SKAN 4 and AdAttributionKit conversion value schema design

Apple gives you six bits. Sixty-four states on the SKAdNetwork 4 fine conversion value, one of which is spent on nothing having happened, and most schemas I am asked to review have allocated all of them to revenue buckets they will never receive with enough precision to act on. Below roughly forty thousand euros a month in a single country on a single campaign, most source identifiers sit at crowd anonymity tier one, where the postback carries a coarse value of low, medium or high and no fine value whatsoever. Designing around a fine value you will not be sent is the common and expensive mistake, and it is where this engagement starts. The decision this makes: Which single question should the first postback answer. Not everything worth knowing, which is the brief I usually get handed, but the one answer that changes a bid tomorrow morning. On a news app with a metered paywall it is generally whether the reader hit the register wall inside forty-eight hours. On a bank it is whether the identity check completed. On a retail app it is first purchase, and rarely its value. The schema follows from that question, and things that fail to change a bid stay out of it. Schema design: I write the mapping across all three postback windows: the fine and coarse encoding for days zero to two, the coarse value for days three to seven, the coarse value for days eight to thirty-five, and the lock window rules that decide when updatePostbackConversionValue stops firing early to buy speed. Where a funnel step and a revenue band both deserve reporting, I encode them jointly rather than splitting the range, and I write down explicitly what the schema cannot distinguish, so nobody six months later reads a distinction into it that was never there. Postback plumbing: Postback copies routed to your own endpoint through the attribution report endpoint key in Info.plist, Apple's signature verified on receipt rather than trusted, raw postbacks stored before any transformation. Hierarchical source identifiers get an allocation scheme in writing, because the two, three and four digit tiers become available at different anonymity levels and an unplanned allocation means the campaign dimension you cared about is the one that gets truncated. AdAttributionKit: Re-engagement postbacks, and distribution through alternative marketplaces in the EU, which matters if you sell in Spain outside the App Store. It runs alongside SKAdNetwork rather than replacing it, and the developer testing mode is the only way to validate a schema before spending against it. I use it for exactly that: replaying the schema against synthetic conversions before a single euro moves. What is out of scope: I do not model SKAN results into lifetime value with an extrapolation nobody can audit. I will not promise that SKAN totals reconcile with your MMP-attributed numbers, because they are counting different things under different rules and a forced reconciliation hides that. I do not build the dashboard. Who should not book this: Below about twenty-five thousand euros a month on iOS, a single geo holdout will teach you more than any schema will, and it costs less. If the intent is to hand the schema to a network so its bidder can optimise against it with no incrementality plan behind it, we would disagree in week one and you should save the fee. Working language: English or Spanish, with your iOS engineers directly.