A third-party measurement partner for a CTV agency engagement is a vendor, paid by neither the streaming service nor the buying platform, that counts what was delivered and estimates what it caused. Buyers ask for one because every other number in a connected TV report comes from a party with something to sell: the publisher reports its own impressions, the demand-side platform reports its own conversions, and the agency assembles both into a deck about its own work.
The difficulty is that "measurement partner" is not one product. It is five different jobs, done by five kinds of company, priced in different ways, and several of them do not function below a certain spend. This guide sets out the five jobs, the vendors that do each, what they cost, where a luxury or high-consideration brand should start, and who should sign the contract.
Why a CTV Agency Needs a Third-Party Measurement Partner
Three sources of bias sit inside an unverified CTV report.
- Platform attribution counts what it can see. A buying platform credits itself for any conversion from a household it served, inside a window it chose. Households that were always going to buy are counted alongside those the advertising persuaded. For a brand with strong existing demand, the platform figure can overstate caused sales by a factor of two to five.
- Each platform deduplicates only itself. Three streaming services each report 400,000 households reached. The combined figure may be 1.2 million or 550,000, and nobody inside any one platform can say which.
- The agency is graded on the report it writes. This does not require bad faith. An agency choosing which window, which baseline, and which metric to present will tend to choose the flattering ones.
A measurement partner is the control on all three. It is worth paying for only where it changes a decision, which is the test to apply to every line below.
The Five Jobs a Third-Party Measurement Partner Does for CTV
| Job | Question answered | Example vendors | Typical pricing | Works from |
|---|---|---|---|---|
| Verification | Did the ad run on a real TV, in a real app, in a suitable program? | DoubleVerify, Integral Ad Science, HUMAN, Pixalate | $0.10 to $0.40 CPM | Any spend |
| Reach and frequency | How many distinct households were reached across services, and how often? | Nielsen ONE, iSpot, VideoAmp, Comscore, Samba TV, Innovid | $0.20 to $1.00 CPM or a flat platform fee | About $75,000 a month |
| Brand lift | Did awareness, consideration, or intent move among exposed viewers? | Upwave, DISQO, Kantar, Cint | $10,000 to $40,000 per study | 3 to 5 million impressions per study |
| Outcome attribution | Which exposed households visited the site, inquired, or bought? | iSpot, Innovid, EDO, Affinity Solutions, Foursquare | $0.25 to $1.25 CPM or flat fee | About $75,000 a month |
| Incrementality | How many of those outcomes would not have happened without the advertising? | Measured, Haus, Recast, or in-house tests using open-source tools | $5,000 to $25,000 a month for a platform; analyst time for in-house | About $50,000 a month in one channel |
Prices are Stillwater planning ranges for US campaigns and vary with contract length and volume. Vendors change their products often, so treat the names as categories to shortlist from, not as rankings.
1. Verification
Verification vendors check that an impression was served to a television device and not a server imitating one, that the app is what it claimed to be, and, where the publisher passes the information, what kind of program carried the ad. On well-chosen private marketplace inventory, invalid traffic should be below 1 percent. On open-exchange CTV it has repeatedly been measured in the high single digits and above. Verification is cheap, works at any budget, and should be on every campaign.
Its limit on CTV is transparency. Many streaming services pass only a genre or a rating, not the program title, so "brand suitability" is often verified at channel level only.
2. Reach and frequency
These vendors combine automatic content recognition data from smart TVs, set-top-box data, and panels to estimate how many distinct households a campaign reached across services. This is the measurement that tells a brand it reached 40 percent of its audience fourteen times and the other 60 percent not at all.
For a luxury advertiser the caution is coverage. Recognition data comes from particular television manufacturers, and panels are built to represent the general population. A target of 150,000 wealthy households may be represented by a few thousand measured homes. Ask for the measured sample inside your audience, not the national one.
3. Brand lift
A brand lift study surveys people who were exposed to the campaign and a matched group who were not, and compares answers. It is the only one of the five that measures a change in what people think.
It is also the one most often bought at a size too small to work. The arithmetic appears below.
4. Outcome attribution
Outcome vendors match exposed households to later behavior: a website visit from the same home network, a search for the brand, a store or showroom visit, or a card transaction. This is independent of the buying platform, which is useful, though it is still correlation. A matched visit shows that an exposed household acted. It does not show that the ad was the reason.
5. Incrementality
Incrementality measurement creates or finds a comparison group that did not see the advertising and reads the difference. The usual method for CTV is a geographic experiment: advertising runs in some markets and is withheld from matched markets for six to twelve weeks. Specialist platforms automate the design and analysis. A capable analyst can run the same test with open-source tools such as Meta's GeoLift or Google's CausalImpact, and build a media mix model with Google's Meridian or Meta's Robyn.
This is the job that answers the question a finance director is asking, and the one a brand spending $50,000 to $200,000 a month should fund first after verification.
What Does a Third-Party Measurement Partner for CTV Cost?
A workable rule is to spend 4 to 8 percent of media on independent measurement. Below 3 percent the brand is usually trusting platform numbers. Above 10 percent, measurement is taking money from the reach needed to produce a measurable effect.
| Monthly CTV investment | Sensible measurement stack | Monthly measurement cost | Share of media |
|---|---|---|---|
| $50,000 | Verification; platform reach reports; one geographic holdout per half-year run with open-source tools | $2,000 to $4,000 | 4% to 8% |
| $100,000 to $150,000 | Add independent reach or outcome attribution; one brand lift study per half-year | $5,000 to $10,000 | 4% to 7% |
| $250,000 | Add an incrementality platform or continuous geographic testing; brand lift each quarter | $10,000 to $18,000 | 4% to 7% |
| $500,000 and above | All five, plus a media mix model refreshed quarterly | $20,000 to $35,000 | 4% to 7% |
Worked example at $150,000 a month
At a $42 blended CPM, $150,000 buys about 3.57 million impressions a month.
| Line | Basis | Monthly cost |
|---|---|---|
| Verification | $0.25 CPM on 3.57 million impressions | $893 |
| Reach and outcome measurement | $0.60 CPM on 3.57 million impressions | $2,143 |
| Brand lift | One $24,000 study per half-year | $4,000 |
| Geographic holdout | Design, analysis and tooling | $2,500 |
| Total | $9,536 (6.4% of media) |
The holdout carries a second cost that does not appear on an invoice: advertising withheld from 10 to 20 percent of markets for the test period. That is the price of knowing, and it is recovered many times over if the result moves budget away from a channel that was not working.
Why Most Small CTV Brand Lift Studies Fail
A brand lift study is a comparison of two percentages, and small differences need large samples.
Suppose 20 percent of the control group say they would consider the brand. To detect a rise to 23 percent, with 90 percent confidence and an 80 percent chance of finding a real effect, the study needs about 2,300 completed surveys in each group. To detect a rise to 25 percent it needs about 860 in each.
Television is not clickable, so surveys reach exposed viewers later, on a phone or laptop linked to the same household. Only a small fraction of exposed households can be matched, invited, and persuaded to answer. This is why vendors set minimums of several million impressions, and why a study run on 800,000 impressions returns "no significant lift" whether or not the campaign worked.
For a luxury brand there is a further problem. The study reports lift among survey respondents, who skew toward people willing to take surveys for small rewards. Ask what share of respondents fall inside your income or net-worth definition. If the answer is 6 percent, the study describes a different audience.
Three practical rules follow.
- Do not commission brand lift below roughly 3 million impressions in the study window.
- Choose one primary question in advance. Testing five metrics and reporting the one that moved is not evidence.
- Where the audience is too small for a survey, use branded search volume and direct site traffic in exposed versus withheld markets as the brand measure.
What a Third-Party Measurement Partner Cannot Fix for High-Consideration Brands
Measurement vendors were built around brands with thousands of weekly conversions. A private aviation company, a wealth manager, or a developer selling 60 residences does not have them. Four limits apply.
- Too few sales to read. A test needs events. If a brand closes 15 sales a month, no eight-week test will detect a 15 percent change in sales. The test has to be read on an earlier, more frequent signal that the brand has shown to predict sales: qualified inquiries, consultations booked, configured quotes.
- Sales happen offline. Outcome vendors see web visits and card transactions. They do not see a signed fractional-ownership contract or a wire transfer. The brand's own records have to be joined to exposure data, usually through a hashed match.
- Long cycles outlast the window. A 30-day attribution window captures almost nothing for a purchase that takes five months. Tests need a post-period, and the patience to read it.
- The audience is thin in every panel. Any method relying on a national panel will have few affluent households in it. Methods based on the brand's own customer file and on geography hold up better.
None of this argues against independent measurement. It argues for choosing the methods that survive low volume: verification, geographic experiments read on a validated early signal, and matching to the brand's own sales records.
Who Should Hold the Contract: the Brand or the CTV Agency?
| Arrangement | Advantages | Risks | Suited to |
|---|---|---|---|
| Agency holds the contract and controls the login | Fast to start; agency volume pricing | The measured party controls the measurement; data leaves with the agency | Verification only |
| Agency holds the contract; brand has full login and data export rights | Volume pricing with visibility | Depends on the agency contract being written properly | Reach, outcome attribution, brand lift |
| Brand holds the contract; agency is a user | Independence; history survives a change of agency | Higher price at low volume; brand needs someone to manage it | Incrementality and media mix modeling |
The principle is simple. The closer a measurement is to answering "did the agency's work pay off," the more it matters that the brand owns it. Verification can sit with the agency. The incrementality test design, the raw results, and the model should belong to the brand.
Whatever the arrangement, the agency agreement should state that the brand owns all measurement data, has direct access to the vendor's reporting, and receives test designs in writing before a test begins.
How to Evaluate a Third-Party Measurement Partner for CTV
Ten questions separate a useful vendor from an expensive dashboard.
- Which of the five jobs does the product do, and which does it not?
- Does the vendor sell media, or take revenue from any platform it measures?
- How many households inside our target audience are in the measured sample?
- Which streaming services allow your tags or share data with you, and which do not?
- Is the relevant product accredited by the Media Rating Council, and for which metrics?
- What is the smallest effect the method can detect at our spend, stated before the test?
- Can results be read on our own sales records, and how are they matched?
- Do we receive the underlying data, or only a report?
- What is the minimum term, and what is the fee if spend falls?
- Who designs the test: your analysts, our agency, or software?
Question six matters most. A vendor that cannot state the minimum detectable effect in advance is selling a report, not a test.
Common Mistakes in Choosing a CTV Measurement Partner
- Buying brand lift first because it is familiar, at a size that cannot produce a result.
- Treating outcome attribution as incrementality. A matched site visit is not a caused site visit.
- Accepting the buying platform's built-in lift study as independent. It may be well designed. It is still the seller's study.
- Paying for cross-platform reach on a two-service plan, where the overlap can be estimated well enough from the platforms themselves.
- Running one test and treating it as permanent. Results decay as creative, audiences, and prices change. Retest at least twice a year.
- Letting the agency choose the holdout markets after seeing the data. Markets are selected and written down before launch.
- Ignoring the cost of data that will never be used. A reach report nobody acts on is a fee, not a measurement.
A 60-Day Sequence for Setting Up Independent CTV Measurement
- Days 1 to 10: Write down the two or three decisions measurement must inform, such as whether to raise CTV from 30 to 45 percent of budget. Confirm data ownership terms with the agency.
- Days 10 to 20: Turn on verification across all CTV buys. Identify the early signal that predicts sales and confirm it is recorded cleanly.
- Days 20 to 35: Design the geographic holdout: matched markets, duration, expected minimum detectable effect, and the decision that follows from each possible result.
- Days 35 to 45: Launch. Add reach or outcome measurement if spend supports it.
- Days 45 to 60: First verification and reach read. Correct frequency and supply. The holdout continues untouched until its planned end date.
How Stillwater Media Approaches Third-Party Measurement for CTV
Stillwater Media plans and buys premium CTV and programmatic media for luxury and high-consideration brands, where customer value exceeds $5,000 and the decision takes more than 30 days. Those are the conditions under which platform-reported results are least reliable, so incrementality testing is built into our engagements from the start. We are not tied to one buying platform or one measurement vendor, we design holdouts before launch and put the design in writing, and we recommend that clients own their measurement data and their test results.
We take on a limited number of engagements each quarter. If you want an independent read on what your CTV investment is producing, apply to work with Stillwater Media.
About Stillwater Media
Stillwater Media is a selective performance media agency for luxury and high-consideration brands. We plan and buy premium CTV on services including Disney+, Netflix, and Prime Video, along with programmatic video and display, digital out-of-home, streaming audio, YouTube Select, podcasts, native, and social, using private marketplace deals, affluent audience engineering, and incrementality testing. We work with brands in private aviation, luxury real estate, wealth management, private clubs, luxury automotive, premium DTC, and luxury hospitality. Signal. Strategy. Scale.
Frequently Asked Questions
What is a third-party measurement partner for CTV?
It is an independent vendor, paid by neither the streaming service nor the buying platform, that verifies where connected TV ads ran and estimates what they caused. The term covers five different jobs: verification, reach and frequency, brand lift, outcome attribution, and incrementality.
How much does third-party CTV measurement cost?
A reasonable budget is 4 to 8 percent of media. At $150,000 a month in CTV, that is roughly $6,000 to $12,000 a month, covering verification, independent reach or outcome measurement, a brand lift study each half-year, and a geographic holdout test.
Should the brand or the agency hold the measurement contract?
Verification can sit with the agency, since it is inexpensive and operational. Incrementality testing and media mix modeling should be owned by the brand, because they grade the agency's work and the history should survive a change of agency. In every case the brand should own the data and have direct access to reporting.
What is the minimum spend for a CTV brand lift study?
Most studies need about 3 to 5 million impressions in the study window to collect enough survey responses. Detecting a three-point rise from a 20 percent baseline takes about 2,300 completed surveys in each group, and only a small fraction of exposed TV households can be reached with a survey.
Is outcome attribution the same as incrementality?
No. Outcome attribution shows that a household exposed to an ad later visited a site or made a purchase. Incrementality compares exposed households or markets with a group that was not exposed, and reports only the outcomes the advertising caused.
