Decide what counts as proof before your CTV campaign launches

I have spent more than twenty years scaling consumer products and watched channels move from curiosity to a core part of acquisition. CTV may be next. But when we tested it at IMVU, I saw how dramatically the same campaign could look different depending on the rules used to assign credit.

CTV has become a significant advertising channel. In the US, streaming accounted for 49.6 per cent of television viewing time in August 2026, according to Nielsen. Across Europe, excluding Russia, 187 million households had a smart TV in 2025, according to Dataxis. As CTV's reach grows, so does the pressure on advertisers to demonstrate that campaigns generate incremental customers and revenue, rather than simply claiming credit for conversions that would have happened anyway.

Smadex and Adjust have published a CTV Performance Playbook, presented at CTV Live! in San Francisco, where I was a guest. It is vendor-published, and I read it that way. It describes the measurement problem clearly and provides a useful comparison with the test I ran at IMVU. I will distinguish between the playbook's recommendations and what we observed ourselves.

04. What we ran at IMVU

At IMVU, we tested CTV as part of our user acquisition strategy. We ran it through a CTV agency over about four to six weeks, on a budget of around $100,000. The agency gave us access to channels and audience segments we could not have assembled alone. As I recall, the design was geographic: some markets saw CTV while others were held dark, giving us a comparison group rather than relying only on a before-and-after view. This was more than six years ago, and I do not want to overstate how the markets were chosen. I would call it a directional incrementality test, not a lab-grade experiment.

We judged the test on the metrics that mattered to the business: cost per install, cost per registration and cost per paying customer. Install cost alone is a weak measure. An install is not a customer, and a channel can make installs cheap while bringing in people who never pay.

We also looked beyond last-click and view-through reporting. Comparing exposed and held-out markets, we examined direct traffic, branded search across web and mobile, and answers to a “How did you hear about us?” question. Each offered a different signal of an effect that click reporting could miss. Someone who sees an ad on television and later types a brand name into a search bar will not appear as a CTV click.

The exposed markets showed the clearest signals in branded search and direct traffic, along with more registrations and new paying customers than the held-out markets. I no longer have the original figures, so I will describe the lift as noticeable rather than attach percentages I cannot back up. The attribution methods also told different stories about the same test. Measured by last-touch alone, CTV looked expensive against our acquisition targets. When we used a multi-touch approach and counted its indirect contribution to acquisition, the economics looked more viable and fell within our allowable acquisition costs.

I do not read that as a victory lap. A model that credits more interactions will give CTV more opportunities to claim credit, so a favourable multi-touch number deserves a second look. That is why we considered it alongside the holdout comparison and the other signals rather than trusting any one view. They pointed the same way, which gave us enough confidence to act, but the test was directional rather than conclusive, and not precise enough to put a firm number on the lift.

We scaled CTV. The bigger lesson was that it was creating demand that last-touch was not capturing. Instead of running it always-on, we concentrated spend around peak seasonal periods, when we expected the incremental impact to be greatest. The lesson I took was not simply that CTV works. It was that the answer depended on measurement rules we had to choose deliberately, and from then on we judged CTV by incremental impact, not by what the attribution dashboard reported.

05. What else the test taught us about running it

Creative was the hardest part. With stakeholders across marketing and the executive team, we had to balance direct-response creative with brand storytelling. Those goals pulled the spots in different directions. Production costs also limited how many variations we could afford to test.

In our test, 15- and 30-second spots performed better than 60-second ads, and frequency, how often an audience sees the ad, mattered more than we expected. Those are findings from one test, not universal rules. They point to a wider problem: if the creative is weak or the frequency is wrong, the channel can look like it failed when the real problem is the execution.

The playbook warns against mistaking poor creative for poor media. Brooks made a related point: creative plays an outsized role in CTV, he said, because the ad is often a 30-second, unskippable video with sight, sound and motion, and weak creative stands out alongside programmes with high production values.

01. The ad runs on one screen and the action happens on another

Most CTV ads cannot be clicked. Shoppable and interactive formats exist, including on-screen prompts and QR codes, but the standard spot is still watched from across the room. The playbook describes the usual path as a viewer seeing an ad on TV, then searching for the app later, scanning a QR code or converting after further exposure somewhere else.

The missing click breaks the default way many teams measure. Mobile acquisition reporting was built around clicks, while CTV rarely produces one. A team that applies mobile measurement rules to CTV is judging a channel with a tool designed for different behaviour.

02. Where last-click under-credits CTV

The playbook explains one mechanism worth knowing about. Attribution systems often prioritise clicks over impressions, meaning recorded ad exposures. Because CTV attribution is impression-based and often relies on cross-device modelling, which attempts to connect television exposure with later activity on a phone or another device, CTV can sit lower in the attribution waterfall: the rules that decide which interaction receives credit for a conversion. If a viewer later taps a mobile ad, that click can take the credit for a conversion the television ad may have helped start.

Adjust can customise how CTV impressions are prioritised. That is a useful control, but it is also a choice that can move credit towards CTV. Make it before launch, and write down why. Changing attribution priority after seeing the results is how a channel ends up with the credit its owner wanted.

03. Where view-through over-credits CTV

The opposite error is just as common. View-through attribution counts conversions that occur after someone has been shown an ad, within a specified period or attribution window. It cannot show whether those conversions would have happened without the campaign.

Windows matter. The playbook recommends tighter windows for fast-converting categories such as gaming or food delivery, because a wide window can pull in conversions that would have happened anyway. It recommends longer windows for categories with more extended consideration periods, such as travel and finance. Either way, the window is a judgement to make before the data arrives.

Mike Brooks, Principal at Brooks Industries, who works with app marketers and advises Smadex, reframed the question when I asked him how often view-through attribution overstates CTV’s contribution. “To me, it’s not a question of is CTV overstated,” he told me. The more useful question is whether the window has been set appropriately and how credit is assigned between CTV impressions and later mobile clicks. He said the most advanced app marketers he works with set those windows in hours, not days, and pair attribution reporting with separate incrementality tests. The length of a window and the priority given to an impression are different decisions.

Here is how the common ways of reading CTV compare.

No single row is the answer. Read several of them against rules agreed before launch. Most importantly, incrementality means the additional customers or revenue a campaign caused that would not otherwise have occurred. An attribution report alone cannot establish that.

The playbook's five-stage testing framework puts incrementality in fourth position, ahead of scaling spend. I agree with the order. Delivery, signal and quality tell you whether the channel is working mechanically. A well-designed incrementality test asks whether it is making a causal difference.

The industry is starting to formalise these distinctions. On 1 October 2026, IAB Europe published its Connected TV Measurement Framework and Transparency Principles. It sets out the metrics on which CTV reporting depends and includes incrementality alongside conversions and return on advertising spend (ROAS) as an outcome measure. Its principles call for disclosure of measurement limitations. Where person-level reach is reported, providers should also state whether it was observed or modelled. That is the discipline I am arguing for: know what a number is built on before you trust it.

06. The strongest objection

The strongest objection is that a four-to-six-week test on a set budget is too small and too short to prove anything, and that you may be paying for a holdout when you could be paying for reach.

A short test may miss long consideration cycles, and a modest budget can produce a noisy result. If your category converts slowly, the test and measurement windows should be longer, and the conclusions more cautious.

But the alternative has its own cost. Without a credible comparison, you are choosing between attribution reports that can each mislead you. A bounded test costs money, but scaling a channel on the wrong evidence can cost much more. It also gives the finance team a stronger basis for deciding what the next tranche of spending should buy.

07. An attribution credit check you can run before launch

An attribution credit check is my name for a short list of decisions to lock down before the first impression runs.

1. Write down the credit rules. Agree the attribution window for your category and, separately, how CTV impressions will be prioritised against clicks and other impressions. Some windows may be better specified in hours than days. Record who decided, why and what would justify changing the rules in a future test.
2. Hold something out. Brooks describes three routes to an incrementality read. You can ask your demand-side platform (DSP), the system used to buy digital advertising, to run a test, while examining possible conflicts of interest. You can run one in-house using a geo holdout, which compares matched geographic markets, or ghost bidding, which withholds an ad from a control group during an ad auction. Or you can use an independent measurement partner, such as Haus or Measured, at additional cost. Whatever the method, the test and control groups must be comparable. Merely comparing people who happened to see the ad with people who did not is not enough.
3. Read at least three signals beyond the click, and compare attribution methods. Direct traffic, branded search and self-reported awareness give you different views of the potential effect. Compare the economics under last-touch and multi-touch reporting, and record how far apart they land. When the signals disagree, investigate before scaling.
4. Judge by outcomes closer to revenue. Look at cost per registration and cost per paying customer, not install volume alone.
5. Fund the creative. Budget for enough variations and sufficient testing time that a weak asset cannot be mistaken for a weak channel. Agree on the creative approach before production starts.

None of this makes CTV easy. It makes the answer harder to argue with.

08. The question to take back to your team