Profila
Sign up
What people asked for that nobody could supply.
Demand intelligence
This is the use case that needs no data science team. You are not buying training data — you are buying the answer to a question no other method can reach.
Every research method available to you today measures what people did with what they were offered. Search volume tells you what people typed into a box that was already full of answers. Sales data tells you what they bought from what existed. Panels and surveys tell you what people say when prompted by your categories, in your words.
None of them can tell you what someone asked for that nobody could supply — because in every one of those systems, an unfilled request leaves no trace. The person searched, found nothing useful, and left. The absence is invisible.
We record it explicitly. When a declared intent cannot be matched, that becomes a record with match.kind: "none" — a first-class row, not a gap.
A material share of every release is demand nobody could fill. That is the row you cannot get anywhere else. The rate — what proportion of declared intents go unfilled, by category and market — is published once the corpus passes 100 contributors; below that it describes a handful of people rather than a market. Ask us and we will tell you where your category stands today.
Demand intelligence
| Question | Where the answer comes from |
|---|---|
| What are people in my category asking for that nobody is serving? | Unmet-demand records, grouped by taxonomy code and market |
| Is that a supply gap or a discovery gap? | Whether the candidates were absent or rejected at the relevance gate — the counterfactual set distinguishes them |
| Which markets have demand my category cannot currently meet? | Unmet demand by inferred_region, subject to the k-anonymity floor |
| Where does my category systematically misunderstand its customers? | Correction chains — people who had to re-state what they meant, and the taxonomy delta between the two |
| Is my product language the same as my customers’ language? | Declaration text against the category it was classified into |
| Am I over- or under-serving relative to stated demand? | Served vs unmet ratio for your categories |
Demand intelligence
The shape is real; the numbers are invented, at a scale this corpus has not yet reached. We would rather show you the form than dress an early-stage corpus up as a market report.
CATEGORY: Automotive (CT-1) · Market: CH · Period: Q3
Declared intents 1,284
Matched 971 (75.6%)
Unmet 313 (24.4%) ← the row you cannot get elsewhere
Top unmet themes (from declaration text, clustered):
"estate / kombi under 40k, petrol" 84 no domestic supply found
"EV with tow bar" 61 candidates existed, all rejected at
the relevance gate — a DISCOVERY gap,
not a supply gap
"certified pre-owned, 7 seats" 47 supply gap
Correction chains 38 resolution rate 0.71
Most common correction:
CT-1 (Automotive) → CT-441 (Real Estate) "parking space", not "car"
— a classification failure your competitors are making too
That last line is the kind of finding that changes a media plan rather than a model.
Readiness
A demand-intelligence deliverable contains no records — only counts, rates and clustered themes, subject to the same k-anonymity floor of 10 that governs the corpus itself. It is derived from data that is already anonymous, and aggregated again on top.
| Blocking criterion | Training corpus | Demand intelligence |
|---|---|---|
| Contributor scale (floor 100) | ❌ blocks | ❌ blocks — aggregate rates need a population to mean anything |
| Deep funnel rungs (none reached yet) | ❌ blocks | ✅ irrelevant — unmet demand is about what was asked for, not what was later bought |
| Consent surface for third-party training | ❌ blocks | ✅ not required — aggregates over already-anonymous data |
| Calibration curve | ⚠️ limits some uses | ✅ irrelevant |
That matters commercially as well as legally. The licensable: false verdict you will see on the Readiness tab is computed for the training buyer, and it is stricter than this product needs.
So the demand-intelligence product is blocked on one criterion rather than four, and that one is growth. Everything else it needs already exists and runs today: unmet demand is captured, correction chains resolve, and category and market breakdowns are produced with every release.
The question to ask us is not “is the corpus licensable” — it is “how many contributors are there in my category and market, and when is that enough to be directional?” That is a straight answer we can give from any release.
One caveat we will not paper over. Whether aggregate reporting to a third party needs any additional consent basis is a question we are putting to counsel, not one we have settled ourselves. We would rather raise it than have you find out we had not thought about it.
Benchmarking
Ask us the useful question
That is the question that decides whether this is directional for you, and it is a straight answer we can give from any release — unlike “is the corpus licensable”, which is computed for a different buyer.
What you can build and send on Profila is in the Business Guide; what it costs is in the Billing Guide.