Every operator we audit tells us the same thing in the first call: “our attribution is mostly fine, we just want a second opinion.” It’s almost never fine. And the parts that are broken are making seven-figure budget decisions look confident.

Three places attribution breaks, in order of how often we see them:

1. The postback.

Postback flows are almost always “working” in the sense that conversions are coming through. What they’re not doing is reconciling against the source of truth. We’ve seen operators where 18% of paid conversions were firing against a stale gclid because the token got dropped on a redirect two years ago and nobody noticed. That 18% is silently reallocating budget every week to a channel that deserves less of it.

The fix is not complex. It’s a weekly reconciliation between what the platform reports, what the postback fired, and what the backend recorded. Three numbers. If they disagree by more than 3%, something is broken. Nobody runs this check because it doesn’t have a dashboard by default.

2. The lookback.

Default lookback windows are tuned for e-commerce. Your product has a 30-60 day payback and a 90-day retention tail. Running 7-day attribution on a 45-day decision is measuring noise. The platform will happily serve you beautifully wrong data forever.

iGaming is worse than most verticals on this. First-deposit-to-second-deposit can take two to three weeks. Retention cohorts compound after day 30. A conversion measured on day 7 and a conversion measured on day 45 are fundamentally different signals. Treating them the same in your ROAS view is the reason “winners” turn out to be losers a month later.

If three different reports give you three different numbers for the same channel, you don’t have an attribution problem. You have a measurement architecture problem.

3. The reconciliation.

Most operators have three sources of truth for the same conversion: the ad platform, the MMP, and the backend. None of them agree. The team spends Monday morning arguing about which one is right instead of making decisions. The fix is not another dashboard. It’s a rule: which source wins, in which case, and what the reconciliation cadence is.

That rule gets written once, gets agreed by paid, CRM, and finance, and then the Monday meeting is about what to do. not which number is right. That shift alone is worth more than most attribution tools.

What to do this week

Pick one channel. Pull the platform number, the MMP number, and the backend number for the same seven-day window. If they disagree by more than 5%, you have work to do. If they agree, check your lookback windows against your actual payback period. If those don’t match either. and they usually don’t. every budget decision of the last quarter was at least partially fiction.

You don’t need a new tool. You need a rule, a reconciliation cadence, and a decision about which source wins. The tool is already in the stack.