Define installs, trials, purchases and subscribers
Agree on the action you intend to buy before asking reports to agree. Name the event, its trigger, its counting unit, and the record that validates it. An event dictionary should let a product owner and a media buyer describe the same outcome without relying on a dashboard label. A shared label alone cannot establish that two systems count the same action.
For this dictionary, distinguish a measured first app launch from a store download, a trial entitlement from a trial button tap, and a completed purchase from a checkout attempt. Define a subscriber as a person or account meeting an agreed subscription condition. Decide whether the business needs new paying subscribers, all paid transactions, or active subscriptions at a particular time. Keep renewals and restored access identifiable.
Our team's Android photo-app measurement record keeps trial starts, active trials, and subscribers separate. The published case does not supply attribution windows or cohort maturity. Use it as an example of preserving event labels, rather than as evidence of an iOS measurement implementation.
Event dictionary template
These are proposed definitions to confirm against your implementation.
| Event | Trigger | Validation | Deduplication | Owner |
|---|---|---|---|---|
| Install | Defined first launch under the selected report's rules | Test-device journey and measurement record | Provider's documented install and reinstall treatment | Measurement owner |
| Trial start | Trial entitlement is confirmed | Entitlement or subscription record | Agreed entitlement identifier | Product owner |
| Purchase | Transaction reaches the agreed completed state | Validated transaction record | Stable transaction identifier across collection paths | Payments owner |
| Subscriber | Account first meets the agreed paid-subscription condition | Subscription status and related transaction | Agreed account and subscription rule | Subscription owner |
Trace the event from app to report
Follow a validated action through the app, collection layer, mobile measurement partner (MMP), partner mapping, and destination report. Record the event name and transformation at each step. The event data path provides a framework for assigning those checks. Identify where an event can be filtered, renamed, retried, or sent through another route.
Ask engineering to demonstrate the intended trigger in a test journey and compare it with the validation record. Then verify receipt in the measurement system and the configured partner destination. Keep timestamps, app version, environment, and a safe test reference in the QA record. Verify failed transactions and repeated attempts as well as the successful path.
If both the app and backend report an action, agree which system owns the event and how duplicates are handled. Check the actual integration rather than assuming matching event names provide deduplication. Separate test traffic from production reporting and document when a mapping change becomes effective.
For each destination, record whether it receives the event itself, a mapped conversion, or an encoded value. Keep a revision history when that mapping changes. The buyer needs to know whether a movement in reported purchases reflects changed behavior or a changed reporting definition.
Keep MMP and SKAN reporting distinct
SKAN means SKAdNetwork, Apple's attribution framework. An MMP is a measurement provider; it may present several reporting streams, including Apple postback data. Label the specific stream being reviewed. A provider's consolidated dashboard and its SKAdNetwork report do not become interchangeable because they share a login.
Apple's SKAdNetwork documentation describes privacy-preserving attribution through signed ads and install-validation postbacks. Treat the postback stream according to its framework rules, rather than expecting a record for every individual product event. Keep the product transaction ledger as the basis for validating what happened in the app.
Compare the report definitions
| Report | Attribution basis | Delay | Limitations |
|---|---|---|---|
| Product transaction or entitlement ledger | Product action and account or transaction rules | Document collection and validation latency | Does not assign advertising credit by itself |
| MMP attribution report | Selected provider method, integration, and configured windows | Check the specific report's freshness | Available coverage and dimensions depend on the collection and attribution setup |
| SKAdNetwork report | Apple install-validation postbacks and conversion-value mapping | Conversion-window timing, possible locking, and randomized postback delay | Version and privacy conditions affect detail and postback availability |
| Buying-platform report | Platform conversion settings and reporting methodology | Check the platform and conversion's processing basis | May use different windows, modeling, and date assignment; verify before comparing |
The first, second, and fourth rows are fields to verify in your own setup, rather than universal timing promises. Apple's multiple-conversion-window documentation explains that eligible SKAdNetwork 4 traffic can produce up to three postbacks, with random delays after a window closes or is locked. Data tiers affect available fields. Multiple postbacks must not be counted as multiple acquired users, and an absent conversion field must not be read as a measured absence of purchases.
Check the iOS reporting setup, including AdAttributionKit
Review the current app build, operating-system coverage, network integration, MMP support, and conversion schema together. Apple's AdAttributionKit overview describes signed attribution for installations and re-engagement, with conditional postback detail. It supports click-based re-engagement on iOS and iPadOS 18 or later. Confirm the intended journey and integration before including that signal in the test design.
AdAttributionKit and SKAdNetwork are distinct frameworks. Apple's interoperability documentation explains how they interact, including selection of one winning impression across both frameworks for a conversion. Ask the integration owner which framework paths are active and how the reporting layer handles them. Do not add both streams together as independent conversions without checking their relationship.
Record who controls conversion-value updates and which dictionary version those values represent. Confirm that the schema can answer the intended question within the available reporting windows. A value described as purchase quality needs an explicit mapping back to the defined purchase action or value range.
Use the proposed iOS measurement scope to list dependencies and owners. Our published cases cover Android. The iOS checks here describe work to scope and validate for the advertiser's setup; they do not establish historical iOS campaign results.
Reconcile windows, timezones and mature cohorts
Start reconciliation with the population. Write down platform, app version, market, campaign, event, and inclusion dates. Choose whether the report groups actions by acquisition date or event date. Give each acquisition cohort the same observation time before comparing trial conversion or revenue.
Separate attribution lookback windows from the time allowed for users to complete the business action. A click-to-install window answers a different question from the period used to observe a trial becoming paid. Record click and view settings independently, and note settings that the chosen integration does not let you align.
AppsFlyer's freshness and timezone guide documents report-specific processing schedules and time-zone support, including SKAN processing. Verify the specific exports and their cutoff before explaining a difference as missing conversions. Keep both export time and reporting period in the reconciliation record.
Reconciliation checklist
- Match the acquisition cohort, platform, market, and event definition.
- Record whether dates refer to acquisition, event occurrence, or postback receipt.
- Align the reporting time zone or document the remaining offset.
- Compare click-through and view-through attribution windows separately.
- Give cohorts equal observation time and check report completeness.
- Confirm currency and gross or net revenue treatment, including refunds.
- Verify event mappings, conversion schemas, and their effective versions.
- Identify duplicate collection paths and overlapping reporting streams.
- Record unresolved differences with an owner and next check.
Sign off the event dictionary
Approve the dictionary before making it the buying brief. Product should sign the meaning of the action, engineering its trigger and validation, measurement its reporting path, and the buyer its optimization mapping. Attach the approved version to the test plan and retain the previous version when definitions change.
The measurement deliverables can organize the event map, report basis, and outstanding dependencies. Give every unresolved item a named role and a condition for closure. If a purchase mapping remains unvalidated, restrict the decisions that depend on it until the required evidence is available.
At each review, state which report supports the decision and why. Keep disagreements visible at the smallest useful scope, such as a specific event, source, or date boundary. Reconciliation is complete when the remaining difference has an understood basis or a documented investigation; forcing every headline total to match can conceal the distinction the buyer needs to make.


