You measure connected TV sales by combining signals, because no single one is reliable on its own. CTV ads almost never get clicked, so the pixel-based attribution that works for Meta and Google will barely see them. The practical stack is four layers: the CTV platform’s own household matching, a post-purchase survey that asks where customers heard about you, lift in branded search and direct traffic in the hours and days after ads run, and a geographic holdout when the spend is big enough to justify one. Each layer has blind spots, but they fail in different directions, which is what makes the combination useful.
Why doesn’t normal attribution work for CTV?
Because the ad and the purchase happen on different screens. Someone sees your ad on a living room TV, then later picks up a phone and searches your brand name, or types your URL. Your attribution tool records that order as branded search or direct. Nothing in the click path connects it to the TV. That is why brands that add CTV often see branded search look unusually strong for a few weeks while CTV looks like it did nothing.
What does the CTV platform’s own reporting tell you?
Most CTV platforms match ad exposures to site visits and orders at the household level, usually through IP address and device graphs, then report the conversions they believe they influenced. It is the most direct signal you will get, and it is also the platform grading its own work. Household matching tends to over-credit, because a household that saw the ad and also saw your Meta ads and got your emails will show up as a CTV conversion. Use it for comparing creative and inventory within CTV, and be careful about using it to justify the channel as a whole.
Do post-purchase surveys work for CTV?
Better than for almost any other channel. Customers remember seeing a TV ad, and they will say so when asked. Add a TV or streaming option to your how-did-you-hear-about-us question before the campaign launches, so you have a baseline of how many people already pick it. The share of new customers choosing TV is a cleaner directional read than any pixel metric. Surveys under-count people who saw the ad and do not remember it, so treat the result as a floor. I wrote more about using them in using post-purchase surveys as a real attribution signal.
How do you read lift in branded search and direct traffic?
Compare branded search impressions, direct sessions, and new customer orders during the flight against the same weeks before it, adjusted for anything else you changed. If CTV is working, you should see branded search and direct climb when ads are live and settle back when they stop. This is crude, and seasonality or a big Meta push can muddy it. But if CTV is running at real spend and none of these move at all, that is information too.
When is a geo holdout worth running?
When CTV becomes a meaningful line in the budget and you need a number you can defend. You run ads in some regions and hold them back in comparable ones, then compare new customer revenue between them. It is the closest thing to a controlled experiment you can run without a measurement vendor. It costs you the sales you would have made in the holdout regions, and it needs enough volume per region to see a difference. I walked through a simple version in running a cheap geo holdout, and the bigger picture is in what is incrementality testing.
Where does an attribution tool fit?
It is where you bring the pieces together so CTV is not living in a separate spreadsheet. ThoughtMetric, for example, lets you add CTV as a custom channel with its spend and pair that with post-purchase survey responses, so TV shows up in the same report as your other channels. (Disclosure: ThoughtMetric sponsors this site.) It will not see TV impressions any more than other pixel-based tools do, so the survey layer is doing the work there. For heavier TV programs, dedicated vendors exist. I covered performance TV options in Tatari alternatives, and Rockerbox measures offline media alongside digital, as I noted in Rockerbox alternatives after the DoubleVerify acquisition.
What mistakes should you avoid?
Launching without a baseline is the big one. If you did not record survey answers and branded search before the flight, you have nothing to compare against. The second is running CTV at a budget too small to move any of these signals, then concluding it does not work. The third is judging it on two weeks. TV effects build, and purchases from a TV ad can land well after the exposure.
Quick answers
- Testing CTV for the first time? Add a TV option to your post-purchase survey two weeks before launch.
- Comparing creative or inventory inside CTV? The platform’s household matching is fine for that.
- Justifying the channel to a founder or CFO? Pair survey share with branded search and direct lift.
- CTV is now a large budget line? Run a geo holdout.
- Want TV in the same report as everything else? Track it as a custom channel in your attribution tool, such as ThoughtMetric (our sponsor).
Leave a comment