task prioritization recommendations will be wrong. If policy documentation is outdated or inconsistent, associate queries will yield misleading answers. |
In both cases, the fundamental requirement is the same: data must be accessible, trustworthy, and connected across systems. As one analysis of retail AI failures puts it, AI is only as effective as the foundations underneath it . Poor quality data, fragmented pipelines, and inconsistent definitions do not become less problematic when AI enters the picture. They become more visible. |
|
4.3 Privacy Considerations |
Customer-facing and employee-facing AI raise different privacy considerations that affect their design and deployment. |
Customer-facing personalization requires access to individual shopping histories, stated preferences, and behavioral data. Amazon's Rufus account memory stores details about customers' families, preferences, and past purchases to inform recommendations . While this enables useful personalization, it also creates privacy risks that retailers must manage carefully. The Harvard Business Review case on Walmart's Sparky explicitly identifies the tension between personalization and privacy as a strategic dilemma, asking how far retailers should go in personalization without making customers uncomfortable . |
Employee-facing tools raise different privacy issues. AI systems that monitor associate performance, track task completion, or analyze customer interactions create workplace surveillance concerns that can undermine trust and adoption. The most successful implementations avoid framing these tools as monitoring mechanisms and instead position them as assistance that helps associates do their jobs more effectively. |
The research literature on consumer attitudes toward retail AI suggests that transparency and user control are critical for acceptance. IBM research cited by GS1 UK found that most respondents are open to AI-enabled assistance provided organizations maintain transparency, protect data, and give users control over how AI informs decisions . These principles apply equally to employee-facing tools, where trust in the technology is essential for adoption. |
|
5. Data Fragmentation: The Universal Constraint |
5.1 How Fragmentation Undermines AI Performance |
Across every category of retail AI application, data fragmentation emerges as the most consistent and significant limitation. The problem is not a lack of data; retailers have more data than ever before. The problem is that data sits in disconnected systems, structured for reporting rather than for real-time decision-making, and governed by inconsistent definitions that vary across departments and channels. |
As Concentrix analysis puts it, most retail data was built for reporting, not intelligence . Reporting environments answer questions about what happened: what sold yesterday, which region underperformed, which channel missed its target. AI requires something different: accessible, trustworthy, connected data that moves consistently across systems and reflects operational reality in real time. |
When this foundational requirement is not met, AI performance degrades in predictable ways. Inventory data updates at a glacial pace, so assistants recommend products that are no longer available. Customer records do not match across systems, so personalization based on one system's view of a customer fails to align with another system's reality. Product hierarchies vary between departments, so comparisons and recommendations operate on inconsistent taxonomies . |
The operational consequences are visible during peak periods. As one analysis of Southeast Asian retail notes, customers may receive app-based promotions but still face failed redemptions, delayed loyalty updates, or inconsistent prices between online and physical stores . These failures trace back to disconnected systems that cannot synchronize in real time. Click-and-collect orders may be delayed when e-commerce and store systems are not connected, and inventory may appear available online even when it has already been allocated elsewhere . |
|
5.2 The Economics of Data Remediation |
The fragmented data problem is well understood. The economics of solving it are more challenging. Remediating retail data requires sustained investment in integration infrastructure, data governance, and organizational alignment. These investments do not produce visible customer-facing innovations; they produce the conditions under which such innovations can succeed. |
This creates a classic underinvestment problem. Retailers face pressure to demonstrate AI progress through visible deployments, which encourages investment in applications rather than foundations. The result is a proliferation of pilots that look impressive but struggle to scale, because the underlying data infrastructure cannot support operational deployment. |
Vody's emergence as an NRF Innovator reflects growing recognition of this problem. The company's focus on cleaning, enriching, and structuring product data addresses a need that many retailers have not yet prioritized but that will become unavoidable as agentic commerce expands . If external AI agents are to navigate retail catalogues on behalf of consumers, those catalogues must be structured in ways that machines can interpret reliably. |
GS1 standards provide one framework for addressing data consistency. Global Trade Item Numbers, when combined with GS1 Web Vocabulary, ensure that products are uniquely identifiable across platforms, allowing AI tools to compare, match, and recommend items with greater accuracy . Standardized identifiers for locations and shipments serve similar functions for logistics and fulfillment AI . These standards do not solve the fragmentation problem on their own, but they provide a necessary foundation. |
|
5.3 Organizational Barriers to Integration |
Data fragmentation is often described as a technical problem, but its roots are organizational. Retail environments do not become fragmented overnight. They become fragmented incrementally, as teams solve immediate operational needs in isolation from one another. Marketing builds its own reporting. E-commerce creates new customer views. Finance maintains its own definitions. Each local solution works for its immediate purpose but adds another layer of inconsistency to the overall data estate . |
Eventually, central IT teams become bottlenecks, responsible for maintaining standards while responding to constantly growing reporting and analytics demands. Backlogs grow. Delivery slows. Business teams become frustrated and begin building their own workarounds. This is how organizations drift into 'shadow IT,' where dashboards, spreadsheets, and locally managed datasets multiply with little coordination or governance . |
The side effects arrive quickly: duplicated logic, conflicting KPIs, data proliferation, and different teams making decisions from different numbers. When the organization attempts to scale AI, these inconsistencies become critical. A customer service AI cannot provide accurate answers if different systems hold different versions of a customer's order history. An inventory AI cannot optimize replenishment if the definition of 'available inventory' varies across systems . |
Addressing this organizational fragmentation requires more than technical integration. It requires governance structures that establish shared definitions, clear ownership, and consistent processes across departments. These changes are difficult because they redistribute authority and require teams to give up local control over data that they have come to rely on. |
|
6. Emerging Patterns: Agentic Commerce and the Next Phase |
6.1 From Assistants to Agents |
The retail AI trajectory points toward increasingly agentic systems: AI that does not simply answer questions or make recommendations but takes actions on behalf of users. This shift is already visible in customer-facing tools. |
Walmart's Sparky represents an early implementation of agentic shopping. Rather than requiring customers to browse product lists and make individual selections, Sparky accepts goal statements and works across Walmart's retail stack to plan and execute complete solutions . The distinction between this and earlier conversational assistants is significant: Rufus answers questions, but Sparky completes tasks. |
Amazon's Alexa for Shopping, which absorbed Rufus in 2026, similarly emphasizes agentic capabilities. The assistant can automate reordering, track prices, alert customers to new products, and build shopping carts based on stated preferences . Amazon frames this as making the search box itself a conversational interface, with the AI able to complete multi-step tasks including comparison, cart construction, and reorder . |
OpenAI's framework for shopping agents, announced in late 2025, extends this concept beyond single retailers. The framework allows agents to search across retailers, compare products, interpret user preferences, and complete purchases with participating merchants . This vision of cross-retailer agentic commerce has significant implications for how consumers discover and purchase products. |
|
6.2 Implications for Retail Strategy |
The shift toward agentic commerce raises strategic questions for retailers. If AI agents mediate the customer relationship, what happens to brand loyalty, advertising revenue, and the customer data that retailers have traditionally used to personalize experiences |
Amazon's response illustrates one approach. By moving its AI assistant into the main search flow and investing in proprietary capabilities including price history, recommendation graphs, and account-level purchase data, Amazon aims to ensure that its own assistant is the most fluent shopper on its marketplace . The strategic logic is defensive: if external agents do the comparison and the click on a customer's behalf, Amazon's advertising business, built around sponsored placements in search results, loses its target . |
Walmart's approach emphasizes openness and interoperability. Sparky was designed to interoperate with other agents rather than locking customers into Walmart alone . This reflects a bet that being the most useful surface for agentic shopping, even if customers also use other agents, is preferable to being a closed system that agents cannot access. |
For retailers without the scale of Amazon or Walmart, the strategic calculus is different. They may lack the resources to build proprietary agentic capabilities and may depend on external platforms for customer discovery. Preparing for agentic commerce, for these retailers, means ensuring that their product data is structured in ways that external agents can interpret accurately, and that their fulfillment systems can respond reliably when agents place orders. |
|
6.3 The Data Imperative Intensifies |
Agentic commerce intensifies the data imperative that already constrains retail AI. When AI agents act on behalf of consumers, they need to navigate product catalogues, compare attributes, verify availability, and complete transactions. Each of these functions depends on structured, consistent, accurate data. |
The NRF Innovators Showcase for 2026 reflects this recognition. Alongside personalization and post-purchase tools, the showcase includes companies focused explicitly on data infrastructure. Vody's mission to prepare product data for conversational interfaces and agent-driven shopping addresses a foundational need that many retailers have not yet prioritized . Slip's machine-readable receipts provide the structured post-purchase data that AI systems need to track orders, manage returns, and deliver accurate follow-up recommendations . |
The implication is clear: retailers that invest in data infrastructure now will be positioned to participate in agentic commerce as it expands. Those that do not will find themselves increasingly dependent on platforms that have made those investments, with correspondingly diminished control over customer relationships and brand presentation. |
|
7. Case Patterns: What Distinguishes Effective Implementations |
7.1 Executive Sponsorship and Organizational Commitment |
Examining successful retail AI implementations reveals a consistent pattern of executive sponsorship and organizational commitment. These are not projects delegated to innovation teams and left to succeed or fail on their own; they are strategic initiatives with visible leadership support. |
Lowe's Mylow Companion was announced with a statement from the company's Chief Digital and Information Officer framing it as part of Lowe's commitment to 'elevate the customer and associate experience' . Walmart's AI tools for associates were introduced by the Senior Vice President of Transformation and Innovation, who framed them as central to how Walmart works . Target's inventory AI improvements were discussed in the context of the company's broader turnaround strategy, with management emphasizing that inventory reliability is a foundational lever for guest experience . |
This executive attention matters for several reasons. It signals to the organization that AI adoption is a priority, not an experiment. It provides the authority to make cross-functional changes that AI implementations often require. And it ensures that AI investments are evaluated against strategic objectives rather than isolated technical metrics. |
7.2 Integration with Existing Workflows |
Effective retail AI implementations integrate with existing workflows rather than requiring users to adopt new systems alongside old ones. Walmart's associate AI tools are available through the Walmart associate app that employees already use . Lowe's Mylow Companion operates on the handheld devices associates already carry . Amazon's Rufus, and later Alexa for Shopping, is embedded in the main search flow rather than requiring customers to navigate to a separate interface . |
This integration principle extends to data flows. The most effective implementations connect AI systems to the operational data they need without requiring manual data entry or synchronization. Walmart's fulfillment AI draws on inventory, logistics, and order data that already exists in Walmart's systems . Target's inventory planning AI connects to the systems that manage supplier orders and store replenishment . |
The alternative, requiring users to manually input data or to work across disconnected interfaces, creates friction that undermines adoption. The value of AI depends on it being easier to use than the alternatives, not harder. |
7.3 Iterative Development and Feedback Loops |
Successful retail AI implementations treat deployment as the beginning of a learning process rather than the end of a development cycle. Amazon has introduced over 50 technical upgrades and new features to Rufus since its launch, continuously refining capabilities based on usage patterns . Walmart describes performance improvements to its translation tool as coming through 'iterative feedback loops' . Macy's incorporated feedback from associates who tested early versions of Ask Macy's, using their input to make responses 'warmer and more helpful' . |
This iterative approach recognizes that AI performance in controlled testing often diverges from performance in operational deployment. Real users ask unexpected questions, encounter edge cases that developers did not anticipate, and interact with systems in ways that reveal design assumptions that do not hold. The ability to learn from these encounters and improve rapidly is a distinguishing characteristic of effective implementations. |
7.4 Human-in-the-Loop Design |
Despite advances in AI capabilities, the most effective retail implementations maintain human involvement at critical points. Home Depot's AI phone agents are designed to connect customers with human associates when needed, with the company emphasizing that 'if they need to speak with an associate, we'll quickly connect them' . Walmart's fulfillment AI describes agents acting as 'digital co-pilots,' handling routine adjustments while allowing human drivers to focus on higher-value decisions . |
This human-in-the-loop design serves multiple purposes. It provides a fallback when AI systems encounter situations beyond their capabilities. It allows human judgment to override AI recommendations when context suggests they are wrong. And it maintains the human relationships that remain central to retail, particularly for complex purchases or when customers are frustrated or confused. |
|
8. Detailed Summary and Future Trajectory |
8.1 Where Retail AI Stands in 2026 |
Retail AI in 2026 is simultaneously more advanced and more constrained than public narratives suggest. The capabilities of customer-facing assistants like Amazon's Alexa for Shopping, Walmart's Sparky, and Macy's Ask Macy's are genuinely impressive. They handle natural language queries, maintain context across interactions, personalize recommendations based on individual behavior, and increasingly take actions rather than simply providing information. Employee-facing tools like Lowe's Mylow Companion and Walmart's suite of associate AI applications demonstrably improve service quality and operational efficiency. |
At the same time, these systems operate under significant constraints. Product data quality limits the accuracy of recommendations and comparisons. Inventory data latency undermines real-time availability information. Customer data fragmentation prevents consistent personalization across channels. These constraints are not technological; they are infrastructural and organizational, and they will not be resolved by advances in AI models alone. |
The comparative assessment reveals a consistent pattern: the effectiveness of retail AI depends less on the sophistication of the AI itself than on the condition of the data and systems it depends upon. Retailers with well-integrated data infrastructure achieve better results from relatively simple AI applications than retailers with fragmented infrastructure achieve from sophisticated ones. |
|
8.2 The Data Foundation Imperative |
The most important insight from this assessment is that data foundation work is not preliminary to AI deployment but integral to it. Retailers that treat data remediation as something to do before AI projects can begin will find themselves perpetually postponing AI initiatives, because data is never perfect. Retailers that treat data remediation as an ongoing component of AI deployment, with continuous improvement driven by the specific needs of AI applications, are better positioned for success. |
This requires a shift in how retailers think about data investments. Data integration and governance have traditionally been framed as IT infrastructure costs, necessary but not value-generating. As AI becomes more central to retail operations, these investments become directly linked to revenue and customer experience outcomes. The business case for data remediation becomes clearer when connected to the AI capabilities it enables. |
GS1 standards provide one framework for structured, interoperable data that AI systems can rely upon . Retailers that adopt these standards, and that invest in the internal data governance needed to apply them consistently, will be better positioned for both current AI applications and future agentic commerce. |
|
8.3 The Trajectory Toward Agentic Commerce |
The trajectory toward agentic commerce is clear, though its pace and ultimate form remain uncertain. Customer-facing assistants are evolving from question-answering to task completion. External AI agents are beginning to mediate product discovery and purchase across retailers. The strategic implications for retailers are significant. |
Retailers that control customer relationships and data will be positioned to build proprietary agentic capabilities that keep customers within their ecosystems. Amazon's approach with Alexa for Shopping illustrates this strategy, leveraging access to price history, recommendation graphs, and account data that external agents cannot match . Walmart's approach emphasizes interoperability, betting that being the most useful fulfillment layer for agentic commerce is preferable to trying to keep agents out . |
For retailers without the scale or resources of Amazon and Walmart, the strategic imperative is different. Preparing for agentic commerce means ensuring that product data is structured in ways that external agents can interpret, that fulfillment systems can respond reliably to agent-placed orders, and that the retailer's value proposition remains visible even when AI agents mediate the customer relationship. |
|
8.4 Employee-Facing AI and Workforce Transformation |
Employee-facing retail AI will continue to expand, but its success will depend on how well retailers manage the workforce dimensions of deployment. The technology can augment associate capabilities, improve service quality, and reduce time spent on routine tasks. But it can also create anxiety about job displacement, generate resistance if perceived as surveillance, and fail through lack of training and support. |
The most effective implementations frame AI as a tool that makes associates' jobs better, not as a replacement for associate judgment. Walmart's framing of AI tools as eliminating friction and simplifying actions, paired with the company's investments in pay, benefits, and training, reflects this approach . Lowe's positioning of Mylow Companion as elevating associate knowledge regardless of tenure similarly emphasizes augmentation rather than replacement . |
As AI capabilities expand, the nature of retail work will continue to evolve. Associates will spend less time on information retrieval and routine tasks, and more time on the interpersonal and problem-solving aspects of service that AI cannot replicate. The retailers that manage this transition effectively will be those that invest in their workforce alongside their technology. |
|
8.5 Conclusion |
Retail AI has moved decisively from experiment to infrastructure. The question facing retailers is no longer whether to deploy AI but how to deploy it effectively. The comparative assessment in this chapter suggests that effectiveness depends less on the choice of AI models or platforms than on the condition of the data and systems that AI depends upon. |
Customer-facing and employee-facing AI applications have different value propositions and different implementation challenges, but they share a common foundation: accurate, accessible, connected data. Retailers that invest in this foundation, even when the returns are not immediately visible, will achieve better results from AI than retailers that focus on applications alone. |
The trajectory toward agentic commerce will intensify these dynamics. As AI agents increasingly mediate product discovery and purchase decisions, the quality of product data and the reliability of fulfillment systems will determine which retailers can participate effectively in this new commerce ecosystem. The retailers preparing for this future are investing not in AI models but in the data infrastructure that makes those models useful. |
The limitation that runs through every retail AI application, from the most sophisticated customer assistant to the simplest inventory tool, is data fragmentation. Addressing this limitation is not a prerequisite to AI success; it is the substance of AI success. Retailers that understand this distinction, and act on it, will be the ones that realize the full promise of artificial intelligence in retail. |
cs@easiersoft.com