There are two ways to end up with a measurement setup that works on iOS today. You can take a system built for the device-ID years and keep adapting it as signals disappear. Or you can start after the signals were already gone and ask a simpler question: with what we can legitimately know now, what should measurement actually be for?
Most of the industry has had to do the first. That’s hard, useful work, and a lot of smart people have done it well. This article is about the second path, because the starting point quietly shapes every decision that follows: what you count, what you optimise for and what you show the person reading the report.
We built Audiencelab on that second path, so we’ll be upfront that we have a view here. But the design choices below are worth thinking through, whichever tools you use.
TL;DR
Measurement built before App Tracking Transparency was designed around knowing who a device was. Measurement built after it has to be designed around what you can know without that.
Starting from scratch leads to four design choices: measure the creative, not the device; define value by in-game behaviour, not the install; engineer the signals you send back to networks; and show the working, including the uncertainty.
None of this replaces incrementality testing. Attribution tells you where credit goes. It doesn’t tell you what would have happened without the ad.
Adapting vs designing from scratch: why the starting point matters
Before 2021, a lot of mobile attribution rested on a comfortable assumption: if you could recognise the device, you could follow the user from ad to install to purchase. When device-level matching wasn’t available, fingerprinting often filled the gaps. The whole stack, from reports to optimisation logic to how we talked about “the truth”, grew up around that assumption.
Then Apple’s App Tracking Transparency, enforced from iOS 14.5 in April 2021, made user consent a requirement for tracking and made it explicit that fingerprinting isn’t allowed. [LEGAL/POLICY REVIEW] Privacy regulation in Europe points the same way. The assumption didn’t weaken. It broke.
A system adapting to that change has a natural goal: get back as close as possible to what it used to do. That’s a reasonable instinct, and it explains a lot of the industry’s effort since. The catch is that every step back towards the old precision sits on a smaller and less stable pool of signal, and the target keeps moving with each OS and policy update.
A system designed after the change has a different goal. It doesn’t ask “how do we recover the device?” because it never had the device to lose. It asks “which decisions does a UA team need to make, and what evidence can support them today?” That question leads somewhere quite different.
Design choice 1: the creative is the unit of measurement, not the device
If you can’t reliably follow individual devices, the natural move is to measure something you can still see clearly: the thing you paid for. For most mobile games, that’s the creative.
This turns out to be more than a workaround. Creatives are where UA teams actually make decisions. Nobody’s Monday meeting is about device 4F2A-something. It’s about whether the merge-gameplay video beats the character-story video, and whether to brief three more like the winner.
Network and campaign-level reporting tell you which channel worked. They’re less good at telling you which ad inside that channel brought in the players you wanted. When the creative is the unit of measurement, every downstream number (retention, purchases, ad revenue) gets attached to the ad that started the journey. That’s the level where creative strategy, budget allocation and the next brief all live.
Design choice 2: in-game behaviour defines value, not the install event
An install is the moment a network’s job looks finished. For the studio, it’s the moment the real question starts.
A creative with a low CPI can bring in players who leave after one session. A creative with a higher CPI can bring in players who finish the tutorial, come back on day 7 and spend something. Judged on installs, the first one wins. Judged on players, the second one usually does.
Designing after ATT makes this choice easier, oddly enough. If you’ve accepted that you won’t see every individual path across every network, you’re pushed towards the data you do own: what players do inside your game. Tutorial completion, level progression, session frequency, D7 retention, ad views, purchases. Your game sees all of that first-hand, and nobody else’s reporting can claim more of it than actually happened.
So the question a modern report answers isn’t “which creative got the most installs?” It’s “which creative brought in players who became valuable?”
Design choice 3: the signals you send back to networks are engineered for quality
Ad platforms learn from the conversions you send them. If the only signal is an install, you’re essentially telling the algorithm “find me more people who install.” It will do exactly that, very efficiently.
A measurement layer designed for today treats those signals as something to design, not just forward. Signal engineering means choosing events that tell the platform what a valuable player looks like, while still being usable for learning. That’s a balance:
- Value correlation: the event should relate to long-term player value.
- Volume: it has to happen often enough for the platform to learn from it.
- Timeliness: it should happen early enough to be useful inside the platform’s optimisation window.
- Reliability: it should be measured the same way every time.
Further downstream isn’t automatically better. A first purchase might be the most valuable event in your game and still be too rare to train on. For example, if a puzzle game sees that players who complete level 10 on day one retain far better than those who don’t, “level 10 on day one” may teach Meta more than a raw install, and arrive far sooner than a purchase. The goal isn’t to send every event you have. It’s to send the one that teaches the algorithm the right lesson.
Design choice 4: show the working, including the uncertainty
When signal is partial, any number in a report is the result of rules: which touchpoints count, how long credit lasts, whether a view is enough, what happens when two sources claim the same install. Those rules matter as much as the number itself.
It’s worth remembering that every system reporting on its own results also sets its own rules. Nobody needs to be doing anything wrong for that to matter. It’s simply why the person reading the report should be able to see which rules produced the number, and how sure the system is about it.
A measurement layer built for this era should make that visible: what was observed directly, what was aggregated, what was estimated and where the gaps are. “We’re less sure about this slice of installs” is more useful than a confident total that hides its assumptions.
How Audiencelab applies these choices today
Here’s how those design choices show up in Audiencelab today. [CONFIRM: every item in this section with product before publishing]
Creative-level attribution on iOS. Audiencelab connects individual ad creatives with what the resulting players do in the game. Where a player has touched several creatives, the last-click creative gets credit by default, and supporting click and view signals are configurable, so the rules are yours to set rather than inherited.
Creative breakdowns. Performance can be read at creative and ad-set level, not just by network or campaign, with retention, revenue and other behavioural events attached to each creative.
Unity SDK. A Unity attribution and analytics SDK captures attribution and in-game behaviour in one place, so creative performance and player value come from the same source.
Customizable postbacks. Audiencelab can send creative-level postbacks back to networks such as Meta, TikTok and Google, so the platforms learn from player quality, not just installs. It also works with the TikTok Events API, including purchase and ad-revenue signals.
Pulsar. Pulsar is a browser-based signal engineering layer. Marketers can define conversion rules (D7 retention, tutorial or level completion, ad-view or purchase thresholds, blended ROAS) and change them without waiting for a new development cycle.
User properties. Segmenting players by IAP behaviour, ad revenue, retention and in-game context helps answer what kind of player it brought in.
What this looks like in practice varies by game. In one implementation with Talofa, using the Unity SDK with Meta and creative-level measurement, reported CPI fell from about $3.80 to about $1.80. That’s one team’s result in one setup, not a promise, but it shows what can change when the platform learns from better signals.
Audiencelab is designed to sit alongside your existing measurement, including your MMP, not to replace it. Think of it as a creative-level lens on the data you already have.
What this doesn’t solve: incrementality still needs its own tests
It would be easy to end on a high note here, so let’s not.
Attribution, however modern, assigns credit. It answers “which creative gets credit for this player?” It does not answer “would this player have come anyway?” A creative can bring in great players who were already on their way to your game through organic search or word of mouth.
That second question needs its own tools: geo holdouts, conversion lift studies, on/off tests with a sensible control. Creative-level attribution makes those tests more useful, because you know what you’re testing and what good players look like. It doesn’t replace them.
Conclusion: design for the era you’re in
The fingerprint isn’t coming back, and chasing it is a tiring way to spend a measurement budget. The more useful move is to change the question. Stop asking “who exactly was this user?” and start asking “which creative brought in players worth having, and how do we teach the networks to find more of them?”
Those are questions you can answer well today, with data you legitimately have: your creatives, your players’ behaviour and well-chosen signals. Whether you build that yourself, use Audiencelab or combine several tools, the principles are the same. Measure the creative, value the player, engineer the signal, and keep running the incrementality tests.
Want to get started with Audiencelab? Let’s talk!