Give AI the product information it’s missing.
Herm turns recurring Shopping Test and Agent Answerability failures into evidence-backed product-data improvements. Review the proposed changes, update your existing commerce systems and re-run the same journeys to see whether the problem was resolved.
Product Enrichment improves the structured product and commerce information AI systems read when selecting and recommending products: read the methodology.
“Find me a waterproof hiking jacket under £200 that works for high-output walking.”
Agent cannot determine which products are intended for hiking versus casual use.
High-output hill and mountain walking in sustained wet weather.
Ready for brand review, not published.
The same missing fact can break hundreds of shopper journeys.
Product Readiness tells you what failed. Product Enrichment is where the underlying product information gets better, on the brand’s approval and through the brand’s own publishing route.
Move from failed tests to the catalogue issue behind them.
A list of failed tests is a symptom list. Herm groups them into the product-information issue they share, and scopes it: how many products carry it, and how much of the tested suite it affects.
Add authoritative fit, waterproofing and field-test information.
What Product Enrichment improves.
Product Enrichment works on structured product and commerce information: the information an AI system reads when it selects and recommends a product.
The structured facts an agent filters and ranks on.
What the product is for, and under what conditions.
What actually separates one variant from the next.
Statements the brand has already signed off.
What the product works with, and what it does not.
Market, availability and destination facts.
Structured product and commerce information used for product selection and recommendation.
Long-form editorial and web content.
The two can complement one another. They are separate engines, and a Product Readiness failure may need one, the other or both.
See exactly what would change on the record.
Every proposal is a field-level diff against the product’s current information. Nothing is hidden inside a summary, and nothing changes until your team approves it.
Customer-facing information labels are shown. Internal field names appear only where a brand asks for them.
Don’t fill gaps by guessing.
Each proposed value resolves to approved evidence. Where nothing authoritative supports a value, Herm does not propose it.
Herm should not convert absence of information into invented product facts. An unsupported value stays unproposed and remains visible as an information gap for the brand to close.
| Proposed value | Evidence | Status |
|---|---|---|
| Multi-day hill and mountain walking on mixed terrain | approved product specification · wet-weather guidance sheet v4 | Proposed |
| Wide fitting, last width 105 mm | approved claim register v4 · fit test summary 2025 | Proposed |
| Suitable for technical winter ground | nothing authoritative addresses winter or technical use | Not proposed |
| Recommended for trail running | specification supports it · Brand Intent restricts it | Held · Brand Intent |
How evidence is selected and prioritised between sources is not published.
Product enrichment should come from information the brand can defend.
You decide what counts as authoritative. Herm proposes from the evidence you supply, and names the source behind every value.
The catalogue you already publish to channels.
The technical record your product team owns.
Guidance sheets, manuals and care information.
Statements legal and marketing have signed off.
What is true and approved per market.
Only sources supported by the product are shown here.
Better product data still has to respect Brand Intent.
Evidence can support a use case the brand does not want the product positioned for. Herm does not propose enrichment that violates approved commercial positioning.
The proposal is still recorded, with the reason it was not recommended, so the commercial decision stays visible rather than disappearing.
Approved specification establishes cushioning and outsole suitable for road running.
Do not position this product for running. Approved positioning is hill and mountain walking.
Fix the family, not just the SKU.
Most information gaps are not SKU-specific. Herm identifies whether a proposed value applies to one SKU, a variant, a family or a category, so the work matches catalogue scale.
Where a variant genuinely differs, in insulation, generation or market, the proposal differs with it.
| Record | Applies at | Proposed value |
|---|---|---|
| Trail Shell Jacket | Family | High-output hill and mountain walking in sustained wet weather. |
| Trail Shell Jacket · Insulated | Variant | As family, plus cold-weather use down to freezing. |
| Trail Shell Jacket · 2025 | Generation | As family. Superseded by the 2026 generation. |
| Trail Shell Jacket · US | Market | As family. US sizing and destination. |
| Trail Shell Pro | Family | Technical hill walking and scrambling. |
| Glen Lightweight Shell | Family | Low-bulk cover for day walks and travel. |
Scope is proposed and reviewable. Inheritance logic is not exposed.
One diagnosis. Hundreds of products.
Review the whole affected set in one place. Approve or reject each proposal, edit the wording, or filter down to the exceptions that need a person to look at them.
| Product | Information | Proposed value | Evidence | Review |
|---|---|---|---|---|
| Kessock Trail Shell Jacket outerwear · 3 variants | Intended use | High-output hill and mountain walking | spec · guidance v4 | Approve |
| Kessock Trail Shell Pro outerwear · 2 variants | Intended use | Technical hill walking and scrambling | spec · claim register | Approve |
| Kessock Glen Lightweight Shell outerwear · 4 variants | Intended use | Day walks and travel, not sustained exposure | spec · description | Approve |
| Kessock Cairn Summit Mid GTX footwear · 1 variant | Waterproofing | withheld | sources disagree | Exception |
| Kessock Moorland Trek Mid GTX footwear · standard fitting | Fit width | Standard. Wide available as its own variant. | claim register v4 | Approve |
Review is per product. Exceptions, meaning conflicts, Brand Intent restrictions and unsupported values, are held back from bulk approval.
Herm proposes. Your team decides.
Nothing reaches your commerce systems without a person approving it. Approved changes are exported, and your team publishes them through the systems you already own.
Take the approved changes back to your existing product stack. Herm works alongside your PIM and feed infrastructure rather than replacing it.
- Product data
- Merchandising
- Legal and compliance
- Brand
- CSV
- JSON
- Feed-ready
- Per-market split
What Herm does not do today. Herm does not write changes directly into your PIM, product platform or feed. Your system stays the record of truth, and your team stays in control of what is published. Once the source system is updated and the new data reaches Herm, the same journeys are run again.
Update the data. Ask the same shopper again.
The proof is not that a field was filled. It is that the same shopper journey, run again, reaches the right product for the right reason.
Product data enriched: approved, exported, published by the brand, then re-tested.
Kessock Outdoor sample workspace. A resolved test demonstrates improved AI shopping performance on the tested journey. It is not a claim about real-world sales.
When product information changes, the affected journeys run again.
You do not have to remember which tests were touched by a change. Herm can automatically re-run the affected Shopping Tests and show whether the previous failure was resolved.
How affected tests are selected is not published.
Not every failed journey is a product-data problem.
Enrichment is applied, the test is re-run, and sometimes it still fails. That result is kept and reported rather than smoothed over.
A still-failing test after enrichment narrows the diagnosis. The information gap is closed, so the cause is somewhere else.
Answerability tells you what’s missing. Enrichment helps supply it.
An unresolved shopper question is the measurement. The proposed information is the action. The question is then asked again.
- Answerability Unknown. No authoritative width information.
- Affected 214 products, 38% of the tested suite.
- Proposed Fit width at variant level, with last width in mm.
- Evidence Approved claim register v4 · fit test summary 2025.
- Re-asked Supported. The question now resolves.
Structured product information, long-form content, and your system of record.
Three different jobs that are often confused with one another. A Product Readiness problem may require structured enrichment, long-form content, or both, and neither replaces the platform your product data lives in.
Structured product and commerce information used for product selection and recommendation.
Stores and distributes product information. Herm does not replace it.
Different infrastructure. Product Enrichment and Content Suite are separate engines that can complement one another. Your PIM or feed platform stores and distributes product information; Herm tests how AI agents use it, identifies gaps, proposes evidence-backed improvements and re-tests the result.
One issue, from detection to resolved test.
Each enrichment workstream carries its own history: when the failure was detected, what was proposed, when the brand approved it, when the source data changed and what the re-run produced.
- 10 Aug Failure detected 19 tests · intended use missing or ambiguous
- 12 Aug Enrichment proposed 214 products · evidence attached
- 18 Aug Brand approved 186 approved · 21 edited · 7 exceptions held
- 22 Aug Source updated published by the brand, re-connected
- 22 Aug Re-tested 16 of 19 previously failing tests resolved ↑ 16
Sellable Intent Rate improved on the affected test set. No claim is made about overall sales.
Kessock Outdoor sample workspace values on one affected test scope. Scoring formulas, weights and thresholds are not published.
Common questions
Does Herm write changes into our PIM?
No. Herm proposes, your team approves, and approved changes are exported for you to publish through the systems you already own. Your system stays the record of truth.
What if the evidence does not support a value?
Then Herm does not propose it. An unsupported value stays unproposed and remains visible as an information gap, so it can be closed with evidence you already hold or deliberately not claimed at all.
Can enrichment contradict our commercial positioning?
No. Where evidence supports a use case that Brand Intent restricts, Herm does not recommend the value for publication. The proposal is still recorded with the reason, so the commercial decision stays visible.
Do we have to review hundreds of products one at a time?
No. The affected set is reviewed in one place, with filtering and bulk approval. Exceptions, meaning conflicts, Brand Intent restrictions and unsupported values, are held back from bulk approval so a person looks at them.
What happens if the test still fails after enrichment?
That result is kept and reported. It is useful: the information gap is closed, so the cause is somewhere else, and the diagnosis narrows.
Is this the same as Content Suite?
No. Product Enrichment works on structured product and commerce data; Content Suite works on long-form editorial and web content. They are separate engines that can complement one another, not a shared generation engine.
Fix the product data behind the failed journey.
Start with one category. See the recurring information gaps, the evidence-backed changes Herm proposes, and what happens when the same shopper asks again.
Sample figures throughout.