For ecommerce & product teams

Make your catalogue usable by AI shoppers.

A feed can pass every structural check and still fail a real buying question. Herm tests whether AI shopping systems can find the right product, understand the attributes that matter, trust the information and complete the next step.

Connect a feed URL, export or API.No change to your catalogue. what a test does

◍ herm · shopping test Illustrative shopping test
Run 12 · 14 Aug 2026 · Kessock Outdoor · 2,340 SKUs · UK / EN journey 31 / 50
Customer request I’m walking the West Highland Way in October. I have wide feet, so I need something genuinely waterproof that I won’t have to break in first. I’d like to stay under £180.
Customer need · 6 requirements
genuinely waterproofwide fit · EEno breaking inunder £180UKhiking · October
Candidate products 14 products retrieved · 4 shortlisted
  • Kessock Moorland Trek Mid GTX waterproof · EE width in stock · £139.00
  • Kessock Braemar Trail Shoe Rejected · waterproofing below stated need
  • Kessock Lomond Light Boot Rejected · waterproofing below stated need
  • Kessock Torridon Insulated Boot Rejected · winter insulation not matched to autumn hiking
Agent evaluation · Kessock Moorland Trek Mid GTX evidence read from the product record
Requirements and whether the catalogue answers them
Requirement State
Correct category Answered
Correct market Answered
Budget respected Answered
Waterproof rating supported Answered
Wide-fit suitability Not stated
Break-in confirmation Unclear
Valid retailer destination Answered
Failure · requirement not satisfied Insufficient wide-fit evidence across 34 variants.

Observation The product is in stock, correctly priced and structurally valid. It fails because one of the customer's stated requirements cannot be resolved from the information the catalogue publishes.

02The problem

Feed health tells you whether the file works. It does not tell you whether the shopping journey works.

Traditional catalogue checks are designed to catch structural errors: missing fields, invalid values and schema problems. AI shopping introduces another test. Can a system use the information you provide to satisfy a customer's actual constraints?

Feed validator
  • size present
  • width present
  • price present
  • upper present
  • waterproof rating present
  • gtin present
Pass Structure is correct.
Customer need
  • will these fit a wide foot?
  • do they need breaking in?
  • genuinely waterproof, or water-resistant?
  • in stock in my size?
2 answers missing Same product record.
Technically valid is not the same as commercially answerable. Both checks are useful. Only one of them reflects the buying task.
03The four shopping tests

Test the journey the way a customer experiences it.

One request, one product, four stages. A stage only runs on what the previous stage produced.

Tracked journey · Kessock Moorland Trek Mid GTX stages run in order, the journey stops where it breaks
  1. 01 Pass

    Find

    Can the agent retrieve the right eligible product?

    What happened here Four candidates shortlisted from the connected catalogue, the target product ranked first on category and price.
  2. 02 Fail

    Answer

    Does the available product information resolve the shopper's important questions?

    What happened here Wide-fit suitability unresolved and break-in behaviour unconfirmed, from a record that carries neither.
  3. 03 Partial

    Trust

    Can the agent rely on the identity, claims and commercial facts it is using?

    What happened here The waterproof claim is supported. The fit claim is asserted without a published specification behind it.
  4. 04 Pass

    Close

    Can the shopper reach an appropriate place to buy?

    What happened here Price, stock and UK destination all resolved. Never reached: the journey stopped at Answer.
Journey outcome The product was found and the commercial step was available. It failed at Answer, so Trust was evaluated on incomplete evidence and the journey never reached a confident recommendation.
04Show the test

See exactly where the product journey breaks.

Every journey is recorded: the request, the requirements read out of it, what the system looked for, what it found in your product record and the point at which it could go no further.

Illustrative shopping test Fig. 01 · journey detail
Product Readiness · Journey detail
journey 31 / 50 · herm test profile · claude
Test environment
environment
Controlled agent
catalogue
Connected feed · read only
journeys
50 · UK / EN
checkout
Not attempted
Candidate products
  • Kessock Moorland Trek Mid GTX waterproof · EE width in stock · £139.00
  • Kessock Braemar Trail Shoe Rejected · waterproofing below stated need
  • Kessock Lomond Light Boot Rejected · waterproofing below stated need
  • Kessock Torridon Insulated Boot Rejected · winter insulation not matched to autumn hiking

Read only. Retrieval reads the connected catalogue and nothing is written back.

Customer request herm test profile · claude
I’m walking the West Highland Way in October. I have wide feet, so I need something genuinely waterproof that I won’t have to break in first. I’d like to stay under £180.
Parsed requirements genuinely waterproofwide fit · EEno breaking inunder £180UKhiking · October
Requirement, and what the record could answer
  • Correct category Complete
  • Correct market Complete
  • Budget respected Complete
  • Waterproof rating supported Complete
  • Wide-fit suitability Not found
  • Break-in confirmation Ambiguous
  • Valid retailer destination Complete
Requirements met 5 of 7 Category, market, budget and waterproofing.
Stage reached Answer Find passed. Close was never evaluated.
Products sharing this gap 214 Same attribute template.
Recommendation
Kessock Moorland Trek Mid GTX £139.00 UK 9 · EE width in stock · UK kessock.com/p/moorland-trek-mid-gtx-wide
Unanswered requirement

Insufficient wide-fit evidence across 34 variants.

journey result: failed
19 of 50 journeys failed this run · 2,340 SKUs · read only
05What your team gets

Turn shopping failures into a product-data backlog.

Every row traces back to the customer requirement that failed and the products it affects.

Product-data backlog · run 12 exports as CSV or JSON
Product-data issues found in this run
Issue Affected SKUs Customer impact Priority Owner
Missing attributes Important specifications absent from the catalogue. 214 Wide-fit suitability cannot be answered across the walking-boot range. High Product data
Ambiguous claims Information exists but is too unclear to answer the customer's constraint confidently. 88 "Waterproof" does not resolve to a declared membrane or rating. High Merchandising
Conflicting data Different product or data surfaces disagree. 21 Feed weight and product-page weight differ, so neither can be trusted. Medium Ecommerce ops
Weak evidence A claim is present but cannot be reliably supported. 46 The break-in claim is asserted with no published test behind it. Medium Product data
Channel blockers The product is understood but the next commercial step cannot be completed correctly. 12 Destination geography is missing, so the request cannot be completed for those markets. Low Marketplace
Each issue opens to the failed journeys behind it: the request, the requirement, the evidence read and the products affected. 5 issue types · 381 SKUs affected · 50 journeys tested
06From diagnosis to improvement

Fix the information that matters to the buying decision.

The goal is not to add more product copy indiscriminately. It is to resolve the information gaps that stop a system from answering a real customer need.

  1. 01 Finding 19 failed journeys, one root issue
  2. 02 Affected products 214 products, 34 variants first
  3. 03 Proposed data field-level changes, each with evidence
  4. 04 Review queued for brand review
  5. 05 Approve brand decides what is accurate
  6. 06 Download / export no write-back to your PIM
  7. 07 Source updated, re-test you publish, Herm re-runs the same journeys
Product record · wide-fit suitability Illustrative shopping test
Before · ambiguous Fit: standard. no width range, no last description, requirement unresolved
After · answerable Wide and extra-wide fittings. Lasted for high-volume feet; EE width in UK 6 to 13. requirement satisfied on re-test, the same claim stated precisely evidence: approved claim register v4 · fit test summary 2025

Herm proposes the information that would close a requirement, with the evidence behind it. Your team reviews it, publishes through your own systems, and the affected journeys are re-tested.

07How it fits your stack

Work with the product information you already maintain.

Read in, evaluate, hand findings back. Your systems stay the record of truth.

Your systems

Existing commerce systems

Your PIM, product platform, catalogue and channel feeds continue to own the product record.

Catalogue input · supported Product-feed URLXMLCSVJSONGoogle Merchant CenterAPI
Herm · Product Readiness

Shopping tests and findings

Customer requirements are tested against the catalogue you connected. Findings name the requirement that failed, the evidence read and the products affected.

  1. 01 Customer requirements defined per category
  2. 02 Find, Answer, Trust and Close tested per product
  3. 03 Failures traced to the missing information
  4. 04 Findings grouped into a product-data backlog
Back to your systems

Product-data remediation

Approved corrections and enrichment are exported for your team to publish. Once the source data changes, the affected journeys are re-tested.

Export · review required CSVJSONStructured enrichment package
Herm evaluates the catalogue.

It does not replace your PIM or commerce platform, and it does not write into your feed, product platform or Merchant Center account. The test reads the product information you already publish and reports what an AI shopping system could and could not do with it.

08Why this matters now

As shopping interfaces change, product information becomes part of the selling experience.

When a buyer asks an AI system for a product that meets five specific constraints, the product that can be evaluated confidently has an advantage over one whose catalogue leaves the important questions unanswered.

The request
Buying questions arrive as several constraints at once, not as a keyword.
The test
A system either resolves those constraints from your product information, or it does not.
The consequence
Unanswered constraints read as uncertainty, and uncertainty is not a recommendation.
09Where this fits in Herm

Product Readiness is one part of AI Commerce Readiness.

Product Readiness focuses on whether your catalogue can support the buying task. Start here if product data, merchandising or digital commerce is your responsibility.

  • Be Visible Brand Visibility Live Does AI know and recommend you?
  • Be Sellable Product Readiness Live Can AI actually complete the recommendation: specs, price, stock, where to buy?
  • Be Relevant Customer Intelligence Live Are your own experiences right for the shopper in front of them?
  • Be Preferred Offers Live Do the shoppers who buy your category buy you, and keep buying you?

Find the customer questions your catalogue cannot answer.

Test real buying requirements against your product information and see exactly where the journey fails.

starts with
A feed URL, catalogue export or API connection
tests
Real customer requirements across Find, Answer, Trust and Close
you receive
The failed requirements, the evidence read and the products affected
boundary
Read only: nothing is written back to your catalogue