Audience overlap analysis counts the people who appear in two or more audiences at once and expresses that shared group as a share of one or both lists. An advertiser might compare its customer file with a publisher's subscribers, two retargeting segments in the same account, or the users reached by two campaigns. The output is usually a single figure, such as "25% of the brand's customers also read the publication".

The practice exists because audiences are built separately and then bought together. Lists assembled by different teams, platforms or partners rarely say how much they duplicate one another. Without a count, a buyer cannot tell whether a second segment brings new people or simply pays twice for the same ones, and a publisher cannot show how much of a brand's customer base it actually reaches.

How the calculation works

At its core the analysis is a set intersection. Take audience A and audience B, find the identifiers present in both, and count them. Two ratios then follow. The first divides the intersection by the size of one audience, usually the advertiser's own. AWS defines it this way in its March 12, 2024 technical guide: users found in both datasets divided by total users in the advertiser's audience. The second, the Jaccard index, divides the intersection by the union of both lists, giving a symmetric score between 0 and 1.

The difference matters in practice. Suppose a retailer holds 200,000 customer records and a publisher 2 million subscribers, with 50,000 people in both. Against the retailer's file the overlap is 25%. Against the publisher's base it is 2.5%. The Jaccard index is 50,000 divided by 2,150,000, roughly 2.3%. The same intersection can therefore be reported as high or trivial depending on which denominator a platform chooses, and most tools pick a "primary" audience without always saying so.

Before any counting, the two sides must agree on a match key: hashed email addresses, phone numbers, mobile advertising IDs, IP addresses or a third-party identifier such as RampID. AWS Clean Rooms accepts any column as a join key. The overlap that results can never exceed the match rate, since records that fail to link are invisible to the intersection. Poor normalisation of emails or phone numbers before hashing is enough to make two heavily shared lists look nearly distinct.

Some platforms add an index. Adobe's Real-Time CDP Collaboration documentation, last updated August 5, 2026, reports an "audience index" comparing how concentrated the overlap is against a baseline. Its worked example of ((500,000 / 1.3 million) / (20 million / 50 million)) x 100 produces 96, which Adobe classes as "medium" relevance within a band of 80 to 120.

Where the analysis runs

Three settings dominate. Inside self-serve ad platforms, overlap tools compare an advertiser's own audiences. Meta's Audience Overlap tool, found in the Audiences section of Ads Manager, requires each audience to hold at least 1,000 people and measures the others against the first audience selected, according to a January 14, 2016 description by Jon Loomer. Display & Video 360 (DV360), Google's demand-side platform, offers Unique Reach Overlap reports that count users reached across a pair of dimensions such as advertiser and campaign, alongside "exclusive" and "duplicate" reach.

The second setting is the clean room, a controlled environment where two parties join data without handing it over. Here overlap is usually the first query run. In the AWS walkthrough, a publisher creates the collaboration, the advertiser alone may query, and analysis rules permit only COUNT DISTINCT on the matched column, so neither party sees joined rows. Uber's Intelligence platform, built on LiveRamp, treats audience overlap analysis as the foundational capability, estimating what share of a brand's customers use Uber services. In the New York Times case, Decentriq found that one in four of a wealth manager's customers were also Times readers, and that overlap became the seed for lookalike segments.

The third is the auction itself. When ad sets from one Meta advertiser target overlapping people and become eligible for the same impression, Meta enters only the ad with the highest total value and holds the others back, according to Jon Loomer's account of Meta's auction overlap rule. The losing ad sets can under-spend or pay more for delivery. In programmatic buying, the IAB Tech Lab's PAIR (Publisher Advertiser Identity Reconciliation) protocol lets demand-side platforms use matched publisher identifier lists from a clean room to identify audience overlaps during auctions, without either side learning about users outside the matched set.

From botany to media plans

The arithmetic is older than advertising technology. Paul Jaccard published his "coefficient de communaute" in 1901 in the Bulletin de la Societe Vaudoise des Sciences Naturelles, comparing alpine plant species across Swiss valleys. Media planners later applied the same logic to readership and viewing panels, estimating how far two titles or channels duplicated one another before buying both.

Digital platforms turned the panel estimate into a direct count. Facebook began testing its overlap tool around January 2016, with no formal announcement, according to Loomer. Google announced the Ads Data Hub beta on May 24, 2017, Amazon Marketing Cloud entered beta on January 5, 2021, and AWS Clean Rooms reached general availability on March 21, 2023. Each made overlap queryable against platform logs rather than modelled.

Interoperability came next. Google's DV360 team built PAIR in October 2022 and donated it; the IAB Tech Lab launched PAIR as an open standard on September 26, 2024, with version 1.1 following in July 2025. Adobe launched Real-Time CDP Collaboration on February 20, 2025, with dashboards for advertisers and publishers to compare audience overlap. LiveRamp's roughly $200 million purchase of Habu, announced in January 2024, consolidated another clean room vendor into the overlap business.

Why it matters to marketers

Overlap sits behind three recurring decisions. In media planning, it shows whether a partner offers reach a brand does not already have. In account structure, it explains why two ad sets fighting for one person inflate costs. In measurement, it determines whether the same person is counted twice across campaigns or channels.

The scale of that last problem is documented. Amazon cited a consumer goods company that found 34% reach overlap across three brands' orders, a figure from Amazon's own documentation. A TransUnion and EMARKETER survey of 196 US marketers, published in October 2025, found that 48% named cross-channel deduplication as a barrier to trusting their measurement.

Limits, thresholds and disputes

Privacy controls shape every result. Clean rooms suppress rows that cover too few people. Ads Data Hub requires about 50 unique users per row under its legacy difference checks, about 20 with noise injection and about 10 for click- and conversion-only queries, according to Google's developer documentation. AWS lets collaborators set the minimum; its example uses 10. Adobe returns nothing for comparisons with fewer than 1,000 overlapping identities. Small or niche intersections therefore vanish, which biases analysis towards large audiences.

Thresholds alone do not stop leakage. As PPC Land's k-anonymity explainer notes, overlapping queries can still isolate individuals unless difference checks or noise are added. A party that queries the overlap of a list, then the same list minus one person, learns whether that person is present. Noise and suppression fix the leak at the cost of precision.

Other weaknesses are methodological. Overlap reflects matchable identifiers, not people, so households with shared devices or several email addresses distort it in both directions. Percentages are asymmetric, and vendors rarely publish which denominator they use. Practitioner rules of thumb, such as treating 40% to 60% overlap as saturation, have no published empirical basis. Overlap also says nothing about incrementality: two audiences may share few people and still reach buyers who would have purchased anyway.

On Meta, the tool's relevance is contested. Meta has pushed advertisers towards consolidated, broadly targeted Advantage+ structures and removed detailed targeting exclusions in January 2025, one of the traditional fixes for overlapping ad sets. Sources also disagree on the tool's limits: Loomer's 2016 account allowed up to five audiences, while a Clix Marketing post by Kristin Kusiak dated August 27, 2024 described a maximum of four.

Not the same as

Audience duplication and deduplicated reach. Duplication is the problem; overlap analysis is one way to size it. Deduplication removes repeated people from a reach count, and reach and frequency reporting expresses the result as unique people and average exposures. Overlap analysis usually precedes buying; deduplicated reach reports what was delivered.

Lookalike modelling. Lookalike and actalike models find new people who resemble a seed list. Overlap analysis counts people already shared. The two often run in sequence, as in the New York Times case, where the overlap became the seed.

Identity resolution. Identity resolution decides which records belong to the same person. Overlap analysis depends on that output but does not perform it; its accuracy is capped by how well the underlying identities were joined.

Recent developments

Reach-overlap reporting has expanded across platforms through 2026. DV360 changed its Unique Reach calculation method on January 13, 2026, and in April 2026 added campaign-level and placement-level overlap dimensions to DV360 and Campaign Manager 360 reports. Amazon launched de-duplicated reach across multiple Amazon Ads accounts on April 2, 2026, in 36 countries.

Clean room overlap is moving into cross-media measurement. LiveRamp extended its Cross-Media Intelligence to accept Meta campaign data inside its clean room on August 4, 2026, ahead of Publicis's $2.5 billion acquisition of LiveRamp. On September 29, 2026, the IAB published guidance on clean room use in commerce media that lists audience overlap first among common use cases, ahead of enrichment, targeting and suppression.

Timeline

  • 1901 - Paul Jaccard publishes the coefficient de communaute, the basis of the Jaccard index, in a study of alpine flora.
  • January 2016 - Facebook begins testing its Audience Overlap tool in Ads Manager and Power Editor.
  • May 24, 2017 - Google announces the Ads Data Hub beta.
  • January 5, 2021 - Amazon Marketing Cloud enters beta.
  • October 2022 - Google's DV360 team creates the PAIR protocol.
  • March 21, 2023 - AWS Clean Rooms becomes generally available.
  • January 2024 - LiveRamp announces its acquisition of clean room vendor Habu for about $200 million.
  • March 12, 2024 - AWS publishes a reference method for audience overlap analysis in AWS Clean Rooms.
  • September 26, 2024 - IAB Tech Lab launches PAIR as an open standard.
  • December 23, 2024 - Amazon adds cross-campaign deduplicated reach reporting for DSP Frequency Groups.
  • January 2025 - Meta removes detailed targeting exclusions.
  • February 20, 2025 - Adobe launches Real-Time CDP Collaboration with audience overlap dashboards.
  • July 2025 - IAB Tech Lab releases PAIR 1.1.
  • December 8, 2025 - Uber announces Uber Intelligence, built on LiveRamp's clean room.
  • January 13, 2026 - DV360 changes its Unique Reach calculation method.
  • April 2, 2026 - Amazon launches cross-account deduplicated reach reporting.
  • April 2026 - DV360 and Campaign Manager 360 add campaign- and placement-level overlap dimensions.
  • August 4, 2026 - LiveRamp adds Meta data to Cross-Media Intelligence.
  • September 29, 2026 - IAB publishes clean room use-case guidance for commerce media, led by audience overlap.

Summary

Who. Advertisers, agencies and publishers run overlap analysis, using tools from ad platforms such as Meta, Google and Amazon, and clean room providers including AWS, LiveRamp, Adobe and Decentriq. The IAB Tech Lab sets the PAIR matching standard that carries overlap into programmatic auctions.

What. It is the count of people shared between two or more audiences, expressed as a percentage of one list or, under the Jaccard index, of their union. It depends on a common match key and is constrained by match rates and privacy thresholds.

When. The underlying measure dates from 1901. Facebook's self-serve tool appeared in 2016, clean rooms made overlap queryable from 2017 onwards, and 2026 brought new overlap dimensions in DV360 and cross-account deduplicated reach at Amazon.

Where. It runs inside ad platform interfaces, in clean rooms operated by platforms and independent vendors, and implicitly in auctions where an advertiser's ads compete for the same person.

Why. Audiences are built separately and bought together. Overlap analysis shows whether a partner adds new reach, whether ad sets are bidding against one another and whether measurement counts the same person twice, though its results are bounded by identity quality, suppression rules and the choice of denominator.