Before You Integrate The Next Agent Protocol, Audit These Five Layers Of Your Own Catalog

Michael Quoc holds 9 US patents and is the founder and CEO of Product.ai—a company building the truth layer for commerce.

getty

​Every CTO running an e-commerce stack is being asked the same question: Are we ready for agentic commerce? Most teams answer by counting integrations: UCP, ACP, MCP, Verifiable Intent. The integration list grows quarter after quarter. The layer underneath those integrations rarely gets audited.

Integrations move purchase intent across the wire. Whether your catalog can be read, cited or trusted by the agent on the other side is a separate question. After looking at 17 years of merchant signals from more than 515,000 retailers, we've discovered that the catalogs that survive agent discovery share five auditable properties.

1. The Identity Layer

Every product in your catalog has an ID. Whether that ID survives the next catalog migration, vendor switch or PIM rebuild is the first auditable question. Across the merchant catalogs we have observed since 2009, a meaningful share of 2020 SKUs no longer carry their original identifier in 2026.

The cause is rarely strategic. Catalog operations drift. An agent that cited a product in 2024 and cannot resolve the same ID today will treat your catalog as unreliable. The audit signal is your SKU survival rate across the last three catalog migrations. If your team cannot produce that number this week, the identity layer is not yet auditable.

2. The Claims Layer

Most product copy reads as marketing assertions. To an AI agent, marketing assertions and testable claims are different inputs. "Our laptop is fast" is a marketing assertion. "The CPU benchmarks at a specific score on a named test" is testable claim.

Independent testing has consistently shown that category-leading product specs hold up to verification at lower rates than buyers assume. A JAMA Network Open study of 57 sports supplements found just 11% had ingredient content within 10% of label claims. The audit signal here is the share of your product claims that an independent test would confirm. Every claim an agent might cite belongs in that count.

3. The Verification Layer

Once your catalog distinguishes marketing assertion from testable claim, the question is whether each claim carries a verification trail. Where did the data come from? When was it last validated? What confidence does the system carry in it? A claim with no provenance is a claim that an agent has no way to defend. The audit signal here is the share of your high-value product claims that resolve to a dated source the agent can inspect. Start with high-value SKUs and expand from there.

4. The Pricing Layer

Pricing has always been the layer where claim and reality drift fastest. A price is correct at the moment it is published and stale by the time the customer arrives. The same dynamic applies to discount code applicability.

In the dataset of millions of promotion codes flowing through commerce checkouts every day, the failure rate at the moment of redemption sits at 26.2%. Over 1 in 4 codes published as live fail at the customer's hand. An agent making a purchase recommendation against your catalog needs the truth at the moment of the call. Yesterday's batch job is already stale. The audit signal at the pricing layer is the latency between your published price and your real-time price.

5. The Transaction Layer

The final layer is the receipt that the customer (or the agent acting on their behalf) walks away with. U.S. retailers absorbed an estimated $890 billion in returns in 2024, roughly 16.9% of retail sales. A meaningful share trace back to a divergence between what the catalog promised and what the customer received. An agent watching this divergence will quickly learn which catalogs to cite and which to route around. The audit signal at the transaction layer is the rate at which catalog-product divergence shows up as the return reason.

The Audit Is The Work

The five layers can be audited with the data your catalog already carries. Most teams have not run any of these audits because the protocol roadmap has consumed every available engineering hour.

The roadmap is real work. It is also incomplete work. Whichever share of your catalog fails these audits is the share of your brand that agentic commerce will quietly route around.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?