Paid Media
Incrementality
Analysis
A causal analysis showing why platform-reported performance overstated true paid media impact, and how budget could move toward audiences with stronger incremental lift.
The dashboard rewarded
proximity, not causality.
Dashboards showed 3.0x ROAS and €106 CPA. Incrementality analysis showed 1.25x ROAS and €246 true incremental CPA. The gap came from attributed conversions that would likely have happened anyway, especially in retargeting audiences.
Reported CPA understated the real cost
Reported CPA was €106, while true incremental CPA was €246.
57% of attributed conversions were not incremental
Paid media reported 61425 monthly conversions, but only 26400 were incremental.
Retargeting was overvalued
Meta Retargeting Hot looked best at €55 reported CPA, but rose to €500 incremental CPA after adjustment.
Prospecting was undervalued
TikTok Broad and Meta lookalike audiences produced stronger incremental lift than retargeting.
Frequency waste started after 9 impressions
Incremental conversion peaked at 6–9 impressions, then declined as frequency increased.
Reallocation added growth without more budget
Moving spend from retargeting to prospecting added +8400 incremental conversions per month.
Four layers, applied in sequence.
Cohort definition
Defined exposed and comparison users across paid media audiences, markets, channels, and observation windows.
Propensity score matching
Matched exposed users with comparable unexposed users using pre-campaign behavior, country, device, loyalty, and organic activity.
Incrementality by audience
Compared reported conversions, incremental conversions, reported CPA, true incremental CPA, and lift rate by audience.
Reallocation simulation
Modeled how moving budget from low-incrementality retargeting to higher-incrementality prospecting would change results.
What the causal
read showed.
Matching aligned treated and control groups on pre-campaign behavior before measuring lift.
Cheapest reported CPA audiences were often the most expensive on an incremental basis.
Incremental CVR peaked at 6–9 impressions. Beyond 9, each additional impression yielded declining return.
Move budget toward
true lift.
Retargeting was taking too much credit for conversions likely to happen anyway. Budget should move toward TikTok and Meta prospecting, where the ad had more room to change behavior.
More incremental growth,
no budget increase.
The recommendation kept spend flat while moving budget toward audiences with stronger causal impact and away from retargeting segments that looked efficient only in platform reporting.
Use incrementality as
the media decision layer.
Replace reported CPA with incremental CPA
Use reported metrics for monitoring, but use incremental CPA for budget decisions.
Move €1.3M from retargeting to prospecting
Shift budget toward TikTok Broad and Meta lookalike audiences with stronger lift.
Cap retargeting frequency at 9 impressions
Reduce waste once incremental conversion starts to decline.
Use UGC as the default for prospecting
UGC outperformed branded studio content by 17 percentage points on incremental lift.
Review market allocation
Germany was underinvested in prospecting despite having the lowest brand awareness across the three markets.
What I owned.
I structured the incrementality analysis, prepared exposed and comparison cohorts, applied propensity score matching, compared reported and incremental performance, analyzed audience-level efficiency, modeled reallocation scenarios, and translated the findings into media budget recommendations.
Explore the rest of the work.
The repository includes cohort definitions, propensity score matching, audience-level incrementality comparison, frequency analysis, and reallocation modeling.