Define the outcome
Choose the decision the test must support before choosing where to buy traffic. You might need to establish whether an app can acquire trial starters in a new market, whether a source can deliver validated purchases, or whether the install journey works at all. Give the test one primary question. Record other useful signals as supporting evidence, with a clear explanation of what they can establish.
An install, a trial start, and a purchase answer different questions. Write the event trigger in plain language and have the product owner approve it. For a purchase test, decide whether the event means a completed transaction, a first paid subscription, or a renewal. Keep a trial button tap separate from a confirmed trial entitlement. These definitions belong in the plan before anyone sets an optimization goal.
Match the decision to the evidence
| Decision | Evidence to record | Check first |
|---|---|---|
| Install cost | Spend and attributed installs | Same market, platform, period, and install definition |
| Trial quality | Confirmed trial starts and subsequent paid conversions | Same trial trigger and observation window |
| Purchase quality | Validated purchases and revenue | Same purchase definition, revenue basis, and cohort maturity |
Use the table to identify missing evidence. If purchases cannot yet be validated, describe purchase quality as unresolved. You can still test delivery or the install journey, but that narrower question needs its own acceptance criteria. Set those criteria from the app's economics and reporting readiness rather than borrowing another app's target.
Record a primary outcome, a quality guardrail, and the action each result would support. A guardrail might require that the acquired audience reaches a useful product action without a material deterioration in payment quality. Assign an owner to define what material means for this app. The outcome planning framework can help connect the business question to the event being purchased.
Separate the test cells
A test cell is the combination you intend to evaluate together. Define its market, platform, optimization event, and source. Add the creative family, destination, and app version in the supporting record. Separate a market comparison from a source comparison wherever the available setup allows. If several conditions change together, label the result as a combined setup comparison.
Use this blank planning table before launch. Fill each row with an actual planned cell; the empty rows deliberately contain no assumed budgets or results.
| Test cell | Market | Platform | Event | Source | Owner |
|---|---|---|---|---|---|
Keep a separate launch record for budget authorization, start time, audience exclusions, creative IDs, and reporting access. Give every cell a stable identifier that survives campaign renaming. A reviewer should be able to find the correct campaign and event export from the planning row without asking the buyer to reconstruct the setup.
Resist opening more cells than the team can maintain and interpret. Ask what each added cell would change in the next decision. If the same action would follow either result, consolidate the question or postpone it. This is a planning rule, not a prescribed campaign count or spending threshold.
Write down overlaps that cannot be removed. Shared audiences, concurrent store tests, and product releases belong in the record. A controlled comparison requires a design that separates the treatment from other changes. When that design is unavailable, the useful output is a directional operational read, with its limitations attached.
Match the channel to the app
Evaluate sources against the intended event, supported app platform, available creative, destination, and reporting path. Ask what the account can actually run and what the app can send back. Treat eligibility and integration checks as dependencies. A source should enter the plan because it can help answer the question under the current setup.
Google's App campaigns documentation distinguishes campaigns for installs from campaigns for engagement and describes optimization for in-app actions. It also explains that the system assembles ads from supplied assets and store-listing material. For a proposed test, confirm the campaign type, selected conversion, and available assets together.
For each source under consideration, collect a short written answer to these questions:
- Can this account run the intended objective for this app and platform?
- Which event will the campaign optimize toward, and who has verified its mapping?
- Which approved formats and localized messages are ready for delivery?
- Where will a person land after the click, including the fallback journey?
- Which report will support the decision, and which dimensions will be unavailable?
Use the buying mix overview to structure that review. Record pending checks as pending. Do not translate an unverified capability into a committed launch cell. If a required format is missing, assign its production and approval before setting the test start.
Prepare the destination alongside the ad. Review the store listing in the target language and confirm that the advertised benefit is visible in the app. A campaign testing a trial promise needs a working path to that trial. If the destination must change during delivery, record when it changed and decide whether the affected period remains comparable.
Read the result on a common basis
Before comparing costs, align the spend numerator with the event denominator. Specify currency, included fees, event filters, attribution basis, and reporting dates. Cost per purchase should use the spend associated with the sources and period that supply those purchases. Keep an unreported event cell visibly unreported.
Our team's five-source cleaner case shows why coverage matters. The published Android case reports installs and trial starts across all sources, but purchase coverage is limited to Mintegral, Meta Ads, and Unity Ads. TikTok Ads and Google Ads purchases are unreported, rather than zero. Its narrower purchase comparison cannot establish an all-source purchase ranking. The page does not supply attribution windows or cohort maturity.
Choose an observation window appropriate to the event being evaluated and apply it consistently. Compare groups acquired during a defined period after giving each group the same time to develop. Distinguish that cohort view from a calendar report of transactions occurring during the period. Record which one you use beside every decision metric.
Also record when each export became available. AppsFlyer's data freshness documentation describes different update schedules across reports and support for app-specific time zones. A fresh export from one dashboard does not establish that another report is equally complete. Agree a reporting cutoff and check completeness before reviewing differences.
Use the event data path to locate gaps between the app, measurement provider, and buying platform. Keep delivery findings separate from measurement defects. If the required event is missing, fix and validate its path before treating the absence as weak audience quality.
Before you launch
Set review conditions while the outcome is still unknown. Identify who can pause a broken destination, who can authorize a budget change, and who decides whether the business question has enough evidence. Specify what happens if the review is inconclusive: extend the agreed observation, repair the setup, narrow the question, or stop the test. Avoid changing the decision rule after seeing a favorable partial result.
Prelaunch checklist
- Validate the primary event against the product or transaction record.
- Confirm the event sent to each source matches the approved definition.
- Test the destination and fallback on the intended platform.
- Record market, language, platform, audience, and creative version for every cell.
- Agree attribution windows, time zone, currency, and revenue treatment.
- Define cohort maturity and the reporting cutoff for the review.
- Confirm campaign, measurement, and export access for the people reviewing results.
- Assign the decision owner and the person responsible for each dependency.
- Record budget authorization and conditions for pausing or changing delivery.
- Save the planning table and the first review date in the shared test record.
At the review, write the decision beside the evidence used to make it. Include what remains unknown and the smallest follow-up test that could resolve it. That record is the basis for the next allocation, even when the current test ends without a source to scale.


