Home/Data platform

A DMP built for the supply side.

Most data platforms were designed to follow a person across the web. Ours was designed to understand an impression inside an app: where it is, what surrounds it, how likely it is to be seen and acted on. The output rides with the deal, not with an identifier.

01Architecture

Five stages, one request, a few milliseconds.

Every bid request that reaches Inception passes through the same pipeline before a deal is offered. Nothing is applied after the fact.

02Ingest

Signals from the supply, with consent, in real time.

We collect what the supply side legitimately knows about an impression, and nothing that requires following a person from one app to another.

Publishers we represent can share consented first-party context: content category, session stage, in-app events they choose to expose. Everything else comes from the request itself and from our own classification of the app and placement.

  • RequestApp, bundle, placement, format, size, position, device class, OS, connection type, coarse geo, consent and ATT state
  • First-partyConsented publisher context: content category, session depth, level or screen, in-app events the publisher elects to share
  • ClassificationApp and content categories mapped to the IAB Content Taxonomy, placement typing, format mix
  • OutcomesWin, render, viewability, completion and interaction events, returned per placement

03Model

Three scores, computed per placement, per request.

The models are trained on outcomes from inventory we sell, so they describe what actually happens in these apps rather than a general assumption about mobile.

Engagement

Will this placement be seen and acted on?

Predicted viewability and interaction likelihood from placement type, session stage, format mix and historical outcomes for that app and position.

Completion

Will a video play through?

Completion probability for opt-in and interstitial video, used to gate which placements qualify for outcome deals.

Quality

Should this request be offered at all?

Invalid-traffic and authorization status, device and connection anomalies, placement consistency. Low-quality requests are dropped before any buyer sees them.

04Segment

Audiences defined by the seller, expressed in standards.

We package context and predictions into seller-defined audiences: cohorts described in standard taxonomies that any DSP can read, without a device identifier.

Because the segments are built from what the supply side knows, they remain available when identifiers are absent. An agency can plan against a cohort today and still be buying the same cohort next year.

05Activate

Two ways the signals reach a buyer.

Deal IDs

The primary path. Signals and constraints are applied when the deal is assembled, so the buyer activates a pre-qualified package in its DSP.

Where
Any DSP, through the SSPs it already uses
Control
Publisher list, format, geo, cohort, floor, pacing
Change
Adjusted by Inception against the buyer's KPI

Supply-side targeting

For buyers who want signals on the open request as well. Cohorts and scores are attached to the bid request in standard fields for the DSP to act on.

Where
On the OpenRTB request, in standard extension fields
Control
Buyer reads and targets the cohorts in its own bidding logic
Change
Taxonomy published to the buyer, versioned

06Measure

Compatible with how app marketers already measure.

Deals report at the level agencies need: app, placement, format, cohort and outcome. For app-install and re-engagement campaigns, delivery is attributable through the mobile measurement partners buyers already run.

  • Deal reportingDaily, per deal, per buyer, per app and placement, via dashboard or API
  • AttributionWorks with leading mobile measurement partners; SKAdNetwork and AdAttributionKit on iOS, install referrer on Android
  • VerificationOMID-compatible placements; standard viewability and brand-safety vendors supported
  • FeedbackOutcomes return to the models, so the next deal is built on what the last one did

07Privacy

Designed to work without following anyone.

The platform's value does not depend on identifiers, which is why it did not lose value when they went away.

No cross-app profiles

Cohorts are computed per request from context and consented publisher data. We do not build or sell user-level profiles.

Consent honoured

TCF 2.2 and GPP strings, ATT status and regional requirements are read on every request and respected in what we process.

No fingerprinting

We do not derive identifiers from device characteristics, and we do not accept them from partners.

Data minimization

We keep what the models need, for as long as they need it, and nothing that would let a person be re-identified.

See the signals on your own brief.

Send a vertical and a KPI, and we'll show which cohorts and scores apply.