Platforms can credit a campaign with people who might have bought anyway. A holdout test compares exposed and unexposed groups, but a small budget calls for discipline and limited expectations.

Choose one clear question

Examine a specific channel, audience and outcome. If many campaigns change simultaneously, the difference cannot be attributed clearly.

Create genuinely comparable groups

Allocation should reduce systematic differences. A geographical holdout or random exclusion needs checks for size, historical behaviour and interaction.

Calculate whether there is enough signal

With very few conversions, a test cannot distinguish a small effect from noise. Define in advance the minimum difference that matters to the business.

Do not adjust the experiment halfway through

Changes to targeting, the offer or measurement destroy the comparison. Record external events and wait for the agreed observation window.

Provide a range rather than an absolute verdict

Present the estimate and uncertainty. An inconclusive result may indicate the need for a larger test, a stronger intervention or a different question.

This framework is an original editorial methodology developed by DigitalNow.

DIGITALNOW EDITORIAL TEAM

Practical guidance from DIGITALNOW, part of VNG Digital Group.

Editorial policy