Data

Data and methodology

You should know exactly what a number is made of before you put it in a board deck. This page is the methodology we show to customers, auditors and regulators.

Data passport

period: 2026-09
panel builds: 283
taxonomy: v2.3
method: panel IDs, web + app
disclosure: ≥ 100 IDs per cell
reproducible: yes, by build id

Source: clean-room panels

Partner data clean rooms process web visits and mobile app usage for a panel of real devices. Each visitor carries a persistent pseudonymous ID; web and mobile observations of the same visitor are linked inside the clean room. Monthly batches of about a billion rows reach Omnimetra as observations of visitor × site or app × time; no names, no contacts, no device identifiers.

Persistent pseudonymous visitor ID, web and app linked
Monthly batches, 1–3 years of archive
Attribute buckets (contact kind, platform, country, industry, company size) returned by our specification, never raw attributes

Labelling: the entity graph

Every site and app is labelled by country and vertical in a fixed order: public registries first, then our own crawl of the target niches (full page content kept for re-labelling without re-crawling), then AI labelling, then organisation enrichment by domain. Labels carry a source, a confidence and a history, and a resolver priority decides conflicts.

Resolver order: manual › registry › seed › list › content › AI › organisation › rule › audience › propagated
Site → brand → operator links, mirror families, licence statuses
Taxonomy versions are explicit; a past month is never recoloured silently

Computation: precomputed sets

Markets, segments, overlaps and cohorts are computed on precomputed sets of visitor IDs per entity and period inside the engine. A segment is an expression over sets; the counter evaluates it in under a second. Results are published per data build.

Weekly and monthly periods
Exact overlaps for any number of entities
Switching, retention and event cohorts on the same sets

The data passport

Every figure carries: the period, the build id, the taxonomy version, the method (panel IDs, web, app or both), the disclosure threshold and whether cells were hidden. Re-running the same query on the same build returns the same number. A newer build shows the difference explicitly.

Compliance matrix

A rule per country × vertical × action (view, segment, activate) with its legal source and effective date, evaluated server-side on every segment, media plan and activation. Segments on gambling intensity or vulnerability are not available in any pack. Activation is only possible through a partner clean room, above the minimum size, under a workspace contract.

Disclosure policy

Minimum cell	100 visitor IDs; smaller cells are hidden, not zeroed
Market reports	cells under 100 IDs and denominators under 1,000 are hidden
Subtraction	hidden cells cannot be recovered by differencing; complements are suppressed
Row-level data	never leaves the engine; exports contain aggregates with passports
Activation	partner clean room only, ≥ 100 IDs, journaled, where the compliance matrix allows
Personal data	none received, none stored, none shown

What we tell you upfront

A panel is a sample. Shares and overlaps are exact for the panel; absolute market sizes are calibrated against public sources and labelled as estimates.
Coverage differs by country and vertical; the passport shows the panel size behind every number.
Apps are measured where the panel carries a mobile mark; where it does not, the number says web only.
Labels can be wrong; every label has a source and a history, and corrections are versioned.

Omnimetra

Ask for the coverage sheet and the methodology PDF

We send them for your country and vertical, together with the demo.

Request a demo

Questions

Is this clickstream data?

It is panel behaviour at the level of visitor × site or app × period, processed in clean rooms. We do not work with URL-level clickstream, keystrokes or content of pages visited by the panel.

Can numbers be audited?

Yes. Every number is reproducible by build id, the methodology is published, and the disclosure policy is enforced in the engine with a CI test that rejects exports containing identifiers.

Which countries and verticals are covered?

All markets of the first vertical pack (iGaming) and the verticals of the panel's taxonomy, with coverage shown per country. Ask for the current coverage sheet.