The studio that found 18 percent of installs were misattributed

How you discover the numbers are quietly wrong

Attribution feels like a solved problem. You buy installs from a network, a tool tracks them, a dashboard shows you the result. But most indie developers never dig deeper. They assume the numbers are real because the system looks professional.

That Berlin studio was different. The founder, Sarah, had built mobile games for twelve years. She knew the shape of real user behaviour. When Attribr showed her day 7 retention at 22 percent and day 30 at 11 percent, she cross-referenced it against her Mixpanel export. The numbers matched.

Then she looked at the source breakdown. Attribr said 62 percent of her installs came from paid ASA campaigns. Her ad account said 54 percent. The difference wasn't rounding error. It was nearly seven thousand installs, and she'd been optimising her budget based on inflated numbers.

The real story: probabilistic attribution (the kind most tools rely on) casts a wide net. It catches a lot of legitimate installs. But it also catches noise. A user on WiFi near your office building, a marketer testing ads from their home, someone who searched for your app and saw an ad five minutes later. All of those look like they came from the paid campaign if you're only matching device signals and timing windows.

Why deterministic matching actually matters

Attribr works differently. We ask a simple question: did this install come directly from a click on a link we can verify? If yes, that's deterministic. If we're not sure, we apply probabilistic matching to fill the gap, but only as a second pass.

Most tools skip the first part. They go straight to probabilistic because it's fast at scale and it inflates the numbers your stakeholders want to see. If your paid channel gets credit for half of organic installs, your CPI looks better, your ROI curves go up, and the tool vendor looks good.

Sarah's 18 percent finding wasn't unique. We've seen it across studios. The number varies depending on your ad networks and how tightly they integrate. But the pattern is always the same: once you separate deterministic from probabilistic, you see how much of your paid channel attribution is actually guesswork.

The reason Attribr caught it is mechanical. The SDK is fifty kilobytes. No third-party dependencies. It runs in under fifty milliseconds at launch. That means we can afford to do the matching locally, on the device, without calling a server. You get your source attribution from the app itself, not from log-file inference. It's slower to build. But it's accurate.

What Sarah did with the real numbers

Once she knew which installs were actually from paid channels, Sarah rebuilt her budget allocation. She'd been spending forty percent of her ASA budget on a creative that Attribr said was driving organic installs at three times the rate of paid. She cut that spend by half and reallocated to the campaigns with real paid traction.

Three weeks later, her CPI dropped fourteen percent. Her day 7 retention on paid installs went from 18 percent to 24 percent. Not because the installs got better, but because she was now comparing like for like. She wasn't accidentally optimising for volume that included organic noise.

That's the thing nobody tells you about attribution: fixing it doesn't make your app better. It makes your decisions better. When you know which installs are actually paid versus organic, you stop burning budget on phantom conversions. You stop running the same creative because the dashboard made it look good. You see what's real.

Sarah also integrated Attribr with Rippl, our performance marketing platform. One of her games has a small community of streamers who run CPI campaigns. Before Attribr, she couldn't see which streamer was driving installs. Now she does. She's paying the best performers forty percent more per install, and they're getting better results because they're incentivised to push her real users, not rinse-and-repeat followers.

The eighteen percent isn't a bug in Attribr

It's worth being clear about this. The misattribution Sarah found wasn't a flaw in our system. It was the water all indie developers swim in. Most attribution tools are built to maximise reach and revenue for ad networks. They're incentivised to give credit generously. A little noise in your channel attribution is treated as a feature, not a bug.

Attribr exists because we thought that was backwards. If you're an indie developer with a £500 monthly budget, you don't need enterprise-grade fraud detection (though we offer that in the Pro plan). You need to know where your installs actually come from. You need to know if your users are still around at day 7 and day 30. And if you're running CPI with Rippl promoters, you need to know which one of them is actually moving the needle.

The SDK integrates in three lines of code. Swift or Kotlin. You get a dashboard that shows you cohorts, funnels, and retention curves. You can bring your own ad network if you want to. And if you hit twenty-five thousand monthly installs, you upgrade to Growth for twenty-nine pounds. At a hundred thousand, you move to Pro and get fraud signals and the ad-network roll-up.

But the core thing doesn't change. You're getting deterministic attribution first, probabilistic second. You're not paying for a tool that's gaming the numbers to make your channels look good.

Why this matters more than you think

Sarah sent me a follow-up message three months after she started using Attribr. She'd launched two new games. With the first, she'd applied everything she learned from the eighteen percent discovery. Her CAC was measurably lower. With the second, she hadn't used Attribr because she figured she'd learned enough. Her CAC was twenty-eight percent higher, and she only realised why after two weeks of painful optimisation.

That's the sneaky part of misattribution. It doesn't just give you wrong numbers. It teaches you to make wrong decisions. If you spend months optimising channels based on inflated numbers, you build intuition around those false signals. When you move to a new project without accurate attribution, you're flying blind and you don't even know it.

Most indie studios run three or four games at once. Your time is split across creative, design, live ops, and business. You don't have someone dedicated to analytics. So you rely on your tools to tell you the truth. If your tools are guessing, you're guessing with a veneer of confidence, which is worse than guessing openly.