Define what correct means for your brand.
Product information tells an AI agent what a product is. Brand Intent tells Herm how that product should, and should not, be represented, recommended and sold across customers, markets, retailers and use cases.
Herm evaluates shopping-agent behaviour against your Brand Intent. It does not publish these rules into external AI agents: read the methodology.
Do not recommend Velora Bright Boost C Intense for a first vitamin C.
Formulated and positioned for skin already tolerant of a lower strength.
A recommendation can be accurate and still be wrong.
Every fact in a recommendation can check out against your feed and the recommendation can still be commercially unacceptable: the advanced retinoid offered to a first-time user, the discontinued generation offered over the current one, the claim nobody approved.
Product Readiness evaluates both. Product evidence asks what is supported. Brand Intent asks what is commercially acceptable.
“I want to start using vitamin C. My skin is sensitive.”
Velora Bright Boost C
“Do not recommend Velora Bright Boost C Intense for a first vitamin C. This product is positioned for skin already tolerant of a lower strength.”
Facts read from the product record: attributes, specifications, availability.
Rules your brand defines about positioning, use case, market and customer fit.
Herm evaluates shopping-agent behaviour against your Brand Intent. It does not publish these rules directly into external AI agents.
Turn merchandising knowledge into testable rules.
Your rules already exist. They are just trapped in documents and in people. Write them the way you would say them, fill in a scope-first form, import a set you already maintain, or upload the approved documents and let Herm propose rules from them.
State the rule the way you would say it to a colleague. Herm converts it to a testable form.
Set scope, market, use case, product, action and evidence requirement field by field.
Bring an existing rule set in through a spreadsheet import.
Merchandising policy, positioning deck, claims library, legal guidance or product documentation.
Human review is required before activation. Rules are never activated automatically.
One rule can apply exactly where it matters.
A rule can be written once at brand level, or scoped down to a single variant. Context filters narrow it further, without a JSON editor in sight.
Customer segment is one of those filters, which is why the right recommendation can differ for a beginner and an experienced hiker asking the same question.
Segments are used to evaluate recommendations against the customer context defined for a Shopping Test, not to personalise anything in real time.
The right recommendation can change by market.
Eligibility, retailer destinations, approved claims, availability and product representation can all differ by market. Brand Intent is scoped accordingly, so a rule that is true in the UK does not quietly become a rule in the US.
| Market | Seller | Claim | Positioning restriction |
|---|---|---|---|
| UK | velora.com | Vitamin C 10% · THD ascorbate | Positioned for daily maintenance, not post-procedure skin |
| US | velora.com/us and Sephora | Vitamin C 10% · THD ascorbate | Positioned for daily maintenance, not post-procedure skin |
| EU | velora.eu | Imperméable 20k/20k | Availability limited to DE / FR / NL |
Retailer rules resolve against the Product Readiness Close dimension, kept distinct from a factual-correctness failure.
Approved claims need approved evidence.
Six categories cover what brands actually govern: what a product is for, what may be claimed and on what evidence, which product to prefer, who it suits, where a shopper may be sent, and which policy applies. Select any category to see a worked example.
Do not recommend Velora Bright Boost C Intense for a first vitamin C. It is formulated and positioned for skin already tolerant of a lower strength.
A road-running recommendation fails Brand Intent even when every product fact in it is true.
Herm checks the evidence when the claim appears in a Shopping Test. Generic “prefer the latest generation” reasoning is not live yet.
Manage Brand Intent at catalogue scale.
Every rule carries its category, scope, market, owner, the number of Shopping Tests it applies to, and its status. Filter the library by any of them.
| Rule | Category | Scope · market | Owner | Tests | Status |
|---|---|---|---|---|---|
| Never recommend Velora Bright Boost C Intense for an already compromised barrier. BI-0142 · updated 12 Aug 2026 | Intended use | Product rule · All | Merchandising | 34 | Active |
| Velora Clarity C Gel is for oily and congested skin rather than a sensitised barrier. BI-0118 · updated 05 Aug 2026 | Positioning | Positioning rule · All | Brand / Marketing | 21 | Active |
| Do not use “hypoallergenic” without approved evidence. BI-0031 · updated 28 Jul 2026 | Evidence requirement | Claim rule · All | Legal / Compliance | 18 | Active |
| Prefer the current generation where alternatives otherwise fit equally. BI-0104 · updated 02 Aug 2026 | Preference | Merchandising rule · All | Merchandising | 27 | Active |
| Sustainability claims must use approved evidence. BI-0029 · updated 28 Jul 2026 | Evidence requirement | Claim rule · All | Legal / Compliance | 58 | Active |
| UK shoppers must not be routed to US-only sellers. BI-0012 · updated 20 Jul 2026 | Retailer restriction | Market rule · UK | Ecommerce | 12 | Active |
| Do not recommend Velora Retinal Renewal to first-time actives users. BI-0166 · updated 14 Aug 2026 | Segment suitability | Segment rule · All | Merchandising | 9 | Review required |
| Exclude Velora Bright Boost C (previous formulation) from active recommendation. BI-0077 · updated 02 Jun 2026 | Exclusion | Exclusion rule · All | Product Data | 4 | Archived |
The people who know what right means can own the relevant rules.
Different departments hold permissions over different parts of the suite, and formal approvals run through those roles. A rule moves from draft to review to active with a named person behind each step.
Conflicts are detected, not silently resolved.
Two teams can write two rules that cannot both hold. Herm surfaces the pair with their evidence, scopes, owners and markets, and marks it review required. It does not invent a precedence order and quietly apply one of them.
Do not recommend Velora Retinal Renewal to first-time actives users.
Prefer Velora Retinal Renewal in the entry-level range.
See exactly which rule the recommendation violated.
This is a Shopping Test where every product fact is supported. Lightweight, cushioned, in stock, in market, valid destination. It still fails, because the product is positioned for experienced actives users and the shopper asked for their first vitamin C.
- 01 Shopper need
- 02 Agent
- 03 Recommendation
- 04 Brand Intent evaluation
“I want to start using vitamin C. My skin is sensitive.”
Velora Bright Boost C Intense
“A higher-strength vitamin C in a fragrance-free base, in stock for UK delivery.”
Answerability asks can AI know. Brand Intent asks should AI recommend it.
Agent Answerability asks whether authoritative product information can resolve the question at all. Brand Intent asks whether the resulting recommendation is commercially acceptable.
They fail differently and they are fixed differently, so Product Readiness keeps them apart rather than averaging them into one score.
Brand strategy changes. The rules should show it.
Every rule tracks who changed it, what changed, when, and why. A positioning decision taken in a Q3 review is still legible in the rule six months later.
| Actor · timestamp | Change | Rationale |
|---|---|---|
| M. Okafor · Merchandising 14 Aug 2026 | BI-0198 rationale updated to “Formulated and positioned for skin already tolerant of a lower strength” (was “Not a beginner serum”) | Clarified after the Q3 positioning review |
| L. Byrne · Brand / Marketing 05 Aug 2026 | BI-0118 status changed from Draft to Active | Approved by the Merchandising lead |
| System 28 Jul 2026 | BI-0029 evidence requirement added: approved supplier documentation | Legal and Compliance policy update |
| S. Adeyemi · Legal / Compliance 20 Jul 2026 | BI-0012 market scope narrowed from All to UK | US retailer agreement changed |
Brand Intent only matters when it is tested.
A rule is written, Shopping Tests run against it, Herm evaluates, violations surface, the team reviews and fixes, and the next scheduled run checks the result. Updating a rule can turn a passing suite amber, which is the point.
An updated rule applies in the next scheduled Product Readiness run, not instantly on every edit.
Aggregate Brand Intent performance at a glance.
How many rules are active, how many Shopping Tests they touch, how many violations are open, and which rules are breaking most often.
Bars show open violations by rule category, relative to the largest.
Some failures need a rule. Others need better product information.
A Brand Intent problem is one where the facts exist and the recommendation violates brand policy. A product-data problem is one where the information required for the decision is missing or ambiguous. Writing a rule does not fix the second kind.
Brand Intent is also not the Content Suite. It governs product recommendation, use case, market, retailer, claims and customer fit. The Content Suite governs article language, sources, author voice, banned terms and editorial QA.
Product recommendation, use case, market, retailer, claims and evidence, and customer fit.
Article language, sources, author voice, banned terms and editorial QA.
Show what the brand declared, and how the test was evaluated.
What Brand Intent commits to, and where the line between what Herm shows and what it protects falls.
- Product evidence and Brand Intent are evaluated separately, never collapsed into one truth score.
- Customer, market and use-case rules are entirely brand-controlled.
- Rules can be scoped from brand level down to a single variant.
- Shopping Tests use the active rules relevant to their scope.
- Conflicts surface for review. Herm does not silently resolve precedence.
- Changed rules apply on the next scheduled Product Readiness run.
- Users can inspect the exact rule any violation broke.
Rules, scopes, evidence, owners, status, conflicts, violations, Shopping Test results, reporting and history.
Evaluator implementation, precedence and matching logic, prompts, ontology, conflict-detection algorithm, weights, thresholds, evidence ranking, document-to-rule extraction prompts and orchestration.
Questions, answered honestly.
What is Brand Intent?
The layer of Product Readiness where your brand defines how products should, and should not, be represented, recommended and sold. Herm then tests shopping journeys against those rules.
Is Brand Intent the same as product data?
No. Product data describes what a product is. Brand Intent describes what is commercially acceptable to recommend, regardless of whether the underlying facts are true.
Does Brand Intent control ChatGPT, Claude or Gemini?
No. Herm evaluates shopping-agent behaviour against your Brand Intent. It does not publish these rules directly into external AI agents.
Can I create rules in plain English?
Yes. Rules can be written as natural-language statements, entered through a structured form, imported in bulk, or proposed by Herm from uploaded documents.
Can I upload existing policies?
Yes. Merchandising policies, claim libraries, positioning decks, legal guidance and product documentation can all be uploaded, and Herm proposes rules from them for human review.
Can a rule apply only to one market?
Yes. Rules can be scoped to a market, and Brand Intent can differ by market, seller, claim and availability.
Can rules apply to customer segments?
Yes. A rule can be scoped to a customer segment or ICP, so recommendations are evaluated against a defined customer context.
Can rules apply to retailers?
Yes. Seller and retailer restrictions are a supported rule category.
Can rules require evidence?
Yes. A rule can require approved evidence before a claim counts as supported.
What happens if rules conflict?
Herm detects the conflict and marks it review required. It does not silently resolve precedence between conflicting rules.
Can teams control who edits and approves rules?
Yes, through role-based permissions. Different departments can own and approve the rules relevant to their scope.
Is rule history available?
Yes. Every rule tracks who changed it, what changed, when, and why.
Are Brand Intent rules re-tested?
Yes. An updated rule applies in the next scheduled Product Readiness run, not instantly on every edit.
Can Brand Intent trigger alerts?
Yes. Rule violations, conflicts, previously passing tests that now fail, and market-specific violation increases can all trigger alerts.
Do not let AI invent your merchandising strategy.
Define the product, market, customer and claim rules your brand actually operates by. Herm tests whether AI shopping journeys respect them.
The free test covers a snapshot. The full Brand Intent workflow is part of Product Readiness.