Jampp recommends holding out around 6% of users for its continuous advertising measurement. Its August 27, 2026 explanation describes random assignment and observation windows tailored to each campaign.
The decision turns part of the audience into an economic reference: what would have happened without that campaign’s advertising. For the buyer, the comparison helps distinguish additional purchases from purchases merely attributed to an ad.
An unexposed reference retains an opportunity cost
The Jampp platform markets measurement for app acquisition and retargeting. Its described method uses Ghost Bids, records of bid opportunities suppressed for the control group, without purchasing substitute advertising for that group.
Avoiding that additional spending does not make the decision free in every sense. The advertiser withholds its campaign from some users. That choice purchases information about advertising’s effect and needs to be justified by the decision it supports.
Around 6% is the provider’s recommendation. The observations needed depend on the outcome’s frequency and the effect size being distinguished. The percentage alone cannot guarantee that every campaign answers the same question with equal precision.
The cost of acquiring an audience varies before campaign impact is estimated. Singular data published by Liftoff and carried by Statista places casual-game cost per install between February 2024 and February 2025 at $0.14 on Android and $1.41 on iOS. These are observed campaign averages without random assignment and with different mixes of users and games. They do not indicate cost per additional purchase or Jampp performance. They explain why spending and comparisons must retain platform, audience and objective when part of an audience is reserved for control.
An install has different costs by platform and genre
Observed average cost per install in four selected genres, February 2024–February 2025.
USD
Tap or hover for details. Select a series above to show or hide it.
Collapse notes
Singular and Liftoff: global sample of 1.1 trillion impressions, 36 billion clicks, 2.4 billion installs and $11.9 billion spending. Genres follow Sensor Tower. Different campaign mixes; observed costs, not causal platform effects or incremental purchases.
Explore the data table
| Category | Android | iOS |
|---|---|---|
| Puzzle | 0.64 USD | 2.80 USD |
| Casual and hypercasual | 0.14 USD | 1.41 USD |
| Strategy | 1.67 USD | 4.16 USD |
| RPG | 4.29 USD | 8.29 USD |
The daily window still contains earlier days
Jampp describes a rolling window adding one day and dropping the oldest. Daily recalculation keeps the observation recent, but consecutive windows share data. Treating every update as an independent experiment would exaggerate the amount of new evidence.
That detail matters when the figure changes a budget. A result persisting for several days can come from many of the same people. The decision needs to establish how much information changed, alongside how often it appeared on the dashboard.
The reading date also determines which outcome can be observed. In Adjust’s aggregated cohorts carried by Statista, 2025 gaming retention was 27% the day after installation and 5% on day 30; e-commerce was 12.6% and 3%. These are activity rates among installed users without identifying how much advertising caused. A window designed for rapid installations does not by itself answer a question about later purchases. Jampp’s continuous measurement must retain the same event and allow its horizon to mature before comparing windows.
Post-install activity changes with the measurement day
Average retention by category in apps observed by Adjust during 2025.
Tap or hover for details. Select a series above to show or hide it.
Collapse notes
Adjust combines its top 5,000 apps with its complete tracked app dataset, using two geographic sources of 45 countries and about 250 territories. Averages of daily install cohorts; specific sample size is not reported. It may not represent the whole market and does not measure revenue, purchases or Jampp causal lift.
Explore the data table
| Category | Gaming | E-commerce | Finance |
|---|---|---|---|
| Day 1 | 27.0% | 12.6% | 12.0% |
| Day 7 | 13.0% | 6.0% | 6.0% |
| Day 30 | 5.0% | 3.0% | 2.0% |
| Visible measure | Condition needed to interpret it |
|---|---|
| Holdout percentage | Who can enter the comparison and how assignment works. |
| Updated lift | Window duration and overlap with the previous window. |
| Additional revenue | Margin and costs associated with the observed outcome. |
Coverage defines the conclusion’s reach
User measurement also depends on available signals. In its historical iOS 14.5 explanation , Jampp limited individual IDFA analysis to consenting users. That document illustrates why a comparison needs to identify its population; it cannot describe the full current architecture by itself.
A sample of observable users can provide useful evidence about them without automatically representing every customer. The commercial proposition becomes more precise when it explains that boundary and avoids turning a partial observation into a total growth promise.
The final step is economic. An additional purchase still requires calculating its contribution after advertising and delivery costs. Positive lift can justify continuing a campaign without justifying every price or budget increase.
Jampp connects media buying to a comparison of its effect. Its commercial advantage becomes more defensible when the advertiser understands the audience, the window and that effect’s contribution. Holding out traffic purchases evidence; spending decisions need to follow what that evidence establishes.
In this article
Marian Vasile
Marian Vasile is a MarTech and growth professional. His work spans product marketing, SEO, paid advertising and deep online tracking, including behavioral analytics and attribution.
