Analysis: Audience data is plentiful but agreement on what it means is not

Ask measurement and streaming technology vendors what broadcasters need most, and “more data” rarely comes up.

A recent Industry Insights roundtable on audience intelligence described an industry that collects more viewer signals than at any point in television history yet still struggles to reconcile them, route them to decision-makers and act on them while they still matter.

The common themes uncovered were less about new metrics and more about joining existing data across platforms, departments and with ad providers. 

The problem is not volume

“The industry has access to more audience data than ever, but having more data does not automatically create better decisions. The disconnect often comes when data lives across different systems, uses different definitions or arrives too late to influence what happens next,” said Peter Nummerdor, senior vice president of product management at VideoAmp.

“Customer data is all too often sitting in separate silos across product, marketing and operational teams and without a more unified approach, it’s difficult to see what impact different decisions have on viewers’ experience,” said Mark Paramor, director of operations assurance at Accedo.

Coactive framed the problem as translation.

A brand’s brief passes through a demand-side platform, the software buyers use to purchase inventory, and then to a publisher, and each handoff uses a different vocabulary.

“Every translation step in the supply chain strips away meaning. The advertiser speaks in moods, the DSP speaks in keywords, the publisher speaks in genres, so the brief arrives at the inventory with the intent already gone,” said Josh Heller, chief revenue officer at Coactive.

Advertisement

Heller’s example: six weeks of creative strategy produced a target of “joyful family gatherings,” and the system answered with “cooking and food.” Anyone who has sat through a holiday dinner may see the overlap. Most media buyers would not call it a match.

Two candidates for a common key

Ampersand started with identity. Linear TV still depends largely on panels and set-top box data, while streaming and connected TV rely on impression-level data tied to device IDs and IP addresses. Those systems were not built to talk to each other.

“Without a reliable cross-channel identity spine, marketers will struggle to determine whether streaming is actually reaching new households or simply reaching the same households they already reached through linear TV,” said Gerrit Niemeijer, chief technology officer at Ampersand.

Heller argued identity will not fully reconcile, because data rights and delivery mechanics differ by platform. His alternative was the program itself.

“The content itself is the one thing that travels intact, so make the content the common key: index it once, consistently, and let the different audience datasets attach to it. You get a consistent view of what performed even where you cannot get one of who watched,” said Heller.

One approach answers who was reached; the other answers what worked. Most organizations need both.

Delivered is not the same as watched

With server-side and server-guided ad insertion, ads are stitched into or signaled within the stream, and the player may not know an ad is playing.

“Broadcasters need measurement that accurately reflects what happens on the client side, at scale, so they can prove an ad was actually played rather than simply delivered,” said Paul Davies, head of marketing at Yospace.

Davies pointed to two emerging standards.

The Streaming Video Technology Alliance’s Ad Creative Signaling supplies standardized ad metadata to the player. The Consumer Technology Association’s Common Media Client Data version 2, or CMCDv2, reports playback state back to a measurement endpoint, which matters when content is syndicated to platforms where the broadcaster does not control the player.

Bitmovin raised the same issue from the playback side.

The content delivery network, the player and the measurement vendor can each report a different result. A drop-off at minute four means something different when startup took six seconds than when playback was clean.

Advertisement

“A lot of what gets read as weak content is actually a delivery problem, and separating the two is one of the most useful things video analytics can do right now,” said Reinhard Grandl, chief product officer at Bitmovin.

Measurement on the right clock

“They need different clocks, not different metrics,” said Heller, referring to live and on-demand content.

Davies pointed to the 2026 FIFA World Cup, where large audiences hit ad breaks at once and break timing depended on play. He described a prefetch approach that uses real-time audience data to request ads earlier and pace those requests, reducing unfilled spots. Paramor said composable platforms can compress decision and development cycles from weeks or months to hours or minutes.

On prediction, the panel was measured.

Looper Insights described on-screen placement as a leading indicator, noting that a title not visible on connected TV home screens is “unlikely to be discovered, let alone watched,” said Francesca Pezzoli, vice president of marketing at Looper Insights.

Grandl drew a line between operational and audience forecasting.

“At Bitmovin, we see more value in developing models that shorten the reaction time on live streams, than in models that try to pick the next hit,” said Grandl.

Show the methodology

As models produce more measurements, several panelists said transparency has not kept pace.

Nummerdor said AI should support decisions with transparent data “rather than create another black box for teams to interpret.” Heller said a system that cannot show its reasoning is “a governance problem no matter how clean your consent posture is.”

Grandl recommended deciding what data is needed before collecting it and keeping session-level quality data separate from identity data.

Advertisement

“If two teams can’t agree on what a view is, no amount of additional measurement will settle it,” said Grandl.

What the industry is not measuring

“Once recommendation and discovery run through AI agents, a growing share of ‘audience behavior’ is a machine acting on someone’s behalf, and the panels and dashboards we have today have no way to tell the difference,” said Grandl.

He also noted that every stack reports impressions delivered, while none reports how many viewers left during the break. That cost shows up in completion rates and return visits that then get attributed to programming.

Heller noted automation’s limits. Coactive once prompted a model with “dirty face” and received Halloween makeup rather than everyday grime, because the model had no brand definition to anchor it.

“That AI gives you the map, not the destination,” said Heller.

The shared view was that measurement’s problem is now coordination. The tools can see more than ever. The work ahead is agreeing on what counts, who acts on it and how fast.