Attribution

What Is Incrementality Testing?

Q: How is incrementality different from attribution?

Attribution divides credit for conversions after the fact. Incrementality proves — through a controlled test — which conversions the ads actually caused. Attribution is bookkeeping; incrementality is science.

Q: How much spend do I need to run an incrementality test?

Rough floor: $30K in the tested channel over the test window, at least 300 conversions. Below that, the test can't detect a lift smaller than 30% — which means most real lifts get missed.

Q: How often should I run incrementality tests?

Quarterly for major channels, annually for smaller ones, plus one after any big creative or offer change. Incrementality is not stable across seasons.

Q: Is a ghost bid test as good as a geo test?

For brand search — yes, and cheaper. For social channels — no, geo lift is the more trusted method because auction dynamics complicate ghost bidding on Meta and TikTok.

Incrementality testing is a controlled experiment that measures the lift ads actually caused — the conversions that would NOT have happened without the ad spend. It's the antidote to attribution reports that overclaim credit for buyers who were coming anyway.

Worked example

You run a geo-lift test: cut Meta spend to zero in 5 cities (control), keep it running in 15 similar cities (test). Over 4 weeks, test cities do $2.1M revenue, control cities do $1.5M revenue-per-city × 5 = $500K. Meta spend in test cities was $200K. Incremental revenue = 2.1M − (control-city rate × 15) ≈ $600K. True incremental ROAS = $600K ÷ $200K = 3.0 — even though platform-reported ROAS was 6.5. The 6.5 was double-counting existing demand.

Benchmarks

Meta Conversion Lift (built-in): needs $10K+/mo spend on a campaign to work.
Google Conversion Lift: available above spend and conversion thresholds.
Geo-lift (DIY or via tools like Haus / Recast): the most flexible, works at any scale.
Typical incrementality of retargeting: 30–60% of platform-reported.
Typical incrementality of brand search: 5–20% (most of it would have happened anyway).

Why it matters

Every attribution model overclaims to some extent. Incrementality is the only method that tells you 'if I turn this channel off, do I actually lose sales?' Without it, budgets flow to the channels best at claiming credit — not the channels driving new demand.

Common mistakes

1.Running a lift test for < 3 weeks. Signal doesn't stabilise.
2.Using non-comparable geos. If test cities skew younger/richer, the 'lift' is demographic, not ad-driven.
3.Testing a channel that's < 15% of your spend. Signal-to-noise ratio is too low.
4.Treating one test as forever true. Incrementality shifts with creative, offer, and season — re-test quarterly.

Put Incrementality Testing to work

Free calculators

Incrementality Calculator MER Calculator

Related services

Performance Marketing Consulting Tracking & Analytics Setup

FAQs about Incrementality Testing

How is incrementality different from attribution?

Attribution divides credit for conversions after the fact. Incrementality proves — through a controlled test — which conversions the ads actually caused. Attribution is bookkeeping; incrementality is science.

How much spend do I need to run an incrementality test?

Rough floor: $30K in the tested channel over the test window, at least 300 conversions. Below that, the test can't detect a lift smaller than 30% — which means most real lifts get missed.

How often should I run incrementality tests?

Quarterly for major channels, annually for smaller ones, plus one after any big creative or offer change. Incrementality is not stable across seasons.

Is a ghost bid test as good as a geo test?

For brand search — yes, and cheaper. For social channels — no, geo lift is the more trusted method because auction dynamics complicate ghost bidding on Meta and TikTok.