The installs were always there. We just couldn’t see them.

Adopting Google’s privacy-preserving measurement multiplied the installs our reporting could attribute by six. Nothing about our media changed that week — which is the whole point.

One of the more disorienting things about growth on iOS since 2021 is that you can be doing well and unable to prove it. The installs happen. The revenue arrives. The attribution layer between those two facts loses enough of the connection that the channel which earned the credit often cannot claim it.

That is not a reporting inconvenience. Budgets follow attribution. A channel that cannot demonstrate its contribution gets defunded regardless of what it actually did, which means measurement error does not just distort the dashboard — it misallocates the money.

Why we volunteered to be early

When Google began rolling out on-device conversion measurement and integrated conversion measurement for iOS App campaigns, we put our hand up early. Both are attempts at the same trick: recover the attribution signal without the underlying user data leaving the device. Event data is de-identified and matched on the handset, so no user-identifying information is disclosed externally — not to a measurement partner, and not to Google.

Being early on measurement infrastructure is a real cost. You integrate against something that is still moving, you carry the debugging, and you get no product launch out of it. We took it because our whole model depends on reading results correctly. We ship a lot, we kill a lot, and both decisions are made from numbers.

What the numbers say, and what they do not

more installs attributed in AAP reporting

6×

lower reported cost per install

80%

where the measurement gap is widest

iOS

These need stating precisely, because the honest version is less exciting and much more useful than the headline. Our installs did not increase sixfold, and our cost per install did not fall by eighty percent. Both figures are deltas in App Attribution Partner reporting: the same activity, more of it correctly credited, and therefore a truer cost per install than the one we had been managing against.

Anyone who reads them as a performance gain has learned the wrong lesson. The right one is that we had been under-measuring this channel substantially, and every budget decision made from that view was skewed by an amount we could not previously quantify.

What followed was ordinary and consequential: with a defensible read on cost, we could raise spend where the returns were real and stop crediting conversions to the last click that happened to catch them. Value-based bidding gets more useful in the same move — target return on ad spend only works on iOS if the values reaching the bidder are close to true.

Measurement is not reporting

Reporting is what you show people after the decision. Measurement is what the decision is made from. Treating the two as the same thing is how a growth team ends up with beautiful dashboards and a misallocated budget.

Privacy-preserving attribution gets discussed as a compliance obligation — the thing you adopt because the old approach stopped being available. That framing undersells it. In our case it recovered a view of a channel we had been systematically underestimating, and the value was not in the reporting getting prettier. It was in being able to spend confidently against something we could finally see.