Why Cross-Screen Reach Is Not Three Platform Totals
Cross-screen video audience measurement asks a deceptively simple question: how many people did a campaign reach across television, OTT, and digital video? This opening chapter focuses on why that answer cannot be produced by adding three platform reports, and gives the reader a way to distinguish a platform’s own audience number from one unduplicated cross-media audience. The practical judgment is straightforward: before accepting a single total, ask what counted as an exposure, which population it describes, and how the same person was recognized across screens.
The arithmetic looks easier than the audience
Imagine a campaign report with three figures: television reached 60 people, streaming reached 50, and a digital video platform reached 40. If all three figures refer to the same 100-person population and the audiences never overlap, the campaign reached 150 people—an impossibility. If every viewer watched on all three screens, the campaign reached only 60. The arithmetic is easy; the audience is not.
The figures in this example are illustrative, but the problem is not. A platform can often report the unique users or homes it observed within its own system. That number may be useful for evaluating that platform. It does not, by itself, say how many of those people also appeared in either of the other reports. A person who watched a television spot, an OTT copy, and a mobile stream is three exposure records but one member of the campaign’s reached audience.
This is why “unique” needs a boundary. Unique within a platform is not necessarily unique across the campaign, across devices, or across the full reporting period. The MRC Cross-Media Audience Measurement Standards define reach as the unique, unduplicated users, homes, or audience exposed at least once during a stated period through a qualifying Viewable Impression. Frequency is the number of qualifying impressions divided by that unique audience. Gross Rating Points, or GRP, connect the two: reach multiplied by frequency, or total qualifying impressions divided by the measured universe.
Those relationships are useful only if their terms remain stable. A report that calls a browser unique, a household unique, and a person unique is not necessarily describing the same unit. A report that counts a server delivery, a partially displayed digital ad, and a television exposure under different rules is not necessarily describing the same event. Adding the resulting totals can create a precise-looking answer whose components do not share a definition.
The arithmetic of overlap makes the problem more explicit. For two platforms, the combined audience is the first platform's audience plus the second's, minus the people counted in both. With three platforms, the calculation also depends on each pairwise overlap and the people present in all three. A report that supplies only the three platform totals has not supplied those overlap terms. Even a reported all-platform union does not reveal which platform produced the incremental people, so it cannot answer every planning question. The missing information is not a small adjustment to the total; it is part of the total's definition.
The central subject here is therefore not a dashboard total but the measurement system behind it. Its job is to turn exposures observed in different environments into a defensible estimate of one audience. That job has two distinct parts. First, the exposures must be comparable enough to combine. Second, the records must be connected well enough to determine which exposures belong to the same people. Failure at either stage changes what the final number means.
A cross-media number has a definition before it has a value
The MRC standard treats cross-media video measurement as a common framework for advertising and content across television, OTT, and digital video. Its stated goals include consistent definitions, comparability, disclosure, quality practices, and a basis for deduplication. The standard’s minimum cross-media universe is the deduplicated total of persons in the media universe for each medium—or, more generally, total persons. That requirement makes the denominator part of the audience claim, not a technical footnote.
A cross-screen reach percentage can be understood as a fraction: the number of unique people with at least one qualifying exposure divided by the measured population, universe, or target. Change the numerator and reach changes. Change the denominator and reach changes again. Compare a television report projected to one population with a digital report measured only among another, and the apparent agreement between their percentages may be misleading even before overlap is considered.
The same is true of the numerator’s qualification. For combined and deduplicated cross-media video, the MRC standard specifies a common viewability basis for digital and linear components: the entire ad must be on screen for at least two continuous seconds. It also requires filtration for both General and Sophisticated Invalid Traffic. A platform may still report a different standalone metric, but that metric must be distinguished rather than quietly mixed into the cross-media total. The purpose is not to declare one platform’s native metric worthless. It is to prevent unlike measurements from masquerading as one currency.
Even a qualifying Viewable Impression makes a narrower claim than everyday language may suggest. The standard describes viewability as an opportunity to see and warns that viewable conditions can occur without a person being present or actually seeing the ad. A unified reach number is therefore not a count of attention, persuasion, or sales. It is a count or estimate of people associated with exposures that met the defined measurement conditions.
That distinction also explains why an ad report cannot always borrow a content report. In a dynamic streaming environment, different users may receive different ads inside the same program. The MRC guidance calls for discrete measurement of those ad occurrences rather than assuming that everyone who watched the content saw every ad. A static broadcast model is treated differently because the same content and ad load can be delivered to everyone in the measured stream. The delivery model changes what may be used as a proxy.
The common ruler is thus more than a viewability threshold. It includes the event that starts the count, the population to which it is projected, the treatment of invalid activity, the distinction between ad and content delivery, and the time period over which uniqueness is defined. Only after those choices align does it make sense to ask how much the audiences overlap.
This does not mean that every media-specific metric must disappear. The MRC standard allows such metrics to sit alongside cross-media metrics when their bases are clearly labelled, with the shared cross-media measure treated as the standard basis and the others as diagnostic. That distinction is valuable because a native platform metric may be useful for optimizing delivery inside that platform while still being unsuitable for a deduplicated campaign total. The choice is not between using one number everywhere and using no platform detail. It is between combining measurements transparently and disguising their differences.
Overlap is the unresolved evidence
The missing quantity in the three-report example is overlap. How many people appear in both television and OTT? How many appear in all three? Which platform’s audience is genuinely incremental? These are not secondary questions reserved for a post-campaign appendix. They determine the difference between total exposures and total people reached.
The subset-reach research paper makes this problem explicit by representing platforms, publishers, sub-platforms, or campaign groups as buying groups. A complete reach Venn diagram would describe the audience of every non-empty subset of those groups. In practice, privacy and compute constraints can make it infeasible to observe every combination. An advertiser may know the reach of each group and the reach of the full union while still lacking the reach of a particular pair—or the incremental reach created by adding one platform to an existing plan.
The paper gives a simple illustration: if three publishers each report a reach of 100,000 users, the incremental reach of the third publisher after the first two could range from zero to 100,000 without extra overlap information. Both extremes are compatible with the individual totals. The platform reports are not necessarily wrong; they are simply insufficient to answer the cross-platform question.
That gap is where the rest of the measurement problem begins. Identity methods attempt to establish which records belong to the same person, but the MRC standard treats cross-device deduplication as difficult and privacy-sensitive. It calls for empirical support and actual consumer duplication information rather than a model with no truth or learning set. A perfect identity graph would still not fix incompatible exposure definitions, a missing population, or invalid activity. Conversely, when privacy limits direct joins, the absence of a person-level match is not evidence that the audiences do not overlap.
The relevant overlap also depends on the decision. A buyer asking whether a streaming placement adds new people to a television plan needs the reach of that streaming group outside the television audience. A buyer asking for total campaign reach needs the union across every included group. Those are related quantities, but they are not interchangeable. This is why a measurement system can be honest about a total union while still being unable to answer which platform delivered the next incremental person. The evidence required should follow the question being decided.
A disciplined system should make the unresolved overlap visible. When some subset reaches are observed, the statistical question is not “What number would look best?” It is “What audience structures remain possible given these observations?” The research paper’s model-free approach answers with lower and upper bounds derived from the non-negative regions of a reach Venn diagram. A later model-based approach can offer a point estimate, but only by adding assumptions about how reach probabilities behave within activity segments. The range comes first because it shows what the observations alone can support.
The evidence supports a limited but useful conclusion. One cross-screen audience is achievable as a defined measurement construct, but it is not the sum of three platform totals and it is not proof of attention or effectiveness. A credible number needs a common exposure ruler, a stated universe, honest coverage, evidence-backed identity and deduplication, and an explicit treatment of whatever overlap remains unobserved. The next question is therefore precise: what must television, OTT, and digital video count in the same way before their exposures can be combined at all?