Crowd Anonymity

Definition

Crowd anonymity is Apple's privacy mechanism that decides how much campaign and conversion detail an attribution postback may contain, based on whether the install cohort is large enough that the report cannot identify an individual.

It is the reason two campaigns with identical instrumentation return different amounts of data, and the reason most SKAN troubleshooting is not a bug hunt. Apple does not publish the install volumes behind each tier, does not tell you which tier a postback came from, and evaluates the tier per install rather than per campaign. What you can do is understand which fields disappear at each level and structure spend so your important dimensions are the ones that survive.

The four tiers

A tier determines two things at once: how many digits of the source identifier you receive, and whether a conversion value is included at all.

Tier 0 is not an error state. It is a correctly functioning postback that tells you an install happened and which two-digit bucket the ad belonged to, and nothing else. A campaign that lives permanently at tier 0 is a campaign whose creative and audience decisions cannot be informed by SKAN, no matter how the SDK is configured.

The hierarchy is the lever you control

You always get the first two digits of the source identifier. Digits three and four are conditional. Since the ad network chooses the encoding, the order of that hierarchy is the single most consequential thing to agree with them.

Under Encoding A, a small campaign still tells you which campaign it was. Under Encoding B, the same install tells you only a region, and campaign-level reporting collapses entirely below tier 2. Both are valid SKAN implementations; only one is usable at low volume.

1. Ask each network how it maps the four digits, and get it in writing. Several networks let you configure this.
2. Put the dimension you would optimise on tomorrow in digits one and two.
3. Consolidate spend. Fewer, larger campaigns reach higher tiers; the same budget split across twenty ad sets may report nothing useful from any of them.
4. Avoid narrow geo and audience splits at campaign level — that fragmentation is exactly what pushes cohorts under the threshold.
5. Design a coarse value that is decision-grade on its own, because tiers 1 and 2 are where most campaigns live.

Diagnosing what you are actually seeing

Because the tier is never stated, you infer it from the shape of what arrived. This table is the diagnostic.

function inferredTier(postback) {
  const digits = String(postback["source-identifier"] ?? "").length;
  const hasFine = postback["conversion-value"] !== undefined;
  const hasCoarse = postback["coarse-conversion-value"] !== undefined;

  if (digits >= 4 && hasFine) return 3;
  if (digits === 3 && hasCoarse) return 2;
  if (hasCoarse) return 1;
  return 0;                       // no value field survived
}

// Report tier mix alongside installs. A campaign whose tier-0 share
// is rising is losing measurability, which no CV schema change fixes.
const mix = postbacks.reduce((acc, p) => {
  const t = inferredTier(p);
  acc[t] = (acc[t] ?? 0) + 1;
  return acc;
}, {});

The same thresholds apply to AdAttributionKit, so none of this is resolved by adopting the newer framework. It is the price of attribution without a cross-app identifier, and the useful response is structural — fewer, larger campaigns and a coarse value you can act on.