Mobile Measurement Partner

On the web, finding out where someone came from was never hard. You added UTM parameters to the end of an ad link — a short tag written into the address itself that records which channel and which creative the click came from. The browser carried that tag along, so the line from ad click to signup stayed unbroken the whole way.

In a mobile app that line snaps in the middle. Someone who taps an ad doesn't land on your product. They land in an app store, tap install, wait for the download, and open the app some time later. The tag the browser was holding is gone by then. All the app knows on first launch is that somebody opened it for the first time.

So the weekly meeting at a team running app install ads in three places always looked the same. Installs were up 1,200 this week, and all three ad accounts reported having produced 900 of them. Add the claims together and you get 2,700, more than twice what actually happened. The app store reported the install total and each ad network reported its own performance, with nothing in between to reconcile the two — so no claim could be checked, and the loudest channel took home more of next month's budget.

A mobile measurement partner (MMP) is the neutral referee built for that gap. A third party, not the company that sold you the ad, joins installs to clicks and applies one set of rules to every channel.

The tool answers a single question: which ad produced the install that just arrived. It gets there in four steps.

1. The moment someone taps an ad, that click is logged on the tool's servers along with the network, the creative, the timestamp, and the device model.
2. The person goes through the app store, installs the app, and opens it. The app ships with the tool's SDK — a package of code meant to be dropped inside someone else's app — which reports that first launch back to the same servers.
3. The tool matches the two records, deciding which of the stored clicks the new first launch belongs to, using device details and timing.
4. It hands the verdict to both the ad network and your team in the same shape. The network collects its share; the team sees installs by channel in one table.

Matching needs rules. Set the attribution window to seven days and a click from eight days ago won't be joined to this install; if one person tapped two ads in the same week, the later click takes the credit by default. Deciding which touchpoint deserves the credit is a whole topic of its own, covered under attribution. What matters here is that these rules are a setting, not a fact. Shorten the window from seven days to one and the same ad's performance for the same week drops by more than half.

The joining doesn't stop at the install. If the person upgrades to a paid plan three days later, that payment attaches to the same ad, which is what makes customer acquisition cost (CAC) and return on ad spend (ROAS) available per channel. A channel that spent $4,500 and produced 3,000 installs looks strong until you see 12 paid conversions behind them, which works out to $375 per customer — while a channel with 900 installs and 80 conversions is far cheaper. Ranked by installs alone, that judgment flips.

Three approaches get used instead, and each holds up in a different situation.

1. Take the ad platforms' own numbers at face value. It costs nothing and needs no setup, but you inherit the double counting described above. With a single channel there is nothing to double count, so this is enough.
2. Lean on a web analytics tool. It shows what happens on the web clearly, but it has no view of the stretch that passes through an app store. If most of your traffic arrives on the web, there is no reason to add another tool.
3. Build it yourself. Storing clicks and matching them to first launches is not complicated in principle. The trouble is that every ad network hands over its values in a different format, and those formats change from year to year, so five partners means five things to keep repairing.

Teams usually tip toward buying at the same point: more than two paid channels and monthly ad spend somewhere around $15,000. Past there, the money lost to a bad split exceeds what the tool costs. An app that runs on organic acquisition alone never needs the calculation at all.

"Of this week's 800 installs, only 310 have attribution. The rest land in organic."

Fewer than half the installs are joined to an ad, so it's too early to call any channel good or bad.

"The network reports 900 and the tool shows 520. Let's settle on the tool's number."

A proposal to treat the third-party verdict, not the seller's self-report, as the basis for payment.

"Cut the window from seven days to one and this campaign's numbers get halved."

Not a complaint that results got worse — a warning that changing the rule changes the result.

The quarter Pagedeck shipped its document app on mobile and started buying ads, the first thing Ravi did was pick one of these tools and ask Theo to install the SDK. Wiring it up took two days, and for the two weeks after that the numbers were wrong. The first-launch call sat behind the login screen, so anyone who installed the app without logging in never registered as an install.

Once that was fixed the channel table finally fit on one page, and Ravi moved budget on the strength of it. The channel with the most installs had the fewest paid conversions.

The most common misreading is that the tool reports the truth. "Third party" sounds like "objective," so the assumption is a natural one. What it actually reports is credit divided according to rules you configured — change the window or the credit model and the same ad shows a different number for the same period. Teams that miss this go looking for the marketer behind a bad month instead of the person who changed the setting.

Reading every unattributed install as organic is the other one. The column is usually labeled organic in the tool, which invites it. But that column also holds installs that did come from ads and simply couldn't be joined — the user opted out of tracking, the window had expired, or the network never passed the click along. Trust the label and you can't explain why organic falls whenever you pause the ads.