The legal precedent is here. The pass rate is 0.2%. And nobody is talking about distributors.
You've seen this pattern
A buyer at a top-20 industrial account uses an AI shopping agent to place an order. The agent sees "100" on the product page and submits. The buyer expected 100 individual units. The catalog had the SKU set to ship by the case. Unit of measure field: "CS-10." Nothing on the page disambiguated it.
The agent parsed what it could read. Assumed each. Fired the order.
The warehouse shipped 1,000 units.
The customer disputed. The distributor ate reverse logistics, a restocking fee, and two weeks of relationship repair with a top-20 account.
The VP of Ecommerce went looking for the root cause. It wasn't the agent. It wasn't the customer. It was one orphan UoM field on one SKU that hadn't been touched since 2019.
Multiply that across 1.2 million active SKUs and you have a risk profile nobody at the distributor is currently measuring.
What just changed
Two things happened in the last two weeks (as of April 23, 2026).
The National Law Review published a piece warning that third-party AI shopping agents are transacting on retailers' sites at material volume. Retailers don't control these agents. One major ecommerce site has already obtained a preliminary injunction blocking agents from transacting on its platform. The court's reasoning: even when a user authorizes an agent, the site has not. That makes the agent's access unauthorized.
First legal precedent in agentic commerce. It won't be the last.
Separately, a developer analysis scanned roughly 4,000 stores with valid Universal Commerce Protocol manifests — the emerging standard that tells AI agents "you can transact with this merchant."
Nine of those 4,000 delivered a flawless end-to-end agent shopping experience.
Nine. Out of 4,000. A 0.2% pass rate.
The blind spot
Read every piece of coverage on agentic commerce this month. Law firms. Analysts. Protocol working groups. Industry press.
Almost none of it is about B2B distribution.
The entire conversation is scoped to consumer retail — shopping assistants buying sneakers, pet food, and birthday gifts. That's where the early data comes from, so that's where the narrative lives.
But distributors are going to get hit harder. And earlier. And with more expensive consequences.
Why B2B is uniquely exposed
Three reasons, in order of severity.
SKU volume
A B2C retailer runs 50,000 SKUs on a good day. A mid-market industrial distributor runs 500,000 to 2,000,000. A large MRO distributor runs 5 million. More SKUs mean more surface area for bad data, and the ratio of SKUs-touched-in-the-last-12-months to SKUs-an-agent-might-transact-against is somewhere between 1:50 and 1:500.
Failure modes compound at scale.
More legal surfaces per transaction
A B2C transaction has two legal surfaces: returns and disputes. A B2B distribution transaction has contract pricing, MAP enforcement, export controls, compliance certifications, UoM conversions, and trade restrictions.
An agent transacting at list price when the customer has a negotiated contract is a billing failure at minimum — and potentially a contract breach. An agent routing a dual-use item to a restricted jurisdiction is not a returns problem. It's a federal compliance problem. An agent quoting a product with a stale UL certification could expose the distributor to product liability.
Every one of those is a line item where an agent can make a wrong call on bad catalog data.
The liability chain
In B2C, the agent operator, the payment rail, and the merchant are still arguing about who owns failure. In B2B distribution, the distributor is the transacting entity of record. The agent is a proxy. The buyer is the principal. The liability chain terminates at the distributor.
You carry the risk. You don't control the agent. You barely control your own catalog.
That's the problem.
The five failure modes
Across more than a million SKUs analyzed in B2B distribution catalogs, the same five failure modes appear in nearly every account.
Missing attributes
Agents can't infer. They parse. If "Voltage" is blank on 40% of your electrical SKUs, 40% of that inventory is invisible to the agent — and to every downstream AI interface pulling from your data.
This is the most common failure, and the most often underestimated. Merchandising teams look at their catalog and see mostly complete. Agents look at the same catalog and see mostly unparseable.
UoM inconsistency
"Each," "pack of 10," "case," "pallet," "drum," "roll." Mixed across a single taxonomy, entered by different merchandisers over five years, with no canonical conversion logic in the data layer.
An agent prices the wrong quantity. Orders the wrong quantity. Or silently fails the transaction and moves to a competitor's listing.
When it fails silently, you don't just lose the order. You lose the future orders you never knew existed.
Taxonomy drift
Three years of category editors each adding their own node. UNSPSC mapped loosely. Internal categories that made sense in 2021 and don't map cleanly to how buyers search in 2026.
The agent parses your taxonomy as authoritative. It isn't. It's an archaeology site.
Duplicate SKUs
The same product listed three times under two manufacturer part numbers and one legacy internal ID. Inventory fragmented across records. Pricing out of sync.
The agent picks one. Which one is effectively a coin flip — weighted by whatever the last crawl indexed.
Stale enrichment
Specs enriched in 2022 by a team that no longer exists. The manufacturer has since revised the spec sheet twice. The agent quotes the stale copy.
This one is especially dangerous because it looks clean. The field is populated. The description reads well. Only when the product arrives and the spec is wrong does anyone notice.
The reframe
For five years, catalog quality has been a discoverability story. Enrich your data so customers can find you on search and on marketplaces.
That story is over.
Catalog quality is now a risk management story. Third-party agents are transacting on behalf of your buyers. You don't control those agents. You barely control your own catalog. The only leverage you have — the only one — is the quality of the data those agents are parsing.
This isn't a marketing problem anymore. It's a finance, legal, and operations problem.
The encouraging part
Here's what buried the lede in the April analysis. Of the 4,000 UCP-verified stores scanned, 422 are exactly one fix away from passing.
That's not doom. That's an addressable problem.
This isn't rip-and-replace. It isn't a multi-year PIM implementation. It isn't another content cleanup sprint that decays the moment the project team disbands.
It's measurement. Then correction. Then continuous maintenance. Three steps, in that order, run as a system rather than a project.
Score. Enrich. Benchmark. Repeat.
Projects decay. Systems compound.
What to do this quarter
If you run ecommerce, product data, or digital operations at a B2B distributor, start with a baseline score. You can't manage what you can't measure, and "we did a cleanup last year" is not a score. A proper catalog quality baseline is categorical — attribute completeness, taxonomy coherence, UoM consistency, duplicate ratio, enrichment recency.
Then prioritize by revenue-weighted risk. Not every SKU is equal. A 500-SKU subcategory doing 18% of revenue gets the first pass. A 50,000-SKU long tail doing 3% goes last.
Fix the top failure mode first. In most distributor catalogs we see, it's attribute completeness or UoM consistency. Clearing one failure mode across a high-revenue category typically moves the agent pass rate by 15 to 30 points.
Rescore on a cadence. Monthly at minimum. Catalog quality is a live condition, not a project milestone. New SKUs arrive. Old specs decay. Editors drift. Without a rescore cadence, you're looking at last year's catalog.
And make one person accountable. Not a committee. Not a shared responsibility. Catalog quality owned by everyone is owned by no one.
The assessment
We're running free Category Assessments for B2B distributors through the end of the quarter. Full catalog scoring across the five failure modes, benchmarked against your category, delivered as a 3-page executive report plus a 15-page technical breakdown.
No pitch. No deck. Just the data.
You'll see where your catalog stands today, which failure modes are costing you the most, and how your scores compare to other distributors in your category.
Request your Category Assessment →
Projects decay. Systems compound. And agents don't wait.
Sources referenced