SKAdNetwork

What SKAdNetwork is

SKAdNetwork (SKAN) is Apple's privacy-preserving framework for measuring iOS app-install advertising. Instead of the user-level trail traditional mobile attribution reconstructs — this device clicked this ad, installed, then purchased — SKAN has the device itself report a conversion through Apple, aggregated, delayed, and stripped of anything that could identify the user. The flow: ad network signs an ad → App Store install → the device sends a postback (after timers and thresholds) → network and advertiser receive aggregate campaign results.

It exists because App Tracking Transparency (ATT) made IDFA-based tracking consent-gated; SKAN is the measurement Apple offers for everyone who doesn't consent — in practice, the majority of iOS installs.

What a postback contains, and what it deliberately doesn't:

• Campaign-level identifiers, not user-level. SKAN 4 provides a hierarchical source identifier (2–4 digits, granularity gated by volume) plus coarse source app info. No device ID, no click timestamps, no user trail.
• A conversion value. SKAN 4's model: a fine value (0–63) or coarse value (low/medium/high) in the first postback window, coarse-only in the second and third. This small number is the only carrier of post-install quality information — encoding it well is the central design problem.
• Privacy thresholds. Below Apple's (undisclosed) crowd-anonymity thresholds, fields degrade or vanish: small campaigns may receive postbacks with null conversion values or no postback detail at all.
• Delays by design. Postbacks arrive after randomized timers — roughly 1–2 days for the first window, with windows two and three (SKAN 4) extending measurement to 7 and 35 days, each arriving days after its window closes.

What this does to UA practice

Every assumption from the IDFA era breaks somewhere:

• No user-level joins. SKAN data cannot connect to your MMP user records, CRM, or LTV tables. Channel evaluation becomes statistics on aggregates, not queries on users.
• Optimization signals are slow and coarse. Bidding algorithms that fed on real-time conversion events now learn from delayed, thresholded postbacks — one reason iOS campaign learning is slower and more volatile than Android's.
• Campaign structure is constrained. Identifier granularity is earned by volume; fragmenting iOS spend across many small campaigns buys you null postbacks. Consolidation is the standard adaptation.
• CPI and ROAS get error bars. Modeled and thresholded data means channel numbers are estimates with confidence bands. Decisions need longer windows and bigger effect sizes to be trustworthy.

MMPs (AppsFlyer, Adjust, Branch) remain the practical operating layer: they manage conversion-value schemas, collect and decode postbacks, reconcile SKAN with consent-based ATT attribution where it exists, and model the gaps into one report.

Designing conversion values well

The 6-bit fine value is scarce — spend it on business priorities, not convenience:

1. Decide the one question that matters per app. Usually: "does this cohort pay back?" Encode the earliest events that predict payback — trial start, first purchase, D1/D3 retention milestones — not every product event you happen to track.
2. Use revenue buckets where monetization is continuous. Map value ranges to fine values so early revenue per cohort is recoverable within postback constraints.
3. Match the measurement windows. SKAN 4's three windows (0–2, 3–7, 8–35 days) reward schemas that capture early signal in window one and trajectory in windows two and three — coarse values included.
4. Keep schemas stable. Every change breaks comparability across the boundary; version them deliberately and annotate dashboards.
5. Back-test against consented users. The ATT-consented slice provides user-level ground truth to validate what your conversion values actually predict.

Common mistakes

• Expecting user-level attribution. Building reports that assume joinable user trails guarantees disappointment; the framework is aggregate by design, and modeling cannot fully reverse that.
• Ignoring postback delays. Judging an iOS campaign 48 hours in — before most first postbacks arrive — reads noise. Evaluation calendars must respect SKAN's clock.
• Designing conversion values without business priorities. Default or engagement-flavored schemas (sessions, screens viewed) measure activity, not value, and waste the only quality signal available.
• Fragmenting campaigns below thresholds. Many small campaigns produce mostly-null data; fewer, larger campaigns produce usable data.
• Treating SKAN, ATT, and modeled numbers as one series. They measure overlapping but different populations with different methods; label sources or watch teams reconcile irreconcilable dashboards.

FAQ

Does SKAdNetwork replace my MMP? No — it changes what the MMP does. The MMP operates SKAN (schemas, postback collection, decoding), blends it with consent-based attribution and modeling, and remains the cross-channel referee.

What changed in SKAN 4 versus 3? Three postback windows instead of one (extending measurement to 35 days), hierarchical source identifiers up to 4 digits under volume conditions, coarse conversion values as a threshold fallback, and web-to-app ads becoming measurable. Adoption ramped through 2023–2024; expect mixed-version reporting in older data.

How do privacy thresholds actually behave? Apple doesn't publish the numbers. Observable behavior: low-volume campaigns receive postbacks with masked fields or coarse-only values. The practical rule — consolidate until your postbacks consistently carry the detail you designed for.

Can I still get user-level iOS attribution? Only for users who consent through ATT on both the publisher and advertiser side, which is a minority slice. That slice is valuable as calibration ground truth, but it is not representative of the whole and shouldn't be scaled up naively.

What is AdAttributionKit? AdAttributionKit is Apple's successor framework extending SKAN's model — re-engagement attribution and alternative-marketplace support among the additions, with SKAN compatibility during transition. The aggregate, privacy-thresholded paradigm continues; teams should design measurement for that paradigm rather than waiting for user-level data to return. The app user acquisition path covers the full iOS measurement stack.