Organizations may feel confident about their AI readiness, but IT leaders are confronting a more complicated reality behind the scenes. From time-consuming data preparation and fragmented product systems to governance gaps, agent sprawl, and the growing pressure to remain discoverable in AI-assisted buying journeys, discover the practical challenges keeping AI initiatives from scaling. Learn why the next phase of AI commerce will depend less on the model itself and more on the trusted data, resilient infrastructure, and operational controls surrounding it.
Table of Contents
Keywords
The most revealing moment in an AI project often arrives five minutes after the demo ends.
The product description provided looks polished, or the product recommended looks accurate and relevant.
But then someone asks: Which source did it use? Is that claim approved in every market? Does it have the current price? Who reviews the change before it reaches a live channel?
Suddenly, the future of commerce looks less like science fiction and more like a missing attribute in a spreadsheet.
That tension sits at the heart of IT departments across the globe today, and what we set out to learn more about in our annual survey of IT and technology decision-makers.
On paper, confidence is high: 95% of IT leaders say their organization is ready for AI-driven commerce, and 87% expect AI budgets to increase.
But here’s the catch: 89% also spend more than a quarter of every AI project preparing data.
Nearly two-thirds of IT leaders say their department now owns AI strategy, meaning they’re the ones responsible for making AI work inside the systems the business already relies on: ERP, PIM, DAM, supplier feeds, commerce platforms, marketplaces, and regional workflows.
A team may be able to vibe-code an impressive prototype over a weekend, but it cannot vibe-code trusted product information, resilient integrations, permissions, approvals, and audit trails across brands and markets.
AI did not create brittle integrations, local taxonomies, manual mapping, or disconnected workflows, however it is making their cost harder to ignore. Every new use case inherits that technical debt unless the underlying product information foundation is modernized.
While 95% of IT leaders say they are ready for AI commerce and 92% rate their data quality and structure as good or excellent, 53% spend between 26% and 50% of an AI project preparing data. Another 36% spend more than half.
That preparation means chasing missing attributes, reconciling duplicate SKUs, normalizing units and taxonomies, checking regional prices and compliance fields, resolving conflicting values, and adapting information for different channels.
It is repetitive, time-consuming work the wider business rarely sees. When it is done well, nothing dramatic happens; the AI simply produces an answer that is not wrong. Yet that invisible effort is often what holds the project together.
Almost every organization surveyed (99%) either has an AI governance policy or is developing one. But a policy cannot always answer the questions that matter when AI touches live product information:
Who changed the record? Which model made the recommendation? What data did it use? Who approved the output? Where was the change published?
That traceability, sometimes called decision lineage, matters when a change involves a price, warranty, compliance claim, promotion, or product description. Governance becomes useful when it is built into workflows through permissions, human review, monitoring, and audit trails, not when it exists only on a board slide.
In our survey, we found that IT automation and IT operations outrank customer-facing use cases such as conversational AI and personalized search as investment priorities. That makes sense: managing millions of attributes, products, channels, and markets has already exceeded what many teams can handle manually.
The complication is that every narrowly focused agent added to solve one bottleneck becomes another component to coordinate. Without a shared foundation, multiple agents may enrich, translate, validate, and activate the same records without understanding what the others have changed.
This is where an AI harness becomes useful. It is everything around the model (permissions, memory, guardrails, approvals, and orchestration) that allows it to act reliably inside business systems.
Seventy-nine percent of organizations are investing in or piloting Answer Engine Optimization (sometimes called Generative Engine Optimization) as product discovery shifts toward AI.
But appearing in an AI-generated recommendation is not simply a new version of ranking for a keyword. An assistant must understand the product, compare it with alternatives, verify its attributes and claims, and trust that the information is current.
Incomplete titles, missing identifiers, inconsistent specifications, unsupported claims, or stale availability can keep a product out of the consideration set entirely.
Enthusiasm for AI is not fading. Budgets are growing, ownership is becoming clearer, and organizations are actively investing.
What is holding AI back is the gap between a promising model and the operational environment required to trust it. More funding will not remove the data-preparation tax when every new project begins with the same incomplete attributes, inconsistent taxonomies, regional variations, and fragile integrations.
The next meaningful AI milestone may not be a dazzling demo. It may be a far less cinematic moment: an agent updates the correct product record, follows the right rules, routes the change to a human, and publishes accurate information to the right channel.
That is the moment AI stops being an experiment and starts becoming infrastructure.
Download the full report, The Agentic Commerce Reality Check, to explore all 10 hard truths from IT leaders and assess whether your product information foundation is ready for agentic commerce.
Akeneo’s latest global survey reveals where AI confidence collides with the realities of product data, governance, integration, and scale.
