Back to Knowledge

Lift Test

Method
Attribution

Definition

Measures the true incremental impact of a marketing campaign by comparing a group exposed to the campaign against a control group that wasn't exposed.

Related Knowledge

Incrementality Attribution Model

Lift formula

Lift=Test group conversion rate−Control group conversion rateControl group conversion rate×100\text{Lift} = \frac{\text{Test group conversion rate} - \text{Control group conversion rate}}{\text{Control group conversion rate}} \times 100

For example, if your test group converts at 4% and your control group at 3%, your lift is:

\text{Lift} = \frac{4\% - 3\%}{3\%} \times 100 = 33\%

This means your campaign drove 33% more conversions than would have happened without it.

Types of lift tests

  • Conversion lift – Measures impact on purchases, signups, or other key actions.
  • Brand lift – Measures impact on awareness, consideration, or purchase intent (typically survey-based).
  • Search lift – Measures whether ads increase branded search volume.

    • Common approaches

  • Platform-native tests
    • Meta, Google, TikTok and others offer built-in lift testing tools.
  • Geo-based tests
    • Show ads in some regions, hold out others, then compare results.
  • User-level holdouts
    • Randomly exclude a percentage of your audience from campaigns.
  • Ghost ads / PSA tests
    • Control group sees a public service announcement instead of your ad.
  • Baseline comparison
    • Compare against historical performance under similar conditions to sanity-check results.

    • Example scenario

    You are spending €50,000 / month on retargeting. Attribution says it is driving 1,000 conversions.

    A lift test reveals that the control group (no retargeting) converts at nearly the same rate. The true incremental lift is only 10%, meaning retargeting actually drives ~100 incremental conversions, not 1,000.

    So your real CPA is:

  • Attributed: €50 CPA (€50,000 ÷ 1,000)
  • Incremental: €500 CPA (€50,000 ÷ 100)

    • Challenges

  • Requires sufficient traffic volume to reach statistical significance.
  • Test duration must be long enough to capture the full conversion cycle.
  • Platform-run tests may be biased, since platforms are incentivized to show positive impact.
  • There is an opportunity cost of holding out users from campaigns during the test.