Let each platform expand its own audience; judge the result on your books.
Expansion models belong to the networks: lookalikes, similar audiences and the targeting broadening inside Advantage, Performance Max and Smart+ are built from behavioural signals only the platform holds, and nothing outside the platform reproduces them. Realvertising decides what those models are seeded with — the cohort ranked by margin per customer rather than the largest available converter list, resized to each platform's seed floor and refreshed as cohorts mature. What comes back is graded outside the platform on the same quality ladder used everywhere else: qualified lead rate, close rate, refund rate and revenue per customer for the customers credited to each expansion audience. Where the platform's view and the graded view disagree, the disagreement is reported as a disagreement.
The seed decides what the model learns
Seed a lookalike with every converter and it learns the shape of a noisy population, refunds and unworked leads included. Seeds here are drawn from the margin-ranked cohort, trimmed of refunded and cancelled customers, and rebuilt on a cadence so the model is not studying customers who arrived under a different offer two years ago. Seed composition and size are recorded with each build.
A predicted value stays labelled predicted
Propensity and expected-value scores are computed by the business metrics service from historical joins between conversions, CRM records and collected revenue. They are inputs to targeting decisions and nothing more: a predicted value never enters reported revenue, never appears inside a ROAS figure, and is shown next to the realised number rather than in place of it.
Expansion is a hypothesis for an experiment
Attribution can credit an expansion audience with revenue recorded after exposure to it; it cannot say whether those customers would have arrived anyway. When the question is whether broadening added anything, Experiments runs a holdout or geo test and the incrementality estimate comes from there. The two numbers stay in separate columns and are never averaged together.
In this capability
- Margin-ranked seed lists
- Per-platform expansion controls
- Downstream cohort grading
- Holdout-tested expansion
Audiences
The rest of Audiences
Customer Quality
Every audience scored on qualified lead rate, close rate, revenue and margin per customer, refunds and retention.
Segments
Business segments defined on CRM and finance fields, then activated through each platform's supported matching route.
Value Signals
Sends collected value, not pixel value, back into network bidding wherever the API and policy allow it.
Suppression
Stops paying to reach people who already bought, accounts that never qualify, and cohorts that refund.
One goal. Every channel. Real business truth.
Connect your ad accounts, say what a good customer is worth, and let the Growth Brain work the objective inside limits you set.
No ad spend commitment · Connect one channel to start · [email protected]