CTV Ad Measurement: Three Paths From Exposure to Outcome

CTV ad measurement runs three ways in 2026: MMP matching for app installs, household matching for web actions, and holdout tests. What each one proves.

CTV ad measurement connects a streaming TV impression to something the advertiser counts: an app install, a site visit, a purchase, or a lift over a control group. Three paths do that work in 2026: mobile measurement partners match exposed households to installs, pixels match them to web actions, and holdout or geo tests measure what the campaign added.

Which path an advertiser can run depends on what it already has in place, and the platform matters less than that.

Agency media buyers and CFOs have spent 2026 asking two questions: which CTV platforms prove attribution, and which partners run incrementality tests. Each answer starts in the same place, a usable record of which households saw the ad, when, and where.

What a CTV Impression Records

A streaming TV ad leaves an exposure record rather than a click. The record holds a timestamp, the app or channel, the device type, the creative, and the household IP address, plus a device identifier when the platform passes one.

Matching happens at the household level. Phones, laptops, and the TV usually share one home IP address, so cross-device attribution links a conversion on the phone to an exposure on the TV. Some platforms add an identity resolution graph on top of the IP match. Branch describes its household measurement in exactly those terms: devices that share an IP address. Kochava's October 2025 documentation treats a match as deterministic when device IDs pass and probabilistic when only an IP address and user agent are available.

The record supports three levels of detail:

  • Delivery: which impressions ran, in which apps, at what completion rate
  • Exposure: which households saw the ad, and how often
  • Outcome: which exposed households took an action within a defined window

Delivery reports list impressions, apps, and completion rates. Measurement starts when households are tied to an action within a defined window, and that needs the record to leave the platform or be matched inside it.

Path One: MMP Matching for App Installs

App attribution on CTV requires a mobile measurement partner. AppsFlyer, Adjust, Singular, Kochava, and Branch each support CTV-to-mobile measurement, matching impressions from integrated platforms to installs and in-app events. Adjust and Singular publish their integrated platform lists, and AppsFlyer limits the feature to designated partners.

For CTV-to-mobile matching, AppsFlyer, Adjust, and Singular use probabilistic methods. AppsFlyer's cross-platform attribution documentation, updated October 2026, describes view-through attribution through probabilistic modeling on the IP address. The window runs from 1 to 72 hours, with a default of 24 hours. The platform sending the impression must pass the IP address and user agent.

Singular's CTV documentation, updated September 2026, uses the same household IP logic and a default 24-hour view-through window adjustable from 1 to 72 hours. An optional setting gives a CTV impression equal priority to a click for up to 24 hours, and it is off by default.

Adjust places CTV impressions at the lowest priority in its attribution waterfall by default, so a paid social click in the same window wins the install. Its CTV ad-to-mobile guide lets a team set a priority window of 1 to 6 hours in which the CTV impression outranks other touchpoints. Kochava recommends a 24-hour window for CTV-only impressions. The priority setting, more than the window, decides what the install report shows.

Each integration carries its own settings:

  • Adjust lists The Trade Desk, MNTN, Roku OneView, Tatari, tvScientific, StackAdapt, Disney, Hulu, Paramount+, and Samsung DSP as CTV partners.
  • Singular lists MNTN Performance TV, The Trade Desk, tvScientific, Vibe, Samsung Ads, StackAdapt, Moloco, and Zeta Global.
  • MNTN's AppsFlyer integration, documented August 2026, sets a 24-hour view-through window, 72 hours cross-platform, and six hours of equal priority.
  • Vibe's AppsFlyer integration, documented April 2026, also offers equal priority for up to six hours.

The report then shows installs from exposed households inside the window, under the priority rules the team chose. That is the same claim the MMP makes for every other view-through channel. Comparison with paid social needs the same window and priority on both.

Path Two: Household Matching for Web Outcomes

Brands without an app use a pixel. The platform matches the exposed household's IP address against visits, sign-ups, and purchases recorded on the site, and each platform sets its own window:

  • MNTN's Verified Visits, documented April 2026, use a household identity graph with a default window of 7 days for retargeting and 14 days for prospecting.
  • Roku Ads Manager's event tracking, documented March 2026, applies a 14-day window to view-through conversions reported through a pixel, a Conversions API, or MMP integrations.
  • Vibe's web attribution windows, documented August 2026, run from 12 hours to 30 days, with 30 days as the default.

Three platforms can report three different conversion counts for the same campaign because one counts 14 days and another counts 30. Set one window across platforms before comparing them, and keep it short enough that organic traffic does not fill it.

Exposure data can also come from the TV itself:

  • iSpot ties conversions to TV ad exposures through automatic content recognition and matches online events to households by IP address.
  • Vizio Inscape reports second-by-second recognition from more than 20 million opted-in smart TVs.
  • Samba TV deduplicates reach and frequency at the household and person level across CTV, digital, and linear.

These sources record what the household watched, including ads the advertiser bought elsewhere. That makes them the basis of TV attribution for linear campaigns, and useful for frequency and for linear-plus-streaming reporting.

The result is a visit, sign-up, or purchase from an exposed household within the window. A matched visit followed exposure, and the third path tests whether exposure caused it.

Path Three: Holdouts and Geo Tests

Incrementality answers the causal question by withholding ads from a control group. Two designs are in use: household holdouts split matched households into exposed and unexposed groups inside the platform, and geo tests split regions. Haus's GeoLift assigns test and control regions by random stratified sampling and lists CTV among eligible channels.

Published test designs set the time and budget:

  • MNTN's Haus integration guide, updated April 2026, calls for a test flight of 4 to 6 weeks at minimum, a post-treatment window of 2 to 4 weeks, and a target power score of at least 90 percent.
  • Measured's TV and CTV incrementality FAQ, published August 2026, says tests typically run four to eight weeks with test markets covering about 7 to 40 percent of national sales.
  • Jamloop added household holdouts and geo lift through Recast in July 2026 with a 30-day measurement window.
  • INCRMNTAL measures without running experiments and reports within five days.

In Haus's 2026 Marketing Decision Confidence Index, reported by eMarketer in May 2026, 60 percent of 500 US marketing decision makers rated independent incrementality testing the most trusted measurement approach.

The output is the number of outcomes the campaign added over the control. No other path separates a matched conversion from one that would have happened anyway.

Delivery Reports Are Not Measurement

A completion rate shows that the file played to the end, and reach and frequency describe how many households saw it. Outcome pricing is replacing completion rate as the number that anchors a plan.

Delivery reports also need to be cleaned before matching. Integral Ad Science's 21st Media Quality Report, published July 2026, covers the second quarter of 2026. It puts CTV invalid traffic at 9.1 percent for non-optimized campaigns and 0.1 percent for optimized ones. DoubleVerify's May 2026 global study found fraud rates under 1 percent in protected CTV campaigns against nearly 9 percent unprotected. It also counted 140 percent more CTV fraud schemes in the first quarter of 2026 than a year earlier. An exposure record that includes invalid impressions inflates every match built on it.

The IAB's 2026 Digital Video Ad Spend and Strategy report, published July 2026, asked 360 respondents what they would pay extra for. Advanced measurement or attribution drew 39 percent, and fraud detection drew 31 percent. Among the 77 buyers who lack confidence in open exchange inventory, fraud and invalid traffic was the top reason at 56 percent.

In April 2026 the Media Rating Council accredited DoubleVerify for CTV impressions with sophisticated invalid traffic filtration. Its January 2026 auction transparency standards strongly encourage the Open Measurement SDK for CTV apps. IAB Tech Lab's OM SDK page, updated November 2025, lists version 1.5 with Samsung and LG TV support. In July 2026 Nielsen withdrew Nielsen One Ads from the MRC accreditation process, and the MRC said the service was removed from in-process status due to planned material methodological changes.

What to Ask Before the Budget Moves

Put six questions to every platform on the shortlist:

  • Exposure export: can impression-level records leave the platform, and how often? MNTN's data export guide, updated May 2026, states that its API and CSV exports are daily aggregates, not event-level, with impression-level data available only through impression tracking.
  • MMP integration: is your MMP named as an integrated partner, and what window and priority does the integration set?
  • Matching method: household IP, device identifier, or identity graph, stated in writing.
  • Window: the default, the range, and whether the same window can be applied across every platform in the plan.
  • Traffic quality handling: whether invalid traffic is filtered before the bid as well as after delivery, and whether impressions later flagged as invalid are removed from the bill.
  • Holdout support: a built-in holdout, such as MNTN's incrementality dashboard or tvScientific's built-in incrementality testing, or an export clean enough for Haus, Measured, or INCRMNTAL.

Roku Ads Manager, per its February 2026 guide, and Disney Campaign Manager start at $500 per campaign, and Amazon recommends $10,000 per self-service streaming TV campaign. A four-week geo test needs enough spend in the test regions to clear a power threshold. Minimums and measurement features for ten platforms are compared in 10 Best Self-Serve CTV Advertising Platforms in 2026, and CPM ranges in How Much Does a TV Commercial Cost in 2026.

Who Measures What

Five groups of companies cover the three paths, and a plan usually needs two or three of them:

  • Verification and media quality: DoubleVerify and Integral Ad Science measure invalid traffic and viewability on CTV, and both publish CTV fraud rates. DoubleVerify's MRC accreditation, granted April 2026, covers CTV impressions with sophisticated invalid traffic filtration.
  • Exposure and audience data: iSpot and Vizio Inscape record what households watched through ACR, including ads bought on other platforms. iSpot matches those exposures to online events by household, and Samba TV deduplicates reach and frequency across CTV, digital, and linear.
  • Mobile measurement partners: AppsFlyer, Adjust, Singular, Kochava, and Branch match CTV impressions to installs and in-app events. They are the only path for app outcomes, and Adjust and Singular publish which CTV platforms they integrate.
  • Incrementality: Haus, Measured, and Recast run geo and holdout tests. INCRMNTAL models lift without experiments.
  • Buying platforms with built-in measurement: MNTN, Roku Ads Manager, and Vibe match exposures to pixels or MMP events inside the platform. tvScientific describes household-level attribution with incrementality testing and has sold cost-per-outcome buying since 2023. Amazon Marketing Cloud exposes streaming TV event-level signals, and Disney Campaign Manager has accepted third-party pixels since October 2024. General programmatic CTV DSPs sit beside them for teams that buy several channels at once.

The ANA reported in February 2026 that CTV made up about 40 percent of programmatic spend in the fourth quarter of 2025. In a survey of about 95 marketers published the same month, 53 percent said they struggle to unify measurement across channels.

How Starti Measures CTV

Starti AI Dash runs pixel and MMP matching inside the buying platform. Reporting connects CTV exposure to supported web pixels and, for app outcomes, the advertiser's MMP. Reported outcomes are visits, installs, registrations, purchases, and revenue.

TrustLayer screens inventory before the bid and analyzes viewability and exposure patterns after delivery. Households with problematic patterns are blocked from future delivery, and their earlier impressions are excluded from billing. Placement reporting shows the shows, episodes, and device types where ads ran. Pricing can be tied to results rather than impressions.

Book a demo to map which of the three paths your current stack supports and what a first campaign would report.

Frequently Asked Questions

How is CTV advertising measured?

By matching the household that saw the ad to an outcome. Mobile measurement partners match impressions to app installs, pixels match them to site actions, and holdout or geo tests compare exposed groups with a control to measure lift.

What is CTV attribution?

CTV attribution assigns credit for an install, visit, or purchase to a streaming TV impression that the same household saw within a defined window. The match uses the household IP address, a device identifier, or an identity graph.

Do I need an MMP for CTV measurement?

For app installs and in-app events, yes. AppsFlyer, Adjust, Singular, Kochava, and Branch each support CTV-to-mobile measurement. Web outcomes can be measured with a pixel instead.

How long should a CTV attribution window be?

AppsFlyer and Singular default to 24 hours for view-through CTV impressions, adjustable to 72 hours. Web platforms default to 7, 14, or 30 days. Use one window across platforms.

What is the best way to measure CTV incrementality?

A holdout or geo test with a control group. Current vendor guidance calls for four to eight weeks of flight, a power score target, and test regions large enough to detect the expected lift.