Game marketing measurement: what each source tells you, and what it leaves out

Nov 18, 2026|5 Min
Game marketing measurement

You have a budget decision due today, and no two of your install sources agree.

Your Mobile Measurement Partner (MMP) says a network drove 1,200 installs last week. The network’s dashboard says 1,500. Your SKAdNetwork postbacks return a partial, delayed, aggregated figure that matches neither.

All of these sources answer different questions in systems that are siloed and disconnected. Your goal is knowing which question each source answers, where each goes blind, and what combination lets you defend a decision. That’s where game marketing measurement comes into play.

What is game marketing measurement?

Game marketing measurement is the set of methods studios use to work out which spend drives installs, revenue, and player value. Example methods include attribution, incrementality testing, media mix modeling, and cohort or Lifetime Value (LTV) analysis.

In this article we’ll explain how game marketing measurement works so you can understand what each source delivers, and then what to fix in your own setup so the systems optimizing on your behalf are working from the whole picture rather than a fragment of it.

No two of your install sources agree

Consent-gated identity, aggregated and delayed platform reporting, and different crediting rules all contribute to inconsistent tracking. The table below explains how:

Why don't your sources agree?

Learn common reasons why tracking can appear inconsistent

Structural cause

Consent-gated identity

What’s happening

Identifiers are permission-based now, in different ways on each platform. Apple asks players for tracking permission, so you only get an IDFA, the unique identifier given by Apple to each device, from the ones who say yes. Android works the other way around: the advertising ID is available by default unless a player deletes it, after which apps requesting it receive a string of zeros. Apps targeting Android 13 or above must also declare the AD_ID permission to access it.
Why it causes inconsistency

When a system can’t identify someone, it estimates instead. Each system estimates its own way, so they land on different answers for the same week.

Aggregated and delayed platform reporting

What’s happening

Apple’s SKAdNetwork reports late and in rounded form. Once a conversion window closes, it can take one to two days for the first postback and up to six for later ones, and it sends a conversion value you defined in advance rather than real revenue. It also assigns crowd anonymity tiers, so if a campaign’s installs are too few to anonymize, Apple may limit or withhold the detail
Why it causes inconsistency
Your own analytics counted those installs. There’s nothing to reconcile, because the details were never sent.

Different crediting rules

What’s happening

Every network and mobile measurement partner (MMP) decides for itself how long after a click or a view it can still claim an install, and which installs to throw out as fraud. Those are the attribution windows and filters, and they’re rarely the same from one setup to the next.
Why it causes inconsistency

A network with a seven-day window claims an install your MMP already credited somewhere else. Each is applying its own rules correctly, which is why both totals look right to the people reporting them.

Mobile attribution 101

Marketing attribution assigns credit for a conversion to the marketing touchpoints that caused it. For mobile attribution, that means matching an install or in-app event back to a specific click, impression, campaign, and creative, so you can compare cost per outcome across sources and decide where the next dollar goes.

What are MMPs?

A mobile measurement partner (MMP) is a neutral third party that receives install and event data from your app’s SDK, applies one consistent set of attribution rules across every network you buy, and sends postbacks back to those networks so their optimization models learn from your outcomes. It also handles deep linking, ad fraud detection, and one reporting layer. Where a shared identifier exists, it can match an install to a click, which is deterministic matching. Where one doesn’t, it infers the match from signals like IP address and timestamp, which is probabilistic and reliable in aggregate but not per player.

What is multi-touch attribution?

Multi-touch attribution (MTA) spreads credit across the touchpoints that preceded a conversion instead of giving it all to the last click, which requires a reasonably complete view of what one person saw. For example, if someone sees a social media ad, reads a blog post, opens an email, then clicks a paid search ad and signs up. Single-touch models give 100% of the credit to the first or last of those. MTA splits it across all four.

With the importance of data privacy, MTA increasingly relies on data from consenting users and the first-party events you collect yourself, subject to applicable law and platform requirements.

What incrementality testing tells you, and what it costs

Incrementality testing shows how many installs your ads actually caused. To test it, you withhold ads from one group of players, compare them against a similar group that saw the ads, and the difference between the two is your incremental lift.

The catch is that you pay for the answer in lost revenue. Because you’re not advertising to people who would have installed, you give up what they would have spent. That trade is only worth it on a big channel, where the answer could impact real budget decisions.

Understanding media mix modeling

Media mix modeling, also called marketing mix modeling, uses statistical regression to estimate how much each channel contributed to aggregate outcomes like installs or revenue. It answers two questions well: where the budget should go, and where a channel stops paying back.

How the three fit together

The mature setup is a set of methods that cross-check each other, running on different clocks, each covering another’s blind spot.

Mobile attribution data

Understand what each data type shows and where it falls short

Method

Attribution (including MMP, SKAN, AdAttributionKit)

Question it answers

Which campaign, creative, and source are my users coming from?

What it can’t see

Whether the conversion would have happened anyway

Incrementality testing

Question it answers

How much of this outcome did the spend actually cause?

What it can’t see

Anything outside the channel and window you tested
Media mix modeling

Question it answers

How should I split budget across channels, and where does each saturate?

What it can’t see

Individual users, creatives, and anything without spend variation

Where your own player data fits

All three methods above read outcomes. But the data they read is the layer you own outright, and it’s where most studios have the most headroom.

Studios today have a data fragmentation problem, meaning their data lives across many systems which do not connect. For example, MMPs help connect acquisition activity to installs and conversions, analytics platforms provide insights into player behavior, and purchase systems capture transaction-related information. For Unity-made games, eligible Runtime data may offer additional context from within the game experience itself.

These sources can overlap, but each may contribute different levels of detail, coverage, and context. The challenge is ensuring that relevant, reliable signals are available to the systems responsible for measurement and optimization, subject to the appropriate permissions and technical requirements.

To sum it all up, each source is useful, but none of them alone describes your whole player and all of their actions. Due to this fragmentation, decisions get made with insights that are siloed and difficult to connect.

As a result, optimization models such as Unity Vector may learn from an incomplete picture and act on only part of the story.

So how can you combat this?

Identify valuable signals your optimization models may be missing

Review which signals your existing integrations provide and where meaningful gaps remain. Some engagement, purchase, or game-related signals may offer additional context, but their availability and usefulness depend on how they're collected, configured, and shared.

Start by identifying which signals are relevant to your optimization goals, then evaluate whether an existing integration can provide them or whether an additional data source is needed.

Send outcomes regardless of where they came from

It’s tempting to pass only the events your MMP credited to a paid source. But organic and unattributed players are real outcomes too, and if Unity Vector never sees them, it may be learning from a filtered sample of your best information. Share these signals only where permitted by applicable law and player consent.

Give Unity Vector the full lifecycle

Set your postback window to lifetime (if available) so the cohorts that pay back slowest and largest show up, not just the players who converted in their first week. Those late payers are often where your lifetime value actually sits.

Treat sharing permissions as a performance lever

The models bidding on your behalf can only learn from what you let them see, so what you share can impact performance. It’s easy to accept whatever was configured at integration and never revisit it. But instead, go back through those settings and make sure you have all the required signals set up, consistent with applicable law, your privacy disclosures, and player consent. A model with more complete, high-quality signals may be better equipped to recognize patterns among your most valuable players.

Learn which signals can give Unity Vector the extra edge

Getting the complete picture

None of your numbers are the truth, and chasing a single source of truth is how measurement projects stall. The goal is narrower and more achievable: a setup whose disagreements you can explain. When your MMP and your network diverge, you should know why, by how much, and which one to use for the decision in front of you.

So start from the gap rather than the tool. Name the one thing your current setup is structurally unable to see, then add the most practical method that can see it. The other half is what you feed back out. Better signal coverage can support better-informed optimization, which is where a user acquisition platform earns its place in the stack rather than just sitting in it.

Diversify your signal coverage

Frequently asked questions

*Source: Google, “Advertising ID,” Google Play Console Help (accessed November 16, 2026; subject to change by Google at any time), https://support.google.com/googleplay/android-developer/answer/6048248

**Source: Apple, “Receiving postbacks in multiple conversion windows,” Apple Developer Documentation (accessed November 16, 2026; reflects SKAdNetwork 4 and later; subject to change by Apple at any time), https://developer.apple.com/documentation/storekit/receiving-postbacks-in-multiple-conversion-windows

Feature image: “Three overlapping circles for attribution, incrementality and media mix modeling, with one shared area where the sources agree” Image generated by Anthropic, Claude (Claude Sonnet 5.5), Date: November 18, 2026