Growth Marketing Glossary

Holdout test

holdout testnoun

The closest marketing gets to a clean experiment — show some, hide from others, measure the gap.

holdout — treated vs controlaudiencetreatedheld out
Schematic — holdout test design
Method
randomised control group
Measures
incremental lift, not credit
Answers
would it have happened anyway
Forms
user-level or geo holdouts

Forms & parts of speech

holdout test · noun
Withholding ads from a control group to measure incremental lift.
"The holdout test showed brand search would have converted without the ad."

What it measures

A holdout test splits an audience at random into a group that sees the advertising and a control group deliberately kept from it. Comparing outcomes between the two isolates the lift the ads actually caused.

This answers the question attribution cannot: would these conversions have happened anyway? If the control group converts almost as well as the treated group, the ad added little real value, whatever the attribution report credited it.

How it is run

Holdouts come in forms. User-level holdouts withhold ads from a random slice of individuals; geo holdouts turn spend off in some regions and on in others, useful when user-level control is impossible after privacy changes.

The discipline is in the design — true randomisation, enough scale for statistical significance, and a clean control that genuinely sees no ads. Done right, a holdout is the strongest causal evidence in marketing; done sloppily, it misleads with confidence.

Worked example. Suppose attribution credits brand-search ads with thousands of conversions. A holdout test withholds those ads from a random control group and finds the control converts almost as often — the buyers were already searching for the brand and would have arrived anyway.

The measured incremental lift is small, so the brand reallocates that budget to channels a holdout shows are genuinely additive. The experiment overturned a confident attribution number with causal evidence.
Failure modes to watch. Running a holdout without true randomisation or enough scale; contaminating the control group so it still sees ads; and generalising a single test's result beyond the conditions it ran under.

Benchmarks

A holdout's result is specific to its design and conditions; there is no portable benchmark. Judge it on randomisation quality and statistical significance, not a target lift.

Design
randomised control
Measures
causal lift
Forms
user-level or geo

Ranges are illustrative; every published figure is cited from a named public source or labelled “RGM analysis.”

Synonyms & antonyms

Synonyms

incrementality testcontrol group testgeo holdout

Antonyms

last-click attribution

Usage trends

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Common questions

What is a holdout test?
An experiment that withholds advertising from a randomised control group and compares outcomes to a treated group, measuring the true incremental lift the ads caused.
Why use a holdout test instead of attribution?
Attribution credits touchpoints on the path but shows correlation; a holdout measures causation by revealing what would have happened without the ads.
What is a geo holdout?
A holdout run by region — ads on in some geographies and off in others — used to measure incrementality when user-level control is not possible after privacy changes.

Related tools & calculators

Resources & people to follow

Curated, non-competitor resources verified per term.

Sources

  1. trendsGoogle Trends — "holdout test"