H

What is Holdout group?

Definition, examples, and more

Definition

A randomly selected share of users who keep seeing the current experience while everyone else gets a change, such as a new paywall or personalized offers. Comparing the two groups shows how much of the difference the change caused.

Example

A fitness app turns on personalized offers for 90% of new users and keeps 10% on its existing paywall. After eight weeks, revenue per paywall view is higher in the personalized group than in the holdout. Because users were assigned at random, the gap measures the effect of personalization rather than seasonality or a change in ad spend.

Why Holdout group Matters

Before-and-after comparisons mix the change with everything else that moved at the same time: marketing, seasonality, app updates. A randomized holdout runs at the same time as the change, so the difference between the groups is the causal lift. Botsi measures every deployment against a randomized holdout for this reason.

Frequently Asked Questions

How big should a holdout group be?

Big enough to measure the difference you care about within a reasonable time, usually 5–20% of traffic. Smaller holdouts cost less revenue but take longer to read.

How long should a holdout run?

Long enough to capture the revenue that matters. For subscription apps that often means first renewals, so 4–8 weeks is common for monthly plans and longer for annual plans if you want renewal data.

Is a holdout group the same as a control group?

In practice, yes. "Control group" is the usual term in A/B tests; "holdout" usually means a group kept on the old experience while a system such as a personalization model runs for everyone else.

Category
Subscription App Terminology
Related Area
Mobile App Growth & Monetization

More terms starting with “H”

Show every user the right offer

Botsi picks the offer and paywall each user sees, from options you approve, and measures the lift against a holdout.