The Data Debt Dilemma: Why Your AI Agents Are Failing at Commerce
The promise of agentic commerce is seductive: AI agents that don’t just suggest products, but autonomously handle the entire shopping journey—from discovery to checkout—on behalf of the consumer. However, early attempts to bring this vision to life have revealed a sobering reality. For many retailers, the gap between a flashy AI demo and a functional revenue stream is a missing data layer.
When AI agents operate on fragmented, siloed data, they don’t become personal shoppers; they become expensive search bars with checkout buttons attached. To win the next decade of retail, companies must stop focusing on the model layer and start solving their data debt.
The High Cost of Disconnected Data: A Case Study
The industry recently witnessed a high-profile lesson in the limitations of agentic AI. When OpenAI launched “Instant Checkout,” expectations were high for a seamless, in-chat purchasing experience. Walmart tested the feature as a checkout channel for approximately 200,000 products, but the results were stark: in-chat purchases converted 3X worse than those on Walmart’s own site.
Daniel Danker, Walmart’s EVP of product and design, described the experience as “unsatisfying,” leading the retailer to back out of the initiative. OpenAI subsequently rolled back the feature, admitting that the initial version of Instant Checkout lacked the necessary flexibility. The shift that followed—moving toward retailer-controlled apps inside ChatGPT—highlights a fundamental truth: the checkout button is the easy part; the context behind it is where the struggle lies.
The Scale of the Agentic Opportunity
Despite these early mishaps, the economic potential of agentic commerce remains massive. Bain projects that the agentic commerce market in the U.S. Alone could reach $300 to $500 billion by 2030. This represents roughly 15% to 25% of overall e-commerce, meaning a significant portion of consumer journeys will soon involve an AI agent acting on a customer’s behalf.
The problem is that most retail technology was built for a “clean” shopping session: a user arrives, browses, adds to a cart, checks out and leaves. But real human shopping is messy. A customer might research on a phone during a commute, add items to a cart on a laptop, compare prices via a marketplace, and finally buy in-store. If an AI agent treats each of these touchpoints as a fresh session, the experience breaks.
Where Fragmented Data Ruins the Experience
- Irrelevant Recommendations: An agent suggests a product the customer already returned last month.
- Inventory Failures: An agent commits to a delivery window that the supply chain cannot honor because inventory visibility is siloed.
- Conflicting Offers: A promotional offer is applied to a product that is already in the customer’s cart on another device.
- Logistical Friction: A single bundle ships in multiple pieces from different fulfillment nodes due to a lack of unified visibility.
These are not AI failures; they are data problems dressed up as AI problems. According to a 2025 Gartner survey of technology leaders, half of the respondents report that their organizations lack the technical and data stack readiness required to deploy AI agents effectively.
Context Intelligence: The New Competitive Moat
In the near future, foundation models and protocols—such as OpenAI’s ACP or Google’s Universal Commerce Protocol—will be commoditized. Every retailer will have access to roughly the same AI “brains.” The only way to differentiate a brand experience will be through the quality of the context provided to those models.
This is what Reltio defines as “context intelligence”: the ability to unify customer, product, and operational data into a single, real-time foundation. When an agent has a trusted, persistent view of a customer’s identity across all channels, the interaction moves from generic to personalized.
The CIO’s New Priority List
For technology leaders, agentic commerce necessitates a shift in the strategic agenda:
- Identity Resolution: This is no longer a back-office project. If an agent cannot recognize a customer across the web, app, store, and third-party surfaces like Gemini or ChatGPT, meaningful personalization is impossible.
- Real-Time Synchronization: Batch updates are no longer sufficient. Agents act on data at the moment of the query; stale data results in broken promises to the customer.
- Experience-Driven Unification: Consolidating data is no longer about operational efficiency—it is the primary layer that determines whether an AI experience feels coherent or fragmented.
Key Takeaways for Retail Leaders
| The Old Model (Session-Based) | The New Model (Agentic) |
|---|---|
| Assumes a linear “Arrive $rightarrow$ Buy $rightarrow$ Leave” path. | Views the customer as an ongoing thread across multiple devices. |
| Relies on overnight batch updates for inventory. | Requires real-time synchronization to prevent broken promises. |
| Treats identity resolution as a back-office task. | Treats cross-channel identity as a customer-facing capability. |
| Focuses on the AI model’s capabilities. | Focuses on the data context informing the model. |
Looking Ahead
Investment in the model layer only amplifies the consequences of existing data gaps. The more powerful the AI, the more exposed the underlying data debt becomes. The retailers who win the agentic era won’t be those with the fastest models, but those with the most trusted, unified view of their business and their customers.
