Akeneo Survey Exposes the Data Burden Behind AI Commerce

Confidence in AI readiness is running ahead of the systems and data required to support it.

That is the tension highlighted in new research from B2B eCommerce Association vendor member Akeneo. Its survey of 1,000 senior IT decision-makers across the United States and Europe found that 95% believe their organisations are ready for AI-driven commerce.

Yet 44% of US respondents said data preparation consumes more than half of the total effort involved in their AI projects.

The findings come from Akeneo’s Agentic Commerce Reality Check Report. As vendor-sponsored research, the results should be viewed in that context, but the operational challenge will be familiar to many manufacturers and distributors. Read the Akeneo announcement.

AI is exposing problems that already existed

AI did not create fragmented product records, inconsistent taxonomies or disconnected supplier feeds. It is making those problems harder to ignore.

A conventional eCommerce experience can sometimes work around incomplete product information. A knowledgeable salesperson may recognize an incorrect attribute. A customer-service representative may know that two part numbers refer to the same product. A buyer may call when they cannot find what they need.

An AI agent does not have that institutional knowledge unless the organisation makes it accessible, structured and reliable.

For manufacturers and distributors, this becomes particularly difficult when product information sits across PIM, ERP, DAM, commerce platforms, spreadsheets and supplier systems. The data may be correct in each system for its original purpose but inconsistent when combined for a new AI use case.

Regional pricing, units of measure, compatibility information, compliance requirements and product relationships add another layer of complexity.

The cost of rebuilding the foundation

The most important finding is not simply that data preparation requires substantial effort. It is whether organisations repeat that work for every project.

A team may clean a product dataset for an AI search initiative and then repeat much of the exercise for automated content generation, sales assistance or an agentic purchasing experience. That creates project-level progress without building a reusable commerce foundation.

Akeneo’s research also found that 87% of respondents expect AI budgets to increase over the next one to two years. Among US respondents, that figure rises to 92%.

More investment will create more demand for the same underlying product, pricing and customer information. If each initiative requires another round of reconciliation, the cost and complexity will continue to compound.

The stronger approach is to treat data preparation as shared infrastructure. Product attributes, classifications, relationships, sources and approval rules should become reusable across people, channels and AI applications.

Product discovery raises the stakes

The data problem becomes especially important as buyers use AI assistants and generative search tools to research products.

Akeneo reports that 48% of US respondents are actively investing in generative engine optimisation. This reflects a wider shift in product discovery: businesses increasingly need their products to be understood by external AI systems as well as their own website search.

For a manufacturer, missing specifications or compatibility data may prevent the right product from being considered. For a distributor, conflicting descriptions or inconsistent category structures may make it difficult for an AI assistant to compare alternatives accurately.

This is different from traditional search optimisation. The objective is not simply to rank a page. It is to provide structured, trustworthy information that an AI system can interpret without inventing the missing context.

What commerce leaders should examine

The survey provides a useful reason for digital, product-data and IT teams to measure where the effort in their AI programs is going.

If more time is spent repairing data than testing and improving the customer experience, the organisation may not have an AI problem. It may have a product-information operating model problem.

Teams should examine whether data preparation produces lasting improvements inside their core systems or creates another temporary dataset for one project. They should also establish who owns product-data quality, how changes are approved and which system provides the authoritative answer when information conflicts.

The next stage of AI commerce will not be decided by access to models alone. Manufacturers and distributors also need dependable product information, resilient integrations and governance that can support multiple use cases.

AI readiness should not be measured by how many tools an organisation can deploy. It should be measured by how reliably those tools can use the information underneath them.