The premise
Two facts shape every attribution decision in 2026: (1) most platforms can no longer reliably observe a single user across sessions or devices via cookies, and (2) the algorithms that drive your media spend still need a conversion signal to optimise against. The attribution problem reduces to: how do you give the algorithm a clean conversion signal when the browser has stopped helping you collect one?
The answer is not one technique. It is a stack of three layered approaches: deterministic where you can, modelled where you cannot, and reconciled at the BI layer where neither alone is enough.
Layer 1 — Deterministic, where you control the user
The strongest attribution signal is deterministic: a unique click ID generated when the user clicks the ad, carried in the URL through to your landing page, captured at registration, and posted back to the buying platform on conversion. No cookie required, no probabilistic match. Either the click ID is in your platform's user record or it is not.
For iGaming this is unambiguously the dominant attribution layer. Your platform (SoftSwiss, Everymatrix, or whatever) controls the user record; the click ID lives in that record from registration onward; conversions fire postbacks with that click ID for the entire user lifetime. We covered the implementation in the CAPI & postback piece; the attribution consequence is that 80–90% of FTDs at scale can be deterministically attributed without any cookie, IDFA, or browser identifier.
The only operators who do not have this are the ones who never properly wired postbacks across all channels. Fixing it is a 14-day engineering project; not fixing it is an indefinite tax on every channel decision.
Layer 2 — Server-side hand-off, for the platform's algorithm
Deterministic click ID solves your attribution. It does not solve the buying algorithm's attribution. Meta's algorithm needs to know that the user it served the ad to is the user who deposited, so it can find more like them. Without that hand-off, the algorithm is steering against the registration event (which Pixel still captures) but missing the deposit event — and your CPA-on-FTD looks fine in your BI but feels expensive in the ads manager.
Conversion API on Meta, Enhanced Conversions on Google, Events API on TikTok — these are the second layer. They take the deterministic conversion you captured at the platform and hand it back to the buying surface as a server-side event, with hashed PII, click ID match keys, and the actual conversion value. CAPI match quality of 8.5+/10 is the operating threshold; below that, the algorithm is flying blind.
Layer 3 — Modelled, for the residual
The remaining 10–20% of conversions — users who deposited via a direct visit, a saved bookmark, or an offline channel — cannot be deterministically attributed back to a single click. This is where most operators either give up (assigning all of it to "direct") or over-assign (last-click attribution to whatever channel happened to fire the latest pixel).
The honest model is multi-touch attribution at the cohort level, with two specific tools:
- Markov-chain attribution on the deterministic 80%, then projected to the residual — gives you channel weights based on path-to-conversion influence rather than last-touch.
- Geo-experiment lift testing — turn one channel off in one market for 14 days, measure the FTD baseline shift, attribute the difference. Gold standard for incrementality.
Most operators do not run either of these because they require BI engineering and discipline to interpret. The ones who do consistently rebuild media plans every quarter — and consistently scale faster than peers.
The two attribution windows that matter
Forget last-click. The two windows that matter for iGaming unit economics are:
- D7-window CPA — cost per FTD where the deposit happens within 7 days of click. Tracks acquisition velocity. The metric the algorithm should be optimised toward.
- D90 LTV-to-CAC ratio — D90 player value divided by fully-loaded acquisition cost. Tracks unit-economic health. The metric the CFO should be optimised toward.
Operators who manage by D1 or D7 LTV alone over-spend in markets with strong early-deposit behaviour but weak retention. Operators who manage by D90 LTV alone under-react to channel-level CPA drift. You need both visible in the same dashboard, refreshed weekly.
Common attribution traps
Pixel-only counts
Reporting media performance based on what Meta Pixel saw alone, with iOS share over 30% in your market. You are missing 25–40% of conversions. Same for app pixel under ATT.
Last-click in mixed channels
If a user clicks a Taboola native ad, then clicks a Google brand search 3 days later, then deposits — last-click attribution gives Google credit for an FTD that Taboola generated. Brand search budget grows; native budget shrinks; CPA blends look fine; new-user acquisition silently dies.
Self-reporting bias
"How did you hear about us?" surveys at registration are unreliable. Users do not remember and self-select. Useful for qualitative segmentation, not for budget decisions.
Affiliate over-attribution
Affiliate platforms typically use 30-day cookie windows that overlap aggressively with paid media windows. The same player can be claimed by paid media and an affiliate. Without click ID priority logic at the platform layer, you double-pay on the same FTD.
What good attribution looks like at scale
For a €500K+/mo operator, the attribution health-check we run looks like this:
| Check | Threshold |
|---|---|
| Click ID coverage on FTD | ≥ 88% |
| CAPI match quality (Meta) | ≥ 8.5 |
| Postback → BI reconciliation gap | < 3% |
| Channel-level incrementality measured | 1× per quarter, top channels |
| D7 CPA / D90 LTV in same dashboard | Yes, weekly refresh |
If even three of those are not in place, attribution is the binding constraint on your scaling — not creative, not budget, not channel mix. We come back to this point a lot because operators consistently underestimate it.
The 60-minute test
Ask your team for one report this week: monthly FTDs, broken down by paid media, affiliate, organic, and direct, with overlap claims explicitly resolved. If your team cannot produce that report in 60 minutes from existing systems, you do not have an attribution stack — you have attribution opinions. The fix is straightforward but not quick. It is the one tracking project we tell every operator to start before any campaign optimisation.