See what happens when AI shops your catalogue.
Connect a sample of your existing product catalogue. Herm runs realistic shopping journeys, evaluates whether the agent finds and recommends the right products, and shows the information gaps that get in the way.
The test runs on a sample of your catalogue and does not change your feed, Merchant Center account or PIM.
A feed-health check tells you whether the data is valid. This test tells you whether AI can use it.
Validation and readiness ask different questions of the same catalogue. One is about structure. The other is about whether a shopper’s need can be satisfied from what the record contains.
The gap. A catalogue can pass the first and fail the second.
- Are required attributes present?
- Are the formats valid?
- Are identifiers resolvable?
- Will the channel accept the file?
Answers whether the channel will accept the file.
- Can the agent find the right product?
- Can it answer the decisive question?
- Is the claim it made supported?
- Is there a valid place to buy?
Answers whether a shopper’s need can be met from the record.
Start with the product information you already have.
No new data model, no migration, no preparation project. Connect the feed or file you already publish to channels and the test runs on a sample of it.
Optional, and it improves the test. Herm can start outside-in, and your customer insight makes the test sharper. Don’t have that ready? Start anyway: Herm can build an outside-in test from the catalogue and category context.
The product feed you already publish.
Standard product XML export.
A spreadsheet export of the catalogue.
Structured product records.
Read the product data already in your account.
Connect directly from your own systems.
Catalogue in, shopping journeys run, evidence out.
Four steps. Nothing to configure before the first run.
A real customer need, attempted by an agent, evaluated against your product information.
The agent searches, compares and recommends from the catalogue you connected. Herm then checks the recommendation against what the customer actually asked for. Select any evaluation line to see the product evidence it was read from.
“I need a waterproof trail shoe under £150 for wide feet.”
Trail Pro
Authoritative product information does not establish fit suitability.
The requested product category is stated explicitly in the connected catalogue.
You don’t just get a score. You get the failure.
A readiness snapshot for the sampled catalogue, and underneath it the journeys that failed, the questions that went unanswered and the product information behind both.
Sellable Intent Rate for the sampled catalogue, with Find, Answer, Trust and Close.
The shopping tests that did not complete, and why each one broke.
The buying questions your product information could not resolve.
The product records behind every pass and every fail.
- 5 failed shopping journeys
- One shared information gap
- Fit suitability not published at variant level
- 38 products affected
Product Readiness expands this into full Agent Answerability across your catalogue.
And what if the agent’s answer is factually possible, but wrong for your brand?
If the test finds a problem, Product Readiness keeps going.
The test is the first run of the same loop the full product operates continuously.
Find the customer questions your product information leaves unanswered.
Explore →Run the same shopping journeys again and see whether the result moved.
Where the test ends and Product Readiness begins.
The test runs on a sample and reports selected findings. The full product runs continuously across the catalogue.
- A sample of your catalogue
- A limited set of shopper tests
- A readiness snapshot
- Selected findings with their evidence
- Full catalogue
- Scheduled Shopping Tests
- Complete Answerability
- Brand Intent
- Product Enrichment
- Multiple markets and environments
- History
- Alerts
- Re-testing
| Limit | Value |
|---|---|
| SKU sample size | unset |
| Number of tests | unset |
| Markets included | unset |
| Test environments | unset |
| Report availability | unset |
| Card / account required | unset |
The test uses the same evaluation as the full product, on less of the catalogue.
Your catalogue is commercial data. Treat it that way.
Product feeds carry pricing, margins, ranging and launch information. The test is built to read what it needs and change nothing.
Data modification boundary. The test does not modify your product feed, Merchant Center account or PIM. Product Enrichment may later propose improvements. You review them, download them and update your own source system.
- The test does not modify your product feed, Merchant Center account or PIM.
- It runs on a sample of the product information you connect, and nothing else.
- Proposed product-data changes are exported for you to publish. Herm does not write them back.
Retention, deletion, model-training use, access and security posture are answered in the policy framework, and by the security team for anything it does not cover.
Questions about retention, deletion, model training and access are answered in the policy framework rather than summarised here. Anything not covered there, the security team will answer directly.
Not another feed validator.
It is a controlled test of whether an AI shopping agent can use the supplied product information to satisfy representative customer needs correctly.
Simulation environment. Herm’s primary shopping simulation environment uses and extends Anthropic’s open-source Commerce Agents reference implementation, providing a real commerce-agent architecture for testing product discovery, comparison and recommendation. Herm independently provides the Product Readiness evaluation, shopper-intent testing, Brand Intent and diagnosis.
Textual attribution only. No partnership, endorsement or certification is implied, and the test is not the live consumer shopping experience of any model provider.
Before you connect anything
What is Sellable Intent Rate?
The share of tested shopper intents an AI shopping agent could complete correctly from the product information supplied.
Does the test change anything in our systems?
No. The test does not modify your product feed, Merchant Center account or PIM. It reads a sample of the catalogue and evaluates what an agent can do with it.
Do we need to prepare our data first?
No. Connect the feed or file you already publish to channels. There is no new data model, no migration and no preparation project, and nothing to configure before the first run.
Is this the same measurement as full Product Readiness?
The same evaluation, on less of the catalogue. The test runs a limited intent set against a sample and reports selected findings; the full product runs continuously across the catalogue with Brand Intent, enrichment, re-testing and monitoring.
Do we need customer research to start?
No. Herm can build an outside-in test from the catalogue and category context alone. Adding your own customer insight makes the test sharper, and it is optional.
Your feed is ready. Find out if AI is.
Connect your catalogue and start the AI Commerce Readiness Test.
Sample figures throughout.