Your measurement stack exists.Something in it is lying to you.

Attribution model differences account for some of the gap. Different windows, different view-through rules. The first thing I do is quantify what's explainable by model differences versus what's actual misconfiguration. In most audits the gap is far larger than model differences can account for, and you'll see exactly how much is which before any rebuild starts.

The audit phase is entirely passive. I don't touch a live system to diagnose it. I only change things in the rebuild phase, after you've signed off the plan, and it's sequenced to minimise disruption. SKAN schema changes take effect going forward only, so historical campaigns aren't affected.

I'll usually recommend fixing the instrumentation first, then migrating if you still want to. Switching MMPs with a broken stack just moves the problem to a new tool. The event taxonomy and tracking plan I rebuild are designed to be MMP-agnostic wherever possible, so the hard work carries over if you do migrate.

That's the normal case, not the exception. The most expensive attribution problems never throw an error. Bridebook came to me for a routine second-opinion audit on a $100K+/month program with nothing alarming on any dashboard. The audit found five separate layers quietly wrong: SKAN returning no data at all, a fake 100% ATT consent rate, view-through, re-engagement, and reinstall attribution silently disabled, revenue not mapped to any partner, and deep-funnel events never reaching the ad platforms. Every one rendered a plausible number. I check what each layer should be emitting against what it actually is, so the gaps that look coherent enough to never get questioned are exactly the ones that surface.

The common failure is several events all independently reporting the same revenue. In a consumer-app review before their first campaign, a code review found seven separate events each flagging revenue back to AppsFlyer. When several fired on one transaction, a single sale got counted multiple times, and the inflation looked exactly like real performance. The second half is SKAN revenue buckets set so wide that most purchases collapse into one range, so the ad platform can't tell a high-value buyer from a low-value one. I trace every event that reports revenue in the codebase, consolidate to one clean source of truth, and rebuild the SKAN ranges around your actual purchase distribution.

When install counts look fine but revenue won't tie out, the leak is usually silent and server-side: purchase postbacks being sent and rejected before they ever reach the MMP, with no error surfaced on any dashboard. A missing or wrong auth header returns a 401, the postback is dropped, and the numbers just look a little low. I diagnose it in the postback logs rather than the dashboard, fix the routing, and validate that events land with a 200 before calling it done. I also check for revenue that isn't being mapped back to partners at all, which produces the same symptom from a different cause.

A fixed stack. The audit produces a prioritized list of what's broken and by how much, but the engagement doesn't end there. The rebuild phase corrects each gap the audit found, in a sequence designed to minimize disruption to live campaigns, and every fix is validated before handoff. You get a working, documented measurement layer, not a deck describing what's wrong.

Scope depends on how many channels you're running, which MMP you're on, and how many layers turn out to need work, which the audit determines. Email us your requirements and I will share the detail proposal with scope and timelines