What is Audience overlap?

Audience A holds 400,000 people and audience B holds 150,000, with 90,000 belonging to both. Overlap is 90,000 ÷ 150,000 = 60% of the smaller audience, and 90,000 ÷ 400,000 = 22.5% of the larger. Quoting the second figure makes the overlap sound modest; the first is the one that says a comparison between the two audiences is measuring mostly the same people.

Audience overlap is the share of people in one audience who also belong to another. The arithmetic is trivial and the denominator is where the argument happens.

Overlap = people in both audiences ÷ people in the smaller audience

Measured against the smaller audience, 90,000 shared members out of 150,000 is 60%. Measured against the larger, the same 90,000 out of 400,000 is 22.5%.

Both are correct and they support opposite conclusions. The figure that matters for most decisions is the first, because the smaller audience is the one substantially composed of people who belong somewhere else. Reports that quote the larger denominator make overlap sound like a minor issue, which it sometimes is and frequently is not.

The folklore holds that overlapping audiences bid against each other and drive up the advertiser's own costs.

There is something in it and less than is usually claimed. Platforms generally arbitrate between an advertiser's own competing entries rather than allowing them into the same auction to raise each other's price, precisely because that outcome benefits nobody including the platform.

What does happen is that the platform chooses. Whichever entry it prefers receives the delivery, and the choice is made on its own estimate rather than on the advertiser's intent. That is not price damage; it is control quietly moving.

The consequential damage is to interpretation.

Two overlapping audiences cannot be compared. The same people are eligible in both, delivery splits between them on the platform's preference, and the resulting figures describe two arbitrary slices of a shared population rather than two audiences.

This matters because comparing audiences is exactly what accounts use them for. A test of interest targeting against a lookalike, of one lookalike percentage against another, of a broad audience against a narrow one — all of these are routinely run with substantial overlap and all of them produce a result that cannot bear the weight put on it.

The winner in such a test may have won because the platform preferred it, or because the overlap handed it the best shared members. Pausing the loser then removes people the winner was never reaching.

The only version of the test worth running uses mutual exclusion, which costs a setting and makes the populations genuinely disjoint.

Overlap also concentrates frequency, and this one is felt rather than measured.

A person belonging to three of an account's audiences is eligible for three campaigns. Each may serve them. The account's per-campaign frequency reports look reasonable; the person's actual experience is three times that, and the negative feedback that follows attaches to the brand rather than to any one campaign.

Per-campaign frequency is therefore an understatement whenever audiences overlap, and an account with heavy overlap has no accurate view of how often anybody is seeing its advertising.

Consolidate, first. Heavy overlap between two audiences usually means they were never distinct. Pooling them into one gives the delivery model more events to learn from, leaves learning faster, and produces less repetition against the shared members. The instinct to preserve the split is usually about reporting tidiness rather than about a real difference.

Exclude, where both must run. Where the audiences genuinely serve different purposes — a prospecting pool and a retargeting pool, two markets with different strategic weight — mutual exclusion makes the definitions real.

Prioritise deliberately. When excluding, decide which audience should win the shared members. Usually the more specific one: a cart abandoner belongs in the cart campaign, not in the general visitor campaign.

Check before splitting. Most splits that look like segmentation divide the same people by an attribute that does not change who they are. Checking overlap before building the split saves the work and the subsequent confusion.

Recheck after building lookalikes. Models from related seeds find related people, and a set of four lookalikes built in an afternoon will frequently turn out to be one audience with four names.

Overlap is computed from the platform's own view of who is in each audience, which means it inherits every limitation of that view.

Audiences built from uploaded lists include only matched members. Audiences built from browser activity include only people whose activity was observed and retained. Two audiences can overlap far more in reality than the tool reports, simply because the parts it cannot see are the parts most likely to coincide.

And the figure is a snapshot. Membership windows expire, lists refresh, and behaviour-based audiences turn over continuously, so an overlap measured last quarter describes last quarter's audiences.

Where several ad sets share one campaign budget, overlap stops being a reporting nuisance and becomes an allocation problem.

The platform is deciding which ad set gets the money, and it decides on predicted result. Two overlapping ad sets are partly competing for the same people, so the comparison that drives the allocation is contaminated in the same way the reporting is. Whichever ad set gets an early lead accumulates delivery, accumulates data, and reinforces its own advantage — and the lead may have come from the overlap rather than from merit.

The outcome is a campaign concentrated on one ad set for reasons nobody chose, with the others starved of both budget and evidence.

The rule that follows is worth stating plainly: put things in the same campaign when you genuinely want the platform to choose between them, and make sure the things it is choosing between are actually different. Overlapping audiences fail the second test even when they pass the first.

When an account has more audiences than anybody can reason about, three questions sort them out fast.

Which pairs overlap heavily against the smaller audience? Those are candidates for consolidation, not for exclusion.

Which pairs are meant to be compared? Those need mutual exclusion before any comparison is believed.

Which pairs serve genuinely different strategic purposes? Those need exclusion and a decision about which one wins the shared members.

Everything else can usually be left alone. The failure mode in this area is not insufficient management of overlap; it is an account that has split itself into so many audiences that overlap became inevitable.

• Measure the overlap against the smaller audience, because that is the one whose comparison is compromised.
• Make audiences mutually exclusive before comparing them. A test between two overlapping audiences is not a test.
• Consolidate rather than exclude when overlap is heavy and the audiences were never strategically distinct. Two near-identical ad sets pooled into one learn faster and fatigue slower.
• Check overlap before splitting an audience further. Most splits that look like segmentation are the same people divided by an attribute that does not change who they are.
• Recheck after building lookalikes from related seeds, because models from similar sources land on similar people.