Skip to main content
The Vertical layer is Concierca’s market- and competitive-intelligence surface. This playbook is how an AI client uses it well: learn the data model first, route by the shape of the question, and ground every recommendation in retrieved signals. The full tool reference is in Vertical.

​
Start here: learn the model

If you don’t already know the vertical’s data model, do not call brand tools blind. Learn it first:
1

vertical__vertical_legend

Returns the live data model — channels, the signal catalog and what each signal means, example questions, benchmarks, lookups, and live coverage / counts. This is your map.
2

vertical__vertical_knowhow

Returns the getting-started guidance and agent rules — the playbook and method for reasoning over this vertical.

​
Route by question

For any brand, resolve first. vertical__vertical_resolve turns a brand name into the canonical entity the other tools expect — never hand-type a brand id into a card or compare call.

​
Grounding rules

These are what separate a grounded market answer from a plausible-sounding guess. Promote them in every response:
  • Check data_coverage before advising on a brand — thin coverage means a weaker claim.
  • Don’t invent missing statistics. If a number isn’t in the data, say so.
  • Brand-specific numbers beat generic best practice — a retrieved figure outranks a rule of thumb.
  • Ground recommendations in retrieved signals, not prior assumptions.
  • Distinguish facts from estimates explicitly.
  • Don’t fabricate exact rankings from quartile or band signals — a band is a band, not a rank.
  • If the dataset lacks a requested fact, say so rather than filling the gap.
  • Respect time windows. Quarterly ad signals attribute each ad to its start quarter with lifetime values. A row whose details carry period_status: "running" is the current, incomplete quarter — young ads under-report reach for weeks. Cite it as “quarter to date, still maturing”; never narrate a running quarter as a trend break or compare it 1:1 against a closed one.
  • Don’t mix grains. Some signals are monthly (e.g. geo_footprint), others quarterly (e.g. geo_reach_profile) — check each signal’s grain in the legend before combining numbers.
  • Reach is a proxy. Reach ≠ audience ≠ customers, and campaign rankings by reach are proxy rankings, never “performance”.
  • Retail signals are research, not permission. A stockist lookalike is a qualified candidate, not a gap claim; emails in the feed are public role addresses, and the list never implies a right to send (outreach rules depend on the recipient’s country).

​
Answer shape

Frame market answers as a chain the user can audit:
Your data shows X → the playbook says Y → therefore Z.
That keeps the retrieved evidence, the method, and the recommendation visible and separable.

​
Naming & counts

Prefer the canonical vertical_* tools. The feed_* names (feed_card, feed_search, feed_compare, feed_brand_resolve, …) are deprecated aliases — use the vertical_* equivalents.
Never hard-code dataset counts or coverage numbers. vertical__vertical_legend reports them live and is the authoritative source — read them at the time you answer, don’t carry a remembered figure. See Vertical for the complete tool list and parameters, and Routing for how market intelligence fits the wider tool map.