Smart TV data is the most talked-about signal in advertising right now. Here's a plain-language guide to what it is, how modeling enhances its value, and why it matters whether you're running a political campaign, launching a new prescription drug, or finding emerging audiences to engage.
Every advertiser has heard some version of the pitch: "We have TV viewing data." Fewer can explain what that data actually is, what happens to it before it becomes something you can buy, or why one provider's audience performs differently from another's.
That gap matters. The difference between a useful audience and an expensive one usually comes down to what happened between the raw signal and the segment in your DSP. Below are the five questions we get asked most often.
1. What is ACR data, and why do advertisers care about it?
ACR stands for automatic content recognition. It's the technology built into most smart TVs that identifies what's on the screen (a show, a movie, and even a specific commercial) by matching what's displayed against a reference library.
For advertisers, that produces something traditional TV measurement never could: a household-level record of what actually played on the screen, including the ads.
That's the important part. Panel-based measurement tells you roughly how many people a program reached. ACR data tells you which households were exposed to a specific spot, how often, and what else they watched around it.
So instead of planning against "adults 35–54 who watch cable news," you can start from something far more precise: households that saw your competitor's commercial, households that watched a particular debate, households that consistently tune into a certain genre.
→ Why it matters: ACR moves TV from an estimate to an observation. That's a better starting point for almost any campaign. But it's only a starting point, which brings us to the next question.
2. If ACR data is so precise, why does it need to be modeled?
Because precision and scale are two different problems, and raw ACR data only solves the first one.
ACR data represents households that own a manufacturer's TVs and have opted in to sharing their TV viewing data, providing a real, verified, behaviorally rich audience foundation. While highly valuable for audience creation on its own, certain targeting requirements can narrow that audience size. Modeling helps extend those insights to additional households, enabling broader reach while remaining grounded in observed viewing behavior.
Modeling solves that. In our case, Tunnl takes a set of real households as the seed, the ground truth of who actually does the behavior you care about. Then we use a proprietary method to find every other person across the U.S. population who looks like them behaviorally. Our vast identity graph with more than 23,000+ proprietary data points helps inform these models and drive their output accuracy.
This is how Inscape Audiences powered by Tunnl work: VIZIO's Inscape data supplies observed viewing behavior from real households; Tunnl's modeling extends that signal across the full population, then scores every adult individually against it.
The result is a finite list of real, identifiable people. You get the scale a campaign needs without giving up the integrity that made the original data worth using. Currently, Tunnl has syndicated 91 of these audiences built from this seed data, and is building custom ACR audiences by request in partnership with Inscape.
→ Why it matters: A seed without modeling can sometimes be too small for you to reach your goals. And modeling without a strong set of input data is guesswork. You need both
3. How is a modeled audience actually different from a normal segment?
This is the question worth slowing down on, because "audience segment" has become a catch-all term for products that work in completely different ways.
Traditional segmentation tells you what share of a population fits a profile. Tunnl identifies which specific individuals do. Here's what sits behind that distinction:
Every U.S. adult is represented as a 256-dimensional behavioral model
Rather than a handful of demographic buckets, each person is described by a deep representation trained on thousands of data points. This is part of a stack of 23,000+ predictive attributes per individual built over more than a decade.
Every seed set gets its own purpose-built model
Your data isn't scored against a generic, off-the-shelf classifier. The behavior you supply (think, the ad people watched) becomes the label that trains a model specific to your audience.
Hundreds of model permutations get tested automatically
The system compares them and selects the best performer based on cross-validated results. This means the model shipped is the one that actually predicted best, not the first one that ran.
Models are trained on real responses
Verified behavior from real people. No synthetic data standing in for the thing you're trying to predict.
Audiences are deterministic intersections, not estimates
A Tunnl audience is the intersection of multiple scored signals across the full population. The output is a precise set of people you can name, count, activate, and measure and not a broad approximation.
→ Why it matters: When an audience is a list of scored individuals, everything downstream gets more honest. Reach numbers mean something. Measurement has something real to measure against.
4. Where can these audiences actually be used?
An audience that only works in one place isn't much of an audience.
These are built to be pushed live across the digital ecosystem and syndicated to the platforms buyers already use — CTV, programmatic, social, and addressable environments — so the same definition travels with the campaign instead of being rebuilt, and degraded, at every stop.
Right now, you can find these audiences publicly available for use here: LiveRamp, The Trade Desk, DV360, Simpli.fi, StackAdapt, Freewheel, Beeswax, Basis, Zeta, and Magnite.
The workflow runs end to end: bring the data in securely, match it into an identity graph, model it, output the audience, and distribute it to market. Today more than 600+ users log into the Tunnl Platform for research, insight, activation, and measurement, and 8,800+ advertisers have run campaigns against Tunnl audiences in the marketplace.
→ Why it matters: Consistency. When the audience definition is identical in your CTV buy and your programmatic buy, cross-channel measurement stops being a reconciliation exercise.
5. What kinds of advertisers get the most value out of this?
Any advertiser whose target is defined by behavior or belief rather than age and zip code — which, increasingly, is most of them. A few clear cases:
Political and advocacy campaigns
These teams live and die by precision. The people who need to hear a message are defined by issue position, turnout likelihood, and potential persuadability and rarely by demographics alone. Being able to reach households that saw a specific ad by a candidate or on an issue, and then extend that group to everyone nationally who matches the same profile, is a meaningful advantage in a short window with a fixed budget.
Marketers focused on health and wellness
Condition-relevant audiences have to be built carefully and reached at scale, across a fragmented set of channels, inside a strict regulatory environment. Alternatively, modeled audiences grounded in verified viewing behavior can give healthcare marketers smart reach without risk of targeting sensitive data.
Brands and agencies with a competitive problem
If you know which households have been exposed to a competitor's campaign, competitive audience targeting becomes more actionable. The same applies to category entry points, seasonal timing, and win-back opportunities.
Corporate reputation and public affairs teams
When the goal is moving a specific set of stakeholders rather than maximizing impressions, the ability to name the audience matters more than the ability to buy cheap reach.
→ Why it matters: The common thread isn't the vertical. It's whether your success depends on reaching the right people rather than a lot of people.
The short version
ACR data gives you an honest picture of what actually played on real screens. Modeling extends those insights to additional households. Syndication puts it everywhere you buy.
Miss any one of the three and you're back to estimating.
Want to see what this looks like against your own campaign? Talk to the Tunnl team →