PIM & MDM / AI Agentic Commerce Is a New Technology Wave. The Data Challenge Isn’t.

At the moment, everyone is talking about agentic commerce. Ten years ago, it was automation. Before that, omnichannel. 

The words keep changing. The technology genuinely shifts. But sit through enough of these waves and a pattern gets hard to ignore: the companies who can act quickly on a new wave are the ones where data is easily available to support business processes, systems and tools. The laggards are usually restricted by systems that were never built to hand data out. 

We have seen this play out every time: EDI rewarded companies who could exchange structured data with partners. APIs rewarded companies whose data could be exposed programmatically. Headless rewarded companies whose product data was decoupled from presentation. Agents are rewarding the same thing right now. 

Every technology wave sorts out companies along the same line: is the data accessible, or is it hostage? New vocabulary, new vendors, identical lesson.

From ZMOT to AI Agents: The Moment of Truth is Changing

In 2011, Google put a name to something retailers were already feeling: the Zero Moment of Truth. The insight was that the decision no longer happened in the store, or even on your site. It happened earlier, in the research phase searching, comparing, and reading reviews long before anyone reached a shelf or a checkout. 

The lesson everyone took from ZMOT was about content. Be findable, collect and share reviews and publish rich product information to convince shoppers throughout the buying journey. 

Fifteen years later, the same moment exists. But the research is increasingly being done by an agent rather than the shopper. And an agent researching your product doesn't care about persuasive content. It wants structured answers. 

ZMOT is still around, it just got automated. The moment of truth is still before the storefront, still decided by information rather than experience. The difference is that the entity gathering that information no longer reads. It queries.

Why Accessible and Accurate Product Data Matters

AI agents discover and evaluate products differently from human shoppers. Instead of navigating a storefront page by page, they depend on data they can retrieve, interpret and compare. 

A human shopper compensates for bad data; clicks into a page, squints at a vague description, tolerates a stale stock indicator. Human patience has quietly subsidized bad product data for two decades. 

An agent, on the other hand, needs clear and immediate response. It asks a structured question and takes the answer it gets. If your data can't answer, it moves to a source that can. 

And those questions increasingly arrive through open protocols, such as UCP, ACP, MCP, not a browser. Your store needs to be optimized for totally different buying journeys; those visiting your storefront as well as those only leveraging an API connection.

BYOA Only Works If the Data is There to Meet It 

In the agentic commerce era, customers arrive at your store through their own agents, aka BYOA (bring your own agent). It could be a procurement agent checking your catalog against a spec, or a consumer's assistant comparing availability across five brands. 

It’s important to remember that you don't control these agents. You can't optimize your UI for them, nor campaign your way into their good graces. The only thing you control is whether your data answers when they ask. 

Which makes BYOA a data question wearing a technology costume. The agents are proliferating on their own. Being ready to be queried is the hard part.

The Product Data Challenge is Spread Across the System Landscape 

Product Information Management (PIM) is so much more than a system. Because product data is spread across technology and functions, the challenge is to gather data from multiple systems for a complete view of the product. Today, we describe this as the Master Product.
It's the complete, orchestrated record of everything your business holds about a product, from every source, mapped into one place: specs from PLM, live inventory from the ERP, customer-specific pricing from a pricing engine, media from the DAM, compliance data from wherever it lives. 

In most companies all of that exists, but not together. 

So, when an agent asks for details about a product spanning three systems, such as in stock, at this price, for this customer, compliant in this market, nothing answers. No single system holds it, and nothing is assembling it.  In short: agents don't need better content, they need access to the Master Product. A lot of AI tooling right now generates nicer product descriptions. That’s useful, but prose isn't structured data. Better copy doesn't make you queryable, it makes your catalog pleasant for humans who increasingly aren't the ones reading it.

What Prevents Companies from Becoming Agent-Ready

Companies rarely lack ambition. Instead, the blockers are structural and consistent: 

Everything routes through IT and the ERP. Critical product data sits in a system built for finance, owned and maintained by IT. Every new channel is a ticket in a queue. 

The PIM is viewed as the complete source. A PIM solution is a critical part of the product information landscape containing descriptions, attributes, images etc. But PIM alone won’t do the job. Agentic use cases will require data beyond the PIM. If pricing, inventory and supplier data exist elsewhere the challenge remains. And the belief that the PIM will solve everything can create a false sense of security – a blocker in its own.  

Data relationships were never modeled. Agents need to understand things like attribute dependencies, which variants relate to which product, what "size" means per category, which SKUs are excluded from which market, and more. Agents query across these relationships constantly, and most companies never formalized them.

How to Prepare Your Product Data for Agentic Commerce

1. Avoid a big migration. Replatforming is the slowest path to a structurally similar place. Leave the source systems where they are. 
   
   

2. Build the Master Product as an orchestration layer, not another database.
   Connect to every source, assemble in real time, on request. Don't copy data into a new system that immediately drifts out of sync. Centralizing creates a new hostage situation. Orchestrating ends the concept. 
   
   

3. Model the relationships, not just the attributes. Unglamorous work that determines whether agents can reason about your catalog at all. 
   
   

4. Then expose it through open protocols. Once the Master Product is live, UCP/ACP/MCP support is configuration, not construction. Everyone wants to start here. It's trivial if the earlier steps are done and impossible if they aren't. 
   
   

5. Let the ERP go back to its actual job. Focus the ERP on its core functions, including accounting, payroll and procurement. It was never good at being a real-time commerce platform. 

The real test isn't whether you support agents today. It's whether the next wave is a configuration exercise or another eighteen-month program.

Build for the Next Technology Wave, Not Just This One

Google described the Zero Moment of Truth in 2011 and it's still the right frame. The decision still happens before the storefront. It's just being made by something that reads schemas instead of copy. 

Companies that handled EDI well also handled APIs well. Those went headless without drama. The same companies are being queried by agents right now and most surely answer with relevant responses. Not because they predicted agents, but because they stopped locking data in systems that couldn't hand it out. Agentic commerce isn't a new chapter. It's the same chapter with a new title. 

The key take-away for commerce businesses getting ready for agentic commerce is to free your data, structure the Master Product and let anything connect. Then stop worrying about what the next wave gets called. 

In collaboration with: