97% of Retailers Use AI. Only 11% Can Actually Scale It.

By Hector Herrera | May 21, 2026

Almost every retailer is using AI. Almost none of them can make it work at scale. Research from customer data platform Amperity finds that 97% of retailers use AI and plan to maintain or increase AI spending—but only 11% have the data infrastructure required to scale AI-driven personalization reliably. The gap between AI ambition and AI execution in retail is not a technology problem. It is a data problem.

That distinction matters because it changes what retailers need to spend on. Buying more AI tools will not fix fragmented customer data. Fixing fragmented customer data—which 58% of surveyed retailers describe as their current state—is harder and slower than any software purchase.

The Personalization Paradox

The Amperity study surfaces a contradiction at the heart of retail AI investment. On the consumer side:

• 73% of consumers say they are more likely to purchase when they receive genuinely personalized offers
• 79% say retailers regularly get personalization wrong

On the retailer side:

• 97% are using AI, primarily for personalization, demand forecasting, and inventory management
• 58% describe their customer data as fragmented or incomplete
• 11% have the unified data infrastructure that AI personalization systems require to function reliably

The math explains the contradiction. If 97% of retailers are deploying AI personalization on top of data that 58% describe as fragmented, then the majority of retail AI deployments are generating personalization recommendations from an incomplete picture of the customer. Consumers are experiencing exactly what the data predicts: personalization that misses because the retailer's AI does not actually know them.

What 'Fragmented Data' Means in Practice

Retail customer data fragmentation is not an abstraction. It is the result of how most retail businesses assembled their technology stacks over the past two decades.

A mid-size specialty retailer might have customer purchase history in a legacy point-of-sale system, online browsing and cart data in a separate e-commerce platform, loyalty program data in a CRM, email engagement data in a marketing automation tool, and in-store foot traffic patterns in a separate analytics system. None of these systems were designed to talk to each other. Each has its own customer identifier format. A customer who buys in-store, browses online, and occasionally uses the loyalty app appears as three or four different people in the retailer's data ecosystem.

AI personalization systems require a unified customer record—a single view that merges these data sources and resolves identity across channels. Building that unified record, what the industry calls a customer data platform (CDP), requires extracting data from legacy systems, standardizing formats, resolving identity conflicts, and maintaining the unified record in near-real-time as new interactions occur. That is a multi-year data engineering project, not a software deployment.

The 11% of retailers with the infrastructure to scale AI personalization have either completed that project or built their data infrastructure correctly from the start.

Where AI Is Actually Working in Retail

The Amperity data does not suggest retail AI is failing broadly—it suggests it is succeeding in specific, bounded use cases where data fragmentation matters less:

Demand forecasting uses point-of-sale transaction data, which most retailers have clean and accessible. AI forecasting tools that improve inventory allocation and reduce out-of-stock rates are delivering measurable returns across a wide range of retailers regardless of their data maturity level.

Pricing optimization similarly works from structured transaction data. Retailers using AI for dynamic pricing—adjusting prices based on inventory levels, competitor pricing, and demand signals—are seeing margin improvements that do not require unified customer profiles.

In-store operations—shelf replenishment alerts, checkout queue management, loss prevention—work from sensor and camera data streams that are architecturally separate from customer data systems.

The use cases that require unified customer data—cross-channel personalization, next-best-offer recommendations, customer lifetime value modeling, churn prediction—are where the 11% infrastructure gap becomes a ceiling on what AI can deliver.

The Spending Problem

Retailers are not underfunding AI. The McKinsey and ICSC 2026 report on retail AI found retailers plan to increase AI budgets significantly through 2028, with spending concentrated on customer-facing personalization tools. The Amperity data suggests a substantial portion of that spending will underperform because it is being layered on top of data infrastructure that cannot support the use cases it is funding.

The retailers likely to break out of the 11% ceiling over the next 24 months are those who treat the next phase of AI investment primarily as data infrastructure investment—CDPs, identity resolution, data governance—rather than AI tool procurement. The AI tools are already good enough. The limiting factor is what you feed them.

What to Watch

Agentic AI in retail—autonomous AI systems that manage customer relationships, pricing, and inventory with minimal human oversight—raises the data infrastructure requirement further. An AI agent making real-time decisions across a customer's purchase history, browsing behavior, loyalty status, and in-store interactions needs not just unified data but real-time unified data. The 11% who can scale today's AI personalization represent the pool from which tomorrow's agentic retail leaders will come.

For the 89% still operating on fragmented data: the investment thesis for AI is still valid, but the first dollar should go to data infrastructure, not the AI tool running on top of it.

Hector Herrera covers AI in retail and consumer sectors for NexChron.