How CTV Measurement Providers Differ on Reach, Frequency, and Device Deduplication
How CTV measurement providers differ in reporting reach, frequency, and device deduplication, with panel, ACR, server data, and identity graph methods.
The short answer
CTV measurement providers differ mainly in what they observe. Panels measure people and project, ACR providers see what a set of smart TVs displayed, census or server data counts every served impression by device, and identity graphs stitch devices into households. Those choices drive how each one counts reach, how it caps frequency, and how much of its device deduplication is observed rather than modeled.
If you have ever put two CTV reach reports side by side and found they disagree by a third, nobody necessarily made a mistake. The providers are measuring different things with different blind spots. The useful skill is knowing which question each method answers well, so you can pick a currency and read the numbers with the right amount of trust.
The four main approaches CTV measurement providers use
Almost every provider combines more than one of these, but each starts from a primary data source.
| Approach | What it observes | Reach unit | How it deduplicates devices | Main blind spot |
|---|---|---|---|---|
| Panel (people meters, opt-in households) | What a recruited sample watched, by person | Persons, projected to population | Directly, inside the panel home | Small samples for niche audiences and small campaigns |
| ACR (automatic content recognition on smart TVs) | What appeared on screen for opted-in TVs from certain makers | TV devices, then households | Links TVs to other devices through IP or graph | Only covers TVs from manufacturers that share data, and not every input or app |
| Census or server data (ad server, SSP, publisher logs) | Every impression served, with device and IP | Devices or IPs, converted to households | Through IP and device graphs | Shared and changing IPs, no view confirmation |
| Identity graph overlay | Linkages between devices, IPs, and addresses | Households or modeled persons | This is its main job | Graph errors and decay, much of it modeled |
Widely known names sit in different places on this map. Nielsen is historically panel-led and has been adding big data. iSpot, VideoAmp, and Samba TV are generally associated with large ACR or set-top box datasets combined with other inputs. Each company's exact method changes often and differs by product, so ask them directly rather than relying on a category label. I would not choose based on brand reputation alone.
How reach gets counted differently
Reach sounds simple: unique households or people who saw the ad at least once. Three choices change the number.
The unit. A household-based provider and a person-based provider will not agree, because one household can contain three persons reached. Make sure every report says which unit it uses.
The universe. Reach as a percentage needs a denominator. Some providers use all US TV households, others use all streaming households, others use their own observed footprint. A campaign can reach 8 percent of US TV households and 25 percent of a provider's ACR footprint at the same time.
Projection. ACR and panel data cover a slice of the population. Providers weight and project to the full universe. The weighting model is where a lot of difference lives, and it is rarely shown in the report. MRC standards distinguish measured from modeled data, which is a good frame for asking about it.
How frequency gets counted differently
Frequency is impressions divided by reach, so any reach error shows up inversely in frequency. Undercount reach and frequency looks high. Overcount reach (for example by treating each IP as a separate household) and frequency looks low and safe when it is not.
Distribution matters more than the average. A campaign at an average frequency of 4 could have half its households at 1 and a long tail at 20 or more. Ask for the frequency distribution, not just the mean. The guide to interpreting reach, frequency, and duplication together covers how to read those curves, and CTV frequency management for pharma covers what to do with them.
Device deduplication: where the reports really split
A household might have a smart TV, a streaming stick on a second TV, a console, two phones, and a laptop. Without deduplication, an OTT campaign touching four of those devices counts four reached units. With good deduplication it counts one household.
Providers deduplicate in a few ways:
- IP-based. Devices sharing a residential IP during the same window are grouped. Cheap and broad, but IPs change, and apartments, offices, and mobile networks share IPs.
- Graph-based. A household graph links devices using IP history, logins, and address data. Better, but graph quality varies and much of it is probabilistic.
- Login-based. Publishers with authenticated users can deduplicate within their own service very well, but not across services.
- Panel-based. In a metered home, devices are known. Very accurate for the panel, then projected.
The question to ask every provider is how much of the deduplication was observed and how much was modeled. A hypothetical example: a provider reports 1.4 million households reached. Of 2.1 million device-level reaches, 500,000 were collapsed by observed links (shared login or metered home) and 200,000 by modeled graph links. If the modeled links are half wrong, true reach is closer to 1.5 million. That is a 7 percent swing from one assumption. The deduplicated reach guide for CTV and online video uses the same observed versus modeled split.
Questions to ask a CTV measurement provider
- What is your primary data source, and how many households or devices does it observe directly?
- What is the reach unit (household, person, device), and what universe sits in the denominator?
- How do you deduplicate devices, and what share of deduplication is observed versus modeled?
- Which TV manufacturers, apps, and platforms are not in your coverage?
- Which of your metrics are MRC accredited, and for what scope?
- How do you handle impressions you can see in ad logs but not in your ACR or panel data?
- Can you report the frequency distribution, not only the average?
- How do you measure across linear, CTV, and online video in one number, if at all?
The last question matters if the plan runs on several screens, which is covered in measuring reach and frequency across linear TV, CTV, and online video.
Healthcare-specific wrinkles
Healthcare audiences are often small and older. Panel samples get thin fast for a condition audience, and ACR coverage depends on which TV brands older households own. If your target skews toward a specific region or age group, ask the provider for coverage estimates for that group, not just the national footprint. Also check that any health segment the provider uses for audience reporting has gone through the same privacy review as your targeting data. The series guide on CTV and OTT advertising for healthcare covers the broader measurement layers.
Practical takeaway
Pick one provider as your CTV reach currency before the next flight and write down its reach unit, its universe, and its observed versus modeled deduplication share. Put those three items in the footnote of every reach chart so nobody compares that number with a different provider's report as if they were the same measure.
Frequently asked questions
Why do two CTV measurement providers report different reach for the same campaign?
They observe different things. A panel measures a sample of people and projects, ACR sees what a subset of smart TVs displayed, server data counts impressions by device or IP, and identity graphs link those devices into households. Each method has different coverage gaps and deduplication rules, so the totals rarely match.
What is device deduplication in CTV measurement?
It is the process of recognizing that several devices or IP addresses belong to the same household or person so reach is not double counted. A household with a smart TV, a streaming stick, and two phones can look like four reached units without it. Providers differ in how they link devices and how much of the linking is modeled.
Is MRC accreditation enough to choose a CTV measurement provider?
Accreditation tells you an independent audit reviewed specific metrics and methods. It does not mean every metric a provider sells is accredited, and it does not tell you whether the method fits your audience. Ask which exact metrics are accredited and check the scope.
Should a pharma brand use one CTV measurement provider or several?
One provider for the planning and reporting currency keeps numbers consistent over time. A second provider can be useful as a periodic check, but mixing providers in the same report usually creates confusion. Pick the currency before the flight, not after the results.
Sources
External guidance and platform documentation change. Links were current at publication; check them again before relying on them for a decision.
Editorial note. Analysis and frameworks are the author's own and do not represent Acxiom or any current or former employer, client, or named platform. Examples labeled hypothetical or illustrative are not results from real campaigns. Nothing here is legal, regulatory, or medical advice.
New pharma programmatic breakdowns, occasionally
One email when I publish something worth reading. Benchmarks, measurement teardowns, and case studies with the caveats attached. No cadence promises, no reselling your address.
Unsubscribe any time. See the privacy policy.
Working through this decision on a real plan?
I work on health and pharma data, identity, and activation, after five years running HCP and DTC programmatic agency-side. Happy to talk through how this applies to your situation.