Definition
A mobile measurement partner is a third-party attribution service for apps. Three parts make it work.
- An SDK in the app. It records the first open, which stands for the install, and the later events the advertiser chooses, such as registration, a trial start or a purchase.
- Matching. For each install, it looks for the ad interaction that most plausibly caused it, using the identifiers and signals each operating system permits, under windows and rules the advertiser sets.
- Postbacks. It sends server-to-server messages telling each network which installs and events it has been credited with, so that the network's optimisation learns from the same result the advertiser reports.
The outcome is a single, deduplicated view of acquisition across every source, with spend pulled in from the networks so that cost per install and cost per event can be read by source.
Why apps use one: deduplicating network claims
Each network sees only its own ads and credits itself with any install that follows one of them within its window. An app advertising on several networks will therefore see installs claimed more than once, and the sum of the networks' reports will exceed the installs the stores record. The MMP settles each claim once, which is a form of DeduplicationRemoving double counts when one conversion is reported by more than one source. across sources rather than within one. Some of the largest networks attribute installs themselves and do not share click-level data; for those, the MMP asks whether they claim an install and arbitrates between their answer and the others. The guide to app attribution, MMPs and SKAdNetwork sets out each of these functions in detail.
Platform guides take the partner as the normal route. The TikTok app installs setup asks for the measurement integration to be verified before spend, and the Google Ads app installs setup treats a partner as the usual alternative to Google's own SDK when an app buys from several networks.
In practice: events, windows and fraud filters
Configuration decides what the partner can report. The advertiser chooses which events to record and which to forward to each network; sending every event the app logs is noise, while missing the event a campaign optimises towards leaves the network blind. Attribution windows are set once and applied to every network alike, which is the main reason the partner's numbers differ from each network's own. Once events are credited, the CPA calculator turns spend and credited events into a cost per event by source. Most partners also reject installs that match known fraud patterns before crediting anyone, so fake installs stay out of both the reports and the bills.
Many partners also provide links, including deferred deep linksA link that opens a specific screen inside a mobile app rather than its home screen. that carry a person's destination through the store install, which joins the link a person tapped to the install that followed. The partner is not a product analytics tool, though: funnels, features and retention analysis sit elsewhere, as the deep links and app analytics guide explains.
Limits under platform privacy rules
Device-level matching depends on identifiers the operating systems increasingly restrict. On iOS, where a person has not agreed to be tracked across apps, Apple's SKAdNetwork reports installs to networks in delayed, aggregated form, with post-install behaviour compressed into a conversion value whose detail depends on volume. Apple's developer documentation describes how those postbacks work. The MMP can configure the conversion value schema and decode the postbacks, but it cannot recover detail the platform withholds. Its iOS figures are therefore partly modelled or aggregated, and the mobile measurement hub shows how teams combine them with incrementality tests and modelled views. Where launches are short, as in games, the entertainment and gaming marketing guide reads these delayed reports beside cohort retention rather than day by day.
A habit app counts its installs once
Pocketfern Habit App buys installs from three networks. In one month their dashboards report 6,300, 4,800 and 3,100 installs, a total of 14,200, while the two app stores record 9,600 installs in all, including organic ones. After the team routes measurement through an MMP with the same seven-day click window for every network, the partner credits 3,900, 2,700 and 1,200 installs to the three networks and treats 1,800 as organic. Day-seven retention for the third network's users is 9%, against 21% and 18% for the other two, so its cost per retained user is far higher than its cost per install suggested, and the team moves budget towards the first two.
Related terms
CPIAverage ad cost of each app install. is only as reliable as the install count beneath it, which the MMP supplies. App AttributionAssigning conversion credit to touchpoints. is the wider question the partner answers under its own rules. A Deep LinkA link that opens a specific screen inside a mobile app rather than its home screen. often comes from the same provider, and DeduplicationRemoving double counts when one conversion is reported by more than one source. across networks is the partner's core job.
Common confusions
- An MMP does not create installs or improve campaigns. It decides credit; the networks still do the buying.
- Partner numbers are not meant to match network dashboards. Different windows and deduplication make them differ by design.
- An MMP is not product analytics. It knows where users came from, not how they use the app.
- Privacy limits apply to the partner too. It cannot see what the operating system does not share.
- More events sent is not better optimisation. A few well-defined events beat a stream of everything the app logs.
