Your catalogue has answers. Does it have the ones shoppers need?
Agent Answerability tests the questions behind real buying decisions against the authoritative information in your catalogue. See what AI can answer confidently, where evidence is weak and which information gaps affect entire product families.
Agent Answerability measures how much of the customer’s buying decision an AI system can resolve using authoritative product information: read the methodology.
Kessock Moorland Trek Mid GTX · “Will these keep me dry through several hours of heavy rain?”
SupportedAn approved claim and a specification both establish the statement. The question resolves.
The field can be filled and the question can still be unanswered.
This is not a claim that your product feed is bad. Product attributes and customer decisions simply operate at different semantic levels. An attribute records a property. A shopper asks whether that property holds under the conditions they care about.
“Will this keep me dry for several hours of heavy rain?”
“Will this work for wide feet?”
“Is this comfortable for an all-day hike?”
Measure the decisions that happen before purchase.
A buying decision is made of several different kinds of question, and a catalogue rarely fails at all of them equally. These are the question types Answerability evaluates today.
| Question type | Representative shopper question | What has to be resolvable |
|---|---|---|
| Suitability | “Is this appropriate for hill walking rather than casual wear?” | Whether the product record establishes what the product is intended for. |
| Constraints | “Will these work in wet conditions and still fit in hand luggage?” | Whether stated conditions and limits can be checked against the record. |
| Comparison | “How is this different from the Cairn Summit?” | Whether two products or variants can be distinguished on authoritative information. |
| Compatibility | “Will these take the crampons I already own?” | Whether compatibility with another product or standard is established. |
| Usage | “Is this appropriate for a beginner or for experienced walkers?” | Whether the record addresses who the product is for. |
| Performance | “Will these hold up over three consecutive days on wet ground?” | Whether performance under a specific condition can be supported. |
| Claims | “Are these actually waterproof, or water-resistant?” | Whether a specific statement is supported by approved evidence. |
| Commercial | “Can I get these in the UK for under £150?” | Whether availability, market and commercial terms are resolvable. |
Herm builds an outside-in view of the questions. You make it yours.
Herm assembles the likely buying questions for a category independently of the brand, so the first measurement does not depend on internal assumptions.
The view gets sharper when the brand adds what it already knows about who buys and what they hesitate over. The question universe is reviewable and editable before a run.
Category and buying-question intelligence built independently of the brand.
What the brand already knows about its customers and their objections.
These are representative buying questions. Generation method and underlying intent datasets are proprietary.
See which parts of the buying decision your catalogue covers.
Answerability rolls up by decision area, which is the level most product teams can act on. The question it answers: which parts of the decision systematically lack authoritative information?
Reading this view. Fit and weather protection are well covered because the category has always described them. Intended use and product comparison are weak because no feed specification ever asked for them.
Find the products with the largest unresolved information gaps.
A metric only becomes useful when it lands on a work queue. Every product carries its own Answerability, the questions it cannot resolve and the shopper intents that depend on them.
| Product | Answer. | Unanswered questions | Evidence flags | Affected intents |
|---|---|---|---|---|
| Kessock Cairn Summit Mid GTX technical boot · 1 variant | 43 | 12 unanswered questions, including intended use and multi-day comfort | 1 conflict · 3 weak | 31 of 50 |
| Kessock Moorland Trek Mid GTX walking boot · standard fitting | 76 | 3 important unanswered questions | 1 weak | 12 of 50 |
| Kessock Trailhead Low trail shoe · standard fitting | 68 | 5 unanswered questions, mostly comparison | 2 weak | 9 of 50 |
| Kessock Moorland Trek Mid GTX Wide walking boot · wide fitting | 92 | 1 weak area | evidence complete | 2 of 50 |
Kessock Outdoor sample workspace. Answerability values are illustrative and are not published as customer proof.
Open the question. Inspect the answer.
Every state resolves to the records it was read from. Select a question to see the evidence Herm found and the outcome it produced.
An approved claim and a membrane specification both establish the statement.
Nothing required. This question is resolvable today.
Herm shows evidence and outcome. Evaluator prompts are not exposed, and Herm does not read hidden model reasoning.
“We don’t know” is different from “no”.
If authoritative product information does not establish something, Herm does not convert the silence into a negative product claim. Missing evidence is reported as missing.
The distinction matters commercially. An unknown is a product-data task with an owner. A false negative is a lost sale on a product that was suitable all along.
“Are these suitable for wide feet?”
No approved width or fit-suitability information on the product record.
The question is open. It becomes a product-data task, and it can be closed with evidence.
A wide fitting may well exist. Inventing a negative claim would misrepresent the product.
Find when your own product information disagrees with itself.
Product information is written in several places by several teams. When those places disagree, an agent will still answer, it will just pick one of them.
Herm surfaces the disagreement rather than resolving it silently. Which record is right is a brand decision.
Answerability for this question is withheld until the brand confirms which record is authoritative. Until then the agent's answer depends on which source it happens to read.
Unanswered questions become failed shopping journeys.
Answerability is measured on the product record, before any agent runs. Shopping Tests then show what that gap costs when a real customer need meets it.
One unresolved question rarely fails one test. It fails every buying situation that depended on it.
- Question Are these suitable for wide feet?
- Answerability Unknown. No authoritative width information.
- Shopper intent Waterproof trail shoe for wide feet under £150
- Agent Cannot confidently identify the best match
- Shopping Test Failed. More appropriate eligible product not considered.
One unanswered question can affect hundreds of products.
Herm groups unresolved questions into families, so a catalogue-wide information gap reads as one piece of work rather than 214 separate defects.
Intended use, stated per product, with the terrain and duration it is appropriate for.
The same missing information explains every product in the group, so it is written once and applied across the family.
Reporting says “intended use”, not a database field name. Internal field names appear only where a brand asks for them.
The number of affected products and intents is known before the work starts, so it can be prioritised against everything else.
Customer-facing information labels are used in reporting. Internal field names appear only where a brand asks for them.
Turn missing answers into structured product information.
An unresolved question is a work item with a known scope. Product Enrichment proposes the information that would close it, with the evidence behind the proposal, for the brand to approve.
Multi-day hill and mountain walking on mixed terrain. Cushioned for consecutive long days. Not intended for technical winter ground.
What Herm does not do. Herm does not write into your PIM, product platform or feed. Approved enrichment is exported for your team to publish through the systems you already own. Once the source system is updated and re-connected, the same questions are asked again.
Answerable is not the same as acceptable.
Agent Answerability asks whether the question can be resolved from authoritative product information. Brand Intent asks whether that is how the brand wants the product represented.
A product can be technically suitable for a use case the brand does not want it recommended for. Both answers are needed to judge a recommendation.
The record establishes cushioning, weight and outsole suitable for hard surfaces. The question resolves.
Cairn Summit is positioned for hill and mountain walking. The brand does not want it recommended for road running.
Answerable, and still not the recommendation the brand wants made.
Answerability should come from evidence the brand can stand behind.
You decide what counts as authoritative product information. Herm reads what you establish, and reports a question as open when nothing authoritative addresses it.
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 already signed off.
What is true and approved in each market you sell in.
Anything else you establish as the record of truth.
How evidence is prioritised between sources is not published.
A product’s Answerability changes when its information changes.
Answerability is not a certificate. A new product launches with little to read. Enrichment closes questions. A product generation changes and yesterday’s evidence goes stale. Product Readiness asks the same questions again.
- Apr Product launched record carries specification and price only 58
- Jun Enrichment approved intended use and fit width published 76
- Aug Claims register extended multi-day guidance added to the family 89
Sample values from a demonstration workspace. Re-measurement is automatic after connected product information changes.
Completeness measures the record. Answerability measures the decision.
Feed management is doing its job. A validated feed is what makes a catalogue readable in the first place. Answerability is the layer above it, and it needs the layer below to be sound.
You need both.
- Does the field exist?
- Is it in the correct format?
- Is every required attribute present?
- Will the channel accept it?
A record that a channel will accept.
- Can the shopper question be resolved?
- Is the evidence sufficient?
- Can variants be distinguished from one another?
- Can an agent explain why this product suits the need?
- Are the important uncertainties visible?
A record a customer's decision can be made from.
Product Readiness works on structured commerce data. Authoritative long-form content published by the brand strengthens the wider information environment around a product, which is Content Suite’s territory: a separate infrastructure, not a shared generation engine.
Common questions
How is this different from feed completeness?
Completeness asks whether a field exists, is correctly formatted and will be accepted by a channel. Answerability asks whether the values in those fields resolve the questions a shopper actually decides on. A record can be 100% complete and still leave the decisive question open.
Where do the questions come from?
Herm assembles the likely buying questions for a category independently of the brand, so a first measurement needs nothing from you. Brands can then add what they know about their customers and objections. The question universe is reviewable and editable before a run.
Why report “unknown” instead of “no”?
Because they are different statements. “Unknown” says the catalogue does not establish the answer, which is a data task with an owner. “No” is a claim about the product, and inventing it would misrepresent products that were suitable all along.
What happens when our own sources disagree?
Herm surfaces the disagreement rather than resolving it silently, and withholds Answerability for that question until the brand confirms which record is authoritative. Until then, an agent’s answer depends on which source it happens to read.
Does Herm publish how questions and evidence are ranked?
No. The question types, the states and the evidence behind each result are public and inspectable. Question selection, evidence ranking, evaluator prompts and aggregation logic are not.
Find the questions your catalogue cannot answer yet.
Start with one category. See which parts of the buying decision your product information already resolves, and which it leaves open.
Sample figures throughout.