How to Build an Addressable TV Audience: A Step-by-Step Walkthrough

Addressable TV audiences often fail quietly. The segment shows a healthy size in the DSP, the flight delivers, and nobody can say afterward how many of the intended households were actually on the impression log. The gap usually opens in the build, long before bidding starts, because each stage hands the next one a file that looks complete and is not. The method below treats the audience as an engineered asset with a defined input, a tracked transformation at every stage, and a reconciliation at the end.

Step 1: Write the Audience Logic Before You Touch Data

Start with a one-page audience specification. It should name the seed signal (viewing behavior, a CRM list, purchase or location attributes, or a mix), the unit you are targeting, the recency window, the exclusions, and the use each input is licensed for. Write it in plain sentences a planner and a data engineer can both sign.

The unit matters more than buyers expect. Addressable TV is bought and delivered at the household, while many seed sources are person-level or device-level. Decide up front how a person-level signal rolls up. If one adult in a home qualifies, does the whole household qualify? For a category like auto intenders that is usually fine. For a suppression list of existing customers, a loose rollup can suppress the neighbors on a shared network.

• Seed signal: where membership comes from, and how fresh it is when the build runs.
• Household rule: how individual or device signals roll up to a household decision.
• Recency: the window that keeps a household in the audience, and the event that removes it.
• Exclusions: current customers, recent converters, and any sensitive categories you will not build against.
• Permitted use: activation, measurement, or both, recorded per input source.

That last line is the one teams skip. Permitted use travels with each input, not with the finished segment, so a list cleared only for measurement cannot quietly become a targeting seed. The FTC's privacy and security guidance for businesses is a practical reference for documenting those limits internally before the audience exists.

Step 2: Resolve Inputs to the Household

Your inputs arrive keyed in different ways. Viewing signals from ACR come keyed to a pseudonymous TV device ID. CRM files arrive as hashed emails or postal addresses. Mobile signals arrive as advertising IDs. Identity resolution is the work of placing all of them in the same household so the audience logic from step 1 can be applied once, consistently.

A household graph does that grouping, and its quality decides the ceiling for everything downstream. On the viewing side, GSDSI's CTV/Smart TV ACR feed covers ~13-14M unique CTV IDs per month at household-level granularity, keyed to a pseudonymous device identifier with content and exposure events attached. Those device IDs become household members once the graph links them to the other keys in your file.

Build a crosswalk table as you go. Each row should record the input key, the household it resolved to, the link type that joined them (deterministic, such as a shared login or address, or inferred, such as a shared network), a confidence tier, and the date of the link. The crosswalk gives you two things later. You can rerun the build with only high-confidence links if a campaign needs precision, and you can explain any household that a reviewer questions.

Expect attrition here and record it. Some CRM records will not resolve at all, and some ACR devices will sit in households with no other known key. Log the count that entered, the count that resolved, and the count that resolved at each confidence tier. That three-line funnel becomes the first row of your validation report.

Step 3: Map Households to Inventory You Can Actually Buy

A resolved household is not yet a reachable one. Addressable TV inventory is fragmented across smart TV platforms, streaming apps sold through supply partners, and set-top addressable linear from TV providers. Each surface recognizes households through its own identifiers, and each covers a different share of homes. A household that resolves cleanly in your graph can still be invisible on the platforms in your plan.

So map before you size. For each platform or partner you intend to buy, estimate which resolved households it can recognize, then build an overlap view: reachable on one surface, reachable on several, reachable on none. In open programmatic buying, household and device signals travel in the bid request through the device and user objects defined in IAB Tech Lab's OpenRTB specification, so ask each supply partner which of those fields it populates on CTV inventory and how consistently.

The households reachable on no surface are not wasted. They tell you where the plan is thin and whether a different partner mix, or a modeled extension, is worth testing. But keep them out of the reach forecast. Forecasting reach from resolved households instead of reachable ones is a common reason an addressable plan overpromises.

Step 4: Activate Without Losing the Keys

Activation is a chain of key translations. Your household file is onboarded to a DSP, a publisher, a clean room, or a TV provider's addressable platform, and each one matches your keys to its own. Most hops drop some households. Three hops at modest loss each can leave a segment much smaller than the reachable count from step 3, and the platform will only show you the final number.

1. Freeze one audience file of record, versioned and dated, and send that exact file to every destination.
2. Request the matched count from each destination and log it against the version you sent.
3. Use one naming convention that encodes the audience, version, and build date, so a segment in any platform traces back to its file.
4. Set a refresh schedule tied to the recency window from step 1, and confirm each destination replaces membership rather than appending to it.
5. Apply frequency caps at the household level wherever the platform supports it, since that is the unit you built.

How the audience then interacts with bid logic, recency conditioning, and suppression at auction time is a separate layer. Our guide to how ACR fits into DSP bidding and measurement covers that side, so this walkthrough stops at a clean, traceable segment in each destination.

Step 5: Validate Reach Against the File You Started With

Validation closes the loop the earlier steps opened. Before launch, assemble the funnel: households in the specification, households resolved, households reachable, households matched at each destination. After the flight, add the last row, households actually delivered an impression, using exposure data keyed back to the same household IDs.

Compare that delivered set against the audience file of record, not against the segment counts the platforms reported. Three readings matter. Delivered households that were never in your file point to a matching or onboarding error. Audience households that were reachable but never delivered point to pacing, frequency, or bid issues. A large drop between resolved and matched points to a weak link in your identity or onboarding chain, which you can isolate because you logged every hop.

Where reach figures will be reported upward or used to settle with a partner, check whether the measurement source holds Media Rating Council accreditation for the metric in question. If the next question is whether the campaign changed outcomes, that requires a control design, which our article on CTV measurement test design walks through.

Making the Build Repeatable

The first build is slow because every decision is new. The second should reuse the specification template, the crosswalk schema, the naming convention, and the funnel report, changing only the audience logic. Teams that keep those artifacts find that addressable TV data stops being a black box and becomes an asset they can audit, compare across partners, and improve flight over flight. For how GSDSI supports media buyers and platforms across this workflow, see our AdTech industry page, or request an ACR sample and run it through step 2 against your own file.