32 milliseconds: why performance matters more than you think
A developer emailed us last month. Their indie puzzle game had 40,000 monthly installs, and they'd just switched to Attribr from a competing attribution SDK. Within a week, they'd noticed something: their app launch felt snappier. Not dramatically different. Just noticeably. They asked us what we'd done. The honest answer was that we'd obsessed over every microsecond from day one.
The problem nobody talks about
Most attribution SDKs add 50ms to 150ms to your app's launch time. You don't notice it once. But multiply that across a million users opening your app three times a day, and you're looking at real friction. A 100ms delay on launch correlates with higher churn, lower retention, and fewer sessions. For indie developers already competing against games with massive budgets, every millisecond counts.
When we set out to build Attribr, we decided early: we would not be that tax. We'd be sub-50ms or we wouldn't ship. That wasn't a marketing promise. It was a hard constraint we built into our architecture from the first line of code.
Why 50KB matters when you're fighting for attention
The smallest SDKs are usually the slowest. They're small because they do nothing. We did the opposite. We kept the payload to 50KB and zero third-party dependencies, then packed in deterministic and probabilistic matching, retention cohort tracking, and a direct bridge to Rippl promoters. That required deliberate choices about what to include and what to leave out.
No network calls on the critical path. No initialisation overhead. No dependency bloat. We process attribution data asynchronously, offline when we can, and we batch server communication. The result is that your app launches, your attribution fires in the background, and your user sees zero difference. That's the opposite of how most SDKs work.
The 32 millisecond moment
We spent months chasing that figure. We profiled on real devices. Pixel 7, iPhone 14, OnePlus, older Android phones people actually use. We measured initialisation time, event logging, local matching against our deterministic database. We removed a reflection call that saved 8ms. We rewrote a sorting routine that saved 12ms. We moved a network request off the main thread that saved another 6ms.
At some point, we hit 32ms on an iPhone 13 and 38ms on a Pixel 6. We stopped there. Not because we couldn't squeeze more. Because 32ms is invisible to the user. Below that threshold, you're chasing diminishing returns and burning engineering time that could go elsewhere.
What indie developers actually need to know
Here's the thing: you don't need to care how we got to 32ms. You need to care that we did it so you didn't have to. You ship your app knowing that attribution won't tank your metrics. Your retention cohorts will be accurate. You'll know where each install came from. On day 7, day 14, and day 30, you'll see which users stuck around.
And if you're running campaigns through Rippl, you'll see exactly which community promoters drove installs that actually retained. That's the Rippl bridge, and it's something no other attribution SDK offers. It closes the loop between performance marketing and retention data.
Performance isn't a feature. It's a promise.
We could have padded the spec sheet with a dozen features we don't offer. We could have claimed we do fraud detection like Adjust, or network-wide attribution like AppsFlyer. We don't. We do one thing for indie developers and small studios: we tell you where your installs came from, whether those users are still active, and we do it in less than 50 milliseconds without drowning you in enterprise pricing.
That developer with the puzzle game? They're still using us. Their launch time is stable. Their cohort data is clean. They know what's working. That matters more than any feature list.
If your app is launching slower than it should be, and you're paying for attribution data that might not even be worth the overhead, have you looked at what sub-50ms actually feels like?
Ready to try Attribr by MRVL?
One tap to download. No sign-up wall.