On June 17, Shopify shipped its Spring ‘26 Edition and made one thing official: the product catalog is now a protocol. The Universal Commerce Protocol, co-developed with Google and backed by Amazon, Meta, Microsoft and Stripe, opens to every developer with no approval required. Its discovery layer, the Catalog API, turns products from millions of merchants into structured, queryable data that ChatGPT, Copilot, Gemini and Google AI Mode can read by default. A day earlier, Adyen launched its own Agentic Feed, a structured catalog, pricing and availability layer pushed across conversational surfaces.
They’re right. Discovery is being rebuilt around structured feeds, and the platforms moving first will own the rails. But I think they’re being too polite about where this leaves everyone whose data does not start clean.
The protocol is a port. It exposes whatever cargo you already have.
One line from Shopify’s own announcement should stop a product data team cold: AI searches powered by the Shopify Catalog convert at twice the rate of searches built on scraped data. That number is not a marketing flourish. It carries the whole thesis. When an agent decides which products to surface, it is grading the structure, completeness and machine-readability of your catalog, then ranking you against everyone else in the feed.
UCP, the Agentic Commerce Protocol that OpenAI and Stripe maintain, MCP, A2A. Four open standards in roughly two quarters, all converging on the same assumption: that the product data on the other side of the connection is ready to be read by a machine. None of them make your data ready. They assume it already is.
The platform launches skip this part. A protocol is a port. It exposes whatever cargo you already loaded onto the dock. If your supplier data arrives as a 40-tab spreadsheet, a PDF catalog with inconsistent units, and three competing names for the same attribute, UCP does not fix that. It publishes it, at agent speed, to every AI surface at once. The merchants who win the 2x are not the ones who adopted the protocol fastest. They are the ones whose catalog was clean enough to be judged well.
What “clean enough to be judged by machines” actually costs
Most teams underestimate this because the cost is buried upstream, before the product ever reaches the PIM. Across 70+ implementations at LemonMind, the pattern is consistent. Bringing 1,000 products to PIM-ready quality by hand runs about three months of work and roughly 14,000 EUR. That is not the license. That is not the integration. That is the human labor of turning what suppliers actually send into something a system, or now an agent, can trust.
The enterprise reality makes it heavier. A typical B2B catalog at our ICP runs a classification store with 500 to 600 attributes per product. Suppliers populate a fraction of them, in their own formats, in their own languages. The gap between “we have a product feed” and “we have a feed an agent will rank” is exactly that manual cleanup work, and it scales linearly with SKU count unless you change how the data enters.
The math that matters, once discovery moves to a protocol, is blunt.
| Path to an agent-readable catalog | 1,000 SKUs | 10,000 SKUs | Time to live |
|---|---|---|---|
| Manual onboarding to PIM-ready | ~14,000 EUR | ~140,000 EUR | ~3 months per 1,000 |
| AI-native onboarding (openProd) | fraction of the above | scales sub-linearly | minutes per supplier file |
In our own runs, a supplier PDF reaches PIM-ready in about 15 minutes on a live file. One test processed 4,000 products in roughly 90 seconds and recognized 128 variants out of a single PDF. The point is not the speed for its own sake. The point is that when the distribution layer becomes a real-time feed, batch cleanup that takes three months is no longer a workflow. It is a missed window.
The CFO question is no longer about software. It is about distribution.
Until this quarter, a PIM business case was an efficiency argument: faster onboarding, fewer errors, lower labor cost. That framing still holds, but UCP and Agentic Feed change the stakes. Catalog quality is now a distribution variable. If agent-driven discovery converts at twice the rate on structured data, then the difference between a clean catalog and a messy one is not a tidiness preference. It is the difference between being surfaced and being skipped in the channel that is growing fastest.
So the question a finance team should ask is not “what does the PIM cost.” It is sharper than that. What is your cost per 1,000 SKUs onboarded today, and what does that number do to your payback period when discovery runs through a protocol that grades the feed? If onboarding is the line item that scales with every new supplier and every new SKU, then it is also the line item that caps how fast you can show up in agentic channels. You can buy the most agent-ready PIM on the market and still be invisible, because the bottleneck was never the PIM. It was the upstream layer that fills it.
This is also why vendor-locked AI access misses the real cost. Pimcore, Akeneo, Sales Layer, Struct and Syndigo have all shipped MCP servers. That commoditizes access to the data already inside the PIM. It does nothing for the data that has not made it in yet, which is where the 14,000 EUR lives.
Adopt the protocol. Then fix the payload.
UCP is good news. A single open standard for discovery beats every retailer maintaining a different feed for every agent ecosystem. Adopt it. Register your catalog. But treat the protocol as the easy half of the work, because it is. The hard half is making sure what you publish through it is structured, complete and consistent enough to earn the 2x instead of forfeiting it.
The sequence is what teams get wrong. The instinct is to wire up the protocol first and assume the data will follow. Same reflex as using a forklift to deliver pizzas: powerful machinery pointed at the wrong part of the problem. The feed is only as good as the supplier data behind it, and supplier data is born messy. Fix the intake, then publish. An agent does not care how modern your protocol stack is. It cares whether your catalog can be read.
So before you celebrate being UCP-ready, answer the question your suppliers’ files are about to answer for you in public: is your catalog ready to be judged by machines, or only ready to be connected to them?
Related reading: Your Product Catalog Is Invisible to AI Agents, Your PIM Architecture Won’t Survive Agentic Commerce, MCP, ACP, A2A: Three Protocols Your PIM Must Speak, What Agentic PXM Actually Needs.
