MMP Selection Criteria in the AdAttributionKit Era

Switching measurement partners used to come down to dashboard preference and how fast support answered a ticket. That test has stopped working. What separates one MMP from another in 2025 is whether the vendor has genuinely rebuilt its plumbing for AdAttributionKit, Apple's successor to SKAdNetwork, or whether it is still running a compatibility layer bolted onto an older architecture and hoping nobody looks too closely.

Why the old checklist stopped being enough

Two years ago an evaluation meant weighing dashboard usability against cost per tracked install, with some judgement about how well each vendor's fraud filters matched your traffic mix. Those factors haven't gone away. A blunter question now sits above them: does this vendor's AdAttributionKit implementation work the way its documentation claims it does. AppsFlyer and Adjust have both published or refreshed AdAttributionKit migration guidance this month, with Singular doing the same. Read that guidance closely, because the detail inside it is a signal in its own right. A guide that reads like a field manual, with specific postback timing behaviour and named edge cases, comes from a vendor that ran the migration internally against real traffic and wrote down what broke. A guide that mostly restates Apple's own documentation with lighter rewording comes from a vendor that read it.

The MMP selection criteria that actually predict a good fit

Five criteria deserve more weight in a 2025 evaluation than they carried in 2023.

• AdAttributionKit coverage and timeline: has the vendor shipped full support, or is a key feature, such as re-engagement postbacks, still sitting on a roadmap slide.
• Conversion value schema flexibility: can the platform help design a coarse and fine value schema suited to your specific monetisation model, rather than handing over a generic template built for a different genre entirely.
• Postback modelling and probabilistic backfill: how the vendor fills the gap between a delayed or missing postback and a usable daily reporting number, and how transparent it is about the modelling assumptions sitting behind that number.
• Historical SKAdNetwork data continuity: whether year-over-year comparisons survive the migration intact, or whether a switch quietly resets trend lines to zero at the worst possible moment for planning.
• Support responsiveness during a live migration: not general account management quality, but specifically how fast a vendor's team answers a broken postback ticket during the first month after a schema change ships.

A decision framework for a genuine bake-off

Running two MMPs in parallel for a month costs real money, and it is still close to the only test that surfaces a genuine difference rather than a sales-deck claim dressed up as capability. Pick one live campaign carrying roughly ten percent of spend and split its reporting across both vendors for four weeks. Compare three numbers at the end rather than five or six, because more metrics invite paralysis instead of a decision anyone will act on. The three: latency from install to postback; the variance between each vendor's modelled numbers and its directly attributed conversions; ticket resolution time on any discrepancy raised inside the test window.

Here is the shape that exercise might take, as a hypothetical. One vendor's modelled backfill diverges from the directly attributed numbers by a double-digit percentage across the first fortnight, then narrows to single digits as postback volume rises in weeks three and four. Expect that pattern going in. An early gap is no sort of disqualifying red flag until the sample has had time to settle.

Cost still matters, but it should not decide the shortlist

None of this argues that price is irrelevant. A migration-era evaluation still has to account for per-tracked-install cost and contract minimums, and a team on a modest budget cannot simply pick whichever vendor has the deepest AdAttributionKit documentation. What changes is the order of operations. Build the shortlist on the five criteria above, then negotiate price inside that shortlist, rather than starting from the cheapest option and hoping its AdAttributionKit support catches up before it starts mattering to your reporting. A cheaper MMP that misreports conversion value for two quarters during a bumpy migration costs a team far more in misallocated budget than the difference in contract price ever saved.

Check how a vendor's pricing model interacts with the new schema, too. Some MMPs price partly on event volume. The coarse and fine value structure in AdAttributionKit can change how many distinct events a team ends up tracking compared with a simpler SKAdNetwork setup, so a quote built on last year's volume may understate what the same integration costs once the new schema is fully built out. Ask the vendor to price a like-for-like AdAttributionKit setup specifically, not last year's SKAdNetwork integration with a note saying AdAttributionKit is "included".

Questions worth asking on the vendor call

Before signing anything, ask the vendor to walk through a real conversion value schema built for a game like yours, live on screen, not a generic template slide pulled out of a deck. Ask how many of their AdAttributionKit clients have paused a migration partway through and reverted, and what caused it. Ask what happens to historical SKAdNetwork reporting on the day you switch over. The vendor that answers plainly, including the parts that don't flatter them, makes a better long-term partner than the one whose answers are all confidence and no specifics.

The measurement stack is not the place to optimise for the newest feature list on a comparison chart. It is the place to optimise for whichever vendor will still give a straight answer six months into a migration that, as this month's guidance makes clear, is nowhere near finished. Treat this quarter's MMP selection criteria as a one-time checkbox exercise and you will probably be back here again when a further schema revision lands.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• AdAttributionKit re-engagement gets sharper after WWDC
• A Year-End Measurement Stack Review Before You Rebudget