Dear performance CTV companies,
Please stop calling everything performance CTV.
By all means, measure outcomes. Optimize toward sales. Make television more accountable. I’m all for it.
But if the product is basically a pixel, an attribution window and a CPA column, that doesn’t mean we’ve transformed television into Google.
TV is still TV, even when it’s online
Somewhere between the satellite dish and the Ethernet cable, we decided we had also reengineered human psychology. But we know how people actually watch television because we are people who watch television. We sit on a couch with a phone in our hand. We answer a text. We scroll. And maybe we look up when something on the television catches our attention.
How often does somebody see a TV commercial and immediately buy the product? Not just Google it later. Not discuss it with their spouse and buy it three days from now, but see an ad, buy a thing.
Of course, it happens. But building an entire definition of television performance around that niche behavior is weird and illogical.
Search catches people while they’re declaring what they want. Social is sitting in their hand, with the click, swipe or checkout button right there. Television is usually 10 feet away while somebody is halfway through an episode and halfway paying attention.
Streaming changed how the television signal gets there. It didn’t move the screen six inches from our face or put a keyboard on the remote.
We understood the TV opportunity pretty well before CTV came along. Linear television was an awareness and demand-generation medium that could absolutely drive business results. But those results didn’t need to happen 30 seconds after the spot aired for the media to have worked. Marketers looked at response curves, sales data, incrementality and marketing mix models because everyone understood there was lag and a halo effect.
When television started coming through the internet, the industry got very excited, for good reason. We suddenly had richer logs, more addressability, more exposure data and more ways to connect an impression to an outcome.
And then we got carried away.
CTV isn’t magic
Better observability somehow turned into a claim that television now behaved like search and social. The advertising dashboard changed. The person sitting on the couch did not.
How the ad tech plumbing works should make us a little more humble about all of this, too. The ad is usually shown on one device while the conversion happens on another. Some identity layer has to connect those two devices. Publishers pass different signals. Platforms use different identifiers. Conversion data comes back with different levels of consistency and different levels of delay.
The IAB has written extensively about fragmentation and inconsistent signal quality in CTV measurement. The org promotes conversion APIs in part because CTV still has technical barriers that don’t exist in the same way on search and social.
Yet the sales pitch somehow became: basically Meta, except it’s on your television.
We’ve seen how messy one piece of that infrastructure can get. Only 23% of the residential IP addresses InterMedia and Adstra recently evaluated for ZIP-code targeting reached their intended geographic target using the IP-mapping approaches we tested. Device-level CTV identifiers were considerably more stable over time.
That finding does not mean CTV works 23% of the time. It means the infrastructure underneath some very confident-looking attribution numbers is not always as confident as the dashboard makes it appear.
Missing mobile connections
And Apple isn’t making it any easier to connect what happens on CTV to what happens on mobile.
iCloud Private Relay was already complicating CTV attribution by obscuring the IP addresses used to connect television exposures to conversions on iPhones. Now Apple’s latest WebKit changes are putting additional restrictions on ad tech and identity infrastructure, including LiveRamp, UID2 and ID5. And because all browsers on iPhone use WebKit, this reaches beyond Safari.
Here’s an example of the challenge CTV faces in tracking conversions in Apple’s ecosystem: For one major DTC advertiser we work with, 63% of purchases happen on an iPhone. When we tracked conversions for a recent campaign that were captured by a DSP pixel, the pixel recorded 109 completed checkouts. But when we tracked conversions by independently matching impression IPs to the advertiser’s CRM logs, the CRM match found 1,250. Different methodologies, sure, but that’s an awfully large discrepancy to ignore.
And this is the environment in which we’re telling advertisers to judge television by the CPA in a dashboard. Apple will keep making privacy decisions based on its own priorities, as it should. Our industry’s job is to build measurement that can withstand those decisions and stop pretending that every conversion we can’t directly observe is a conversion television didn’t influence.
Keep building performance CTV
But problems arise when a CTV platform’s performance sales pitch turns into the advertiser’s expectation.
Someone sees a CTV ad. Two days later, they search the brand. Paid search gets the click. The brand site records the conversion. CTV gets whatever credit the identity and attribution stack manages to reconstruct. Then everyone gets together for a performance meeting and discovers, through the miracle of modern analytics, that search had a lower CPA.
The instinct, then, is to treat search as the better-performing channel over CTV. And less credit is given to CTV for causing that search to happen.
Television has always been unusually good at creating demand before somebody starts actively shopping. Instead of treating that as a strength, we’ve built measurement systems that can punish television for being farther away from the receipt. The channel that happens to be closer to the cash register gets congratulated for making the customer walk into the store.
We have much better tools available to us now than “Trust me, TV works.” We can look at exposure quality, reach, frequency, content, attention, branded search, site visitation, conversion, geography, incrementality and longer-term sales contribution. Server-to-server conversion feeds can close some gaps. Holdouts can tell us whether the media caused something to happen. MMM can capture effects that user-level attribution misses. AI can help us make sense of all of those signals together.
If the industry’s big AI breakthrough is using all of that information to manufacture a nicer-looking last-touch answer, we’ve wasted an extraordinary amount of energy.
So please keep building performance CTV.
Build better identity systems. Build better server-to-server conversion infrastructure. Fix the data-sharing gaps. Tell advertisers where deterministic measurement stops instead of quietly smoothing over the edges. Run incrementality tests. Build feedback loops that actually make buying smarter. Solve enough of those problems and performance CTV becomes a very compelling thing.
Just stop promising advertisers that television already behaves like Meta.
Because when that promise eventually collides with reality, the advertiser does not blame the positioning deck. They decide CTV doesn’t work as well as other channels. And that perception screws over publishers, agencies, serious technology companies and everyone else who is actually trying to make television more measurable.
CTV has a great future as a performance medium. It doesn’t need a fake self sabotaging narrative around it.
Sincerely,
Someone who cares
“On TV & Video” is a column exploring opportunities and challenges in advanced TV and video.
Follow InterMedia Advertising and AdExchanger on LinkedIn.
For more articles featuring David Nyurenberg, click here.
