Measurement, not proxies, is what unlocks real performance.
The industry has been talking about attention metrics for years.
The problem is the conversation kept rotating through proxies. First viewability. Then completion rate. Then time-in-view. Each one a reasonable approximation of whether a human being might have engaged with an ad. None of them actually measuring attention.
We have seen this before. Identity went through the same cycle. Everyone claimed deterministic matching until you asked them to prove it. The methodology was opaque. The match rates were suspiciously high. The moment you stress-tested the signal, it broke.
Attention is starting to follow the same path.
What Attention Actually Requires You to Measure
Attention is not one thing. It is a set of distinct signals. Most vendors compress that into a single score and move on. That is where the slippage starts.
Start with the most basic question: is a human being physically in front of the screen? It sounds trivial. It is not. The entire CTV measurement ecosystem treats the device as a proxy for the viewer. It is not. A TV being on is not the same as someone watching it.
Then there is a second, separate question: is that person actually looking at the screen during the ad? That is a different signal entirely, and it requires a different measurement approach.
The signal almost nobody deals with seriously is how many people are in the room. Co-viewing changes the math in both directions. If a household is watching together, that is not one impression. It is several. If those same people are on their phones during the ad break, the effective delivery collapses back down. The delivered impression number captures none of this. It only tells you the ad ran.
Now step back and think about what that means. Every campaign optimized to delivered impressions is built on a single assumption: that delivery equals exposure. It does not. Presence in the room discounts that number. Attention discounts it further. That is not a failure of the system. It is just how people watch TV.
What almost never gets discussed is that the same dynamic works in the other direction. Co-viewing expands the value of an impression when multiple people are present and paying attention. The delivered impression is simultaneously overstating and understating what you bought, depending on the household. The current model prices neither correctly.
Most attention vendors measure one layer of this stack and call it attention. That is not measurement. It is one signal standing in for a system.
Every Metric Is a Proxy for the One Above It
Terence Kawaja put a framework on stage at a recent Marketecture Live event that I have not been able to stop thinking about. He called it a conversion data hierarchy. The idea is straightforward, but the implications are significant.
At the base, you have an impression. An impression is not really a measurement of anything meaningful. It is just a unit of counting. So the industry built viewability on top of it. A viewable impression is better than an unknown impression, but it is still a proxy. It tells you the ad had an opportunity to be seen. It does not tell you whether anyone actually saw it.
From viewability, the industry moved to attention. Better still. But attention, as Kawaja framed it and as it is currently measured by most vendors, is itself a proxy for context, for audience quality, and ultimately for intent. Intent is what he called the “golden elixir.” It is the only signal in the stack that consistently correlates with outcomes. Everything below it is a progressively less accurate approximation of the thing you actually care about.
The reason this matters for CTV is that the channel entered the market promising performance but got measured like a branding vehicle. Delivery became the metric. Reach and frequency became the optimization target. The gap between delivery and actual viewer attention, which in a shared living room environment is enormous and cuts both ways, got papered over because the infrastructure to measure it did not exist.
When you cannot measure attention accurately, you cannot connect CTV investment to outcomes with any real confidence. Attribution is built on delivery data, not viewing data. Optimization is working on inputs that do not actually correlate with the outcome you are trying to drive. Fixing the attention layer is not about adding another metric to a report. It is about building the foundation that makes everything above it, audience quality, intent signal, outcome measurement, actually mean something. Without it, you are optimizing a proxy of a proxy and calling it performance.
Where the Real Money Still Lives
There is a tendency in this industry to talk about linear TV in the past tense. That is a mistake. The majority of TV advertising dollars are still flowing into linear. These budgets are enormous, they move slowly, and they are almost entirely priced on demographic proxies: age, gender, DMA. That is it. No behavioral signal. No intent data. No attention layer whatsoever.
Linear has always operated with a different measurement contract than digital. Demo and daypart are the inputs. Reach and frequency are the outputs. That has been the standard for decades and it has worked well enough to sustain massive budgets. But the industry is at a point where advertisers are asking harder questions about what they actually got, and the current measurement infrastructure was not built to answer them.
This is where the attention infrastructure argument becomes a business argument, not just a measurement one. If you can see who is actually watching specific shows, on specific networks, in specific dayparts, and connect that to purchasing behavior or in-market signals, you get something the linear market has never had. The ability to evaluate inventory based on who is actually there and how likely they are to respond, not who was expected to be watching.
That changes what linear planning looks like entirely. Instead of buying a demo on a network and hoping, you are evaluating shows and dayparts based on where your category of buyer is actually present and paying attention. That might change how you buy it. It might change how much you buy.
The dollars will follow the data. They always do. The question is who builds the infrastructure to deserve them.
What Deterministic Attention Actually Requires
I have written a lot about what deterministic identity requires: a persistent, verifiable signal that can be validated against a ground truth source. Attention measurement has the same requirements and the same failure modes.
A deterministic attention signal requires a panel. Not a modeled audience. Not a probabilistic extrapolation from ACR data. A real panel of households where you have ground truth on who is in the room, where they are looking, and what they are watching, second-by-second.
That panel needs to be census-matched. It needs to cover linear and CTV simultaneously. It needs to reach into environments that do not share signal with the open ecosystem. And it needs to be large enough to project nationally with statistical confidence.
What That Infrastructure Actually Looks Like
There are very few datasets that actually clear that bar. TVision is one of them. A panel of U.S. households with direct observation of who is in the room, where they are looking, and what they are watching, second by second, across both linear and CTV. The kind of ground-truth dataset the industry has described in theory for years without consistently delivering.
Viant’s acquisition of TVision reflects a deliberate execution of a cohesive vision from Tim and Chris Vanderhook, supported by a team that has been building toward this moment. Viant already has IRIS_ID, which analyzes video content at the frame level, and Household ID, which resolves audiences across devices at the household level. Bringing this dataset into that foundation connects those capabilities to direct observation of real viewing behavior.
What that enables is a closed loop. Planners can identify inventory where their audience is demonstrably present and engaged, execute against it, and measure whether that signal held. That is a different form of optimization than anything built on delivered impressions.
The implication for linear is straightforward. TVision’s panel covers broadcast and cable, not just CTV. The budgets still flowing through those channels, priced on demographic proxies, can now be evaluated against observed behavior. Not who was expected to be watching. Who was actually in the room, and whether they were paying attention. That signal does not stop at measurement. It can be applied to CTV and programmatic activation, enabling sequential messaging tied to real outcomes.
The pieces to do this have existed in isolation for years. What changed is that they are now connected and tied directly to buying decisions. That is when attention stops being a proxy and becomes an input into how media is actually executed.