Measurement
Part of Programmatic experimentation
Measuring the effect of placement exclusions
Test an optional placement exclusion and measure its whole-buy effect on delivery, spend, outcomes and replacement inventory.
Run a concurrent controlled test with a baseline arm where candidate placements remain eligible and a variant arm where they are excluded. Compare the whole-buy outcomes, delivery and the placements that absorbed any shifted spend.
Choose an exclusion that may be tested
Define the candidate domain, URL, app ID or group using exact identifiers, and state the performance question. Before launch, specify the completed campaign outcome, the improvement needed to adopt the exclusion and guardrails for suitable inventory, spend and reach.
Keep mandatory brand-suitability and other approved exclusions in every arm. Test an optional exclusion only when both conditions are permitted by the brief. Judge the whole eligible buy after removal, including replacement inventory.
Key guardrails for testing placement exclusions
- Mandatory exclusions
- Brand-suitability and approved exclusions must remain in both arms
- Spend threshold
- Ensure sufficient spend in each arm to avoid sparse data
- Reach requirement
- Maintain minimum reach to ensure statistical validity
- Budget allocation
- Keep allocation steady during test; adjust only if unavoidable
- Outcome target
- Pre-agree on required completed-outcome volume before launch
Set up the comparison
Align market, inventory routes, formats, audience, creative, destination, bid strategy, budget basis and measurement across the arms. Change only candidate-placement eligibility, set arm budgets in proportion to the experiment split, and match pacing.
Keep budget allocation between arms steady during the test. If allocation cannot be held steady, do not treat arm-level spend differences as a clean placement-only comparison.
In Display & Video 360, use an ad experiment and its app, channel and website targeting controls. Inspect the saved settings: the candidate should remain eligible in the baseline and be excluded in the variant.
Set the required completed-outcome volume and a stopping date before the test starts. If either arm falls short, treat the result as inconclusive rather than choosing a winner from sparse outcomes.
Measure what replaced the placement
Use the same placement-level delivery report for both arms, with matching dates, filters and spend basis. Break out spend and impressions by candidate identifier and by other placements, then identify which non-candidate placements gained spend in the variant.
Compare actual spend, impressions, completed outcomes and cost per outcome using the same definitions and measurement window. Reconcile placement-level changes with each arm’s total spend; lower total delivery is not spend displaced into replacement inventory.
Where available, use diagnostics to check whether the relevant targeting filtered requests. A filter count supports an implementation check, while a candidate’s later absence from placement reporting does not by itself prove that the exclusion caused a business improvement.
A before-and-after comparison can be affected by changes in auctions, season, creative or tracking. A controlled split improves the comparison, but sparse outcomes can still limit the decision; the changed placement mix is part of the exclusion’s effect.
Adopt the optional exclusion only if the whole-buy outcome meets the pre-agreed improvement rule and the guardrails hold. Keep the baseline if the result worsens, and specify what a bounded follow-up must establish if delivery or completed outcomes are too sparse.
