The Audience Overlap Report Your DSP Provides Describes Identifier Overlap, Not Person Overlap
DSP audience overlap reports count shared identifiers across segments, not shared people, so the deduplication logic your media plan relies on may be understating true audience redundancy.
When a media buyer looks at an audience overlap report inside a DSP, the natural reading is straightforward: this percentage of people in Segment A also appear in Segment B. That reading feels safe because the numbers are specific and the interface is clean. The problem is that the report is almost never counting people. It is counting identifiers, and those are different things in ways that change how you should plan, buy, and deduplicate.
What the overlap report is actually counting
A DSP manages audiences as collections of identifiers. Depending on the environment, those identifiers might be cookie IDs, device IDs, hashed email addresses, or proprietary graph tokens. When the platform compares two segments and reports that they share, say, 18 percent overlap, it is reporting that 18 percent of the identifiers in one segment also appear in the other.
A single person who has a browser cookie, a mobile device ID, and a hashed email address in the graph is represented as three separate identifiers. If Segment A picked her up via her cookie and Segment B picked her up via her device ID, the overlap report will show zero shared identifiers between those two records, even though both segments are targeting the same individual. The deduplication looks clean. The actual reach is not.
The reverse is also possible. If a household shares a device, one physical person might generate overlapping identifiers across segments while the report counts that as overlap for a single individual who is actually two or three people. Both errors flow from the same root cause: identifier-level reporting does not have stable access to the person construct that media planning assumes.
Why this matters for budget allocation
Media plans that use overlap reports to justify running multiple audience segments simultaneously are implicitly assuming that the overlap percentage describes redundant reach against real people. If the true person-level overlap is higher than the identifier overlap the report shows, the plan is buying more duplication than intended, paying twice to reach the same individuals through different identifier paths.
This is particularly relevant when a campaign runs across both cookie-addressable and identity-resolved inventory in the same flight. The cookie graph and the people-based graph do not share a common deduplication layer inside most DSPs, so audiences that look non-overlapping in the platform's reporting can be substantially redundant when measured against a person-level identity spine.
For buyers managing frequency alongside reach goals, the gap compounds. Frequency caps set at the identifier level already allow one person with multiple IDs to see more impressions than the cap intends. Overlap reports that also operate at the identifier level mean the deduplication logic your plan uses to estimate unique reach is working from the same incomplete picture as your frequency cap. The two errors reinforce each other.
A practical way to pressure-test the report
One useful diagnostic is to compare your DSP overlap report against a person-level deduplication run inside a clean room or through an identity resolution partner that resolves all active identifiers to a common household or individual key before counting. If those two outputs show materially different overlap percentages for the same pair of segments, the gap is telling you something real about how much of your planned audience separation is identifier separation rather than person separation.
This does not require a complex implementation to be informative. Even a rough comparison on a subset of your audience, using your CRM-onboarded segment as the reference spine, can reveal whether the DSP's overlap figures are directionally reliable for your specific inventory mix. If the clean room consistently shows higher person-level overlap than the DSP reports identifier-level overlap, that pattern is worth carrying into how you size your reach estimates going forward.
As a hypothetical example: suppose a buyer runs two behavioral segments targeting in-market purchasers and plans them as largely non-overlapping based on a DSP report showing 12 percent shared identifiers. A person-level reconciliation might reveal that 30 percent of actual individuals appear in both segments when hashed emails and device graphs are resolved together. That 18-point gap represents real budget delivering redundant impressions to people the plan assumed were being reached only once.
What to ask your DSP or identity partner
A few questions are worth raising with your platform or data partner before leaning on overlap reports for planning decisions.
First, ask what identifier types are included in the overlap calculation and whether the platform resolves those identifiers to a person or household level before comparing segments, or whether the comparison happens at the raw identifier level. Many platforms will be straightforward about this, and the answer tells you immediately how much weight to put on the output.
Second, ask whether your identity onboarding partner offers a segment overlap view that resolves through their graph rather than the DSP's internal identifier pool. Some do, and that view will reflect a different and potentially more accurate picture of person-level redundancy depending on how comprehensively their graph covers your customer population.
Third, consider whether your measurement partner can produce a post-campaign deduplication report that counts unique reached individuals across segments, not unique identifiers. This does not fix the planning gap, but it does tell you retrospectively how much of your planned audience separation held up in delivery, which informs how conservatively to treat overlap estimates on the next flight.
The planning posture that accounts for identifier granularity
Until DSPs expose person-level deduplication natively across all identifier types and environments, the safest posture is to treat DSP overlap reports as a lower bound on redundancy rather than a precise measurement. If the report shows 15 percent overlap, the true person-level figure is unlikely to be lower, and could be meaningfully higher depending on the identifier mix in your specific segments and inventory.
Building in a conservative redundancy buffer when estimating unique reach from multi-segment plans is not pessimism. It is an acknowledgment that the reporting layer and the planning assumption are operating at different levels of resolution. Closing that gap with a person-level check, even an approximate one, produces plans that behave more like the spreadsheet intended once they run in the real world.