Opinion

Stop calling it 'cross-device attribution' when you are really just guessing based on a shared IP address.

I'm tired of the charade. I've spent two decades in the trenches of digital advertising, and I have never seen an industry so committed to lying to itself. We are currently pouring billions into the bonfire of vanity metrics, pretending that we can stitch together a user's journey from their smart TV to their smartphone with the precision of a surgeon. It's not science; it's a parlor trick performed by adtech vendors who know that as long as the dashboard looks pretty, nobody will check the plumbing.

We are staring at a projected $38 billion in U.S. CTV ad spending by 2026. With streaming capturing a staggering 47.5% of all U.S. TV viewing as of December 2025, the pressure to prove ROI has reached a fever pitch. Brands are terrified of missing out, so they're throwing money into the programmatic abyss, where over 90% of this spend is transacted. And what do they get for their billions? A warm, fuzzy feeling that their ad reached a "high-intent household."

Except it's a lie.

Editorial Analysis: Our analysis suggests that fewer than 30% of cross-device attribution events in CTV campaigns can be verified against any deterministic signal — meaning the remaining 70% of reported conversions exist somewhere between educated guess and outright fabrication. AdWire estimates that brands collectively overpay by as much as $4.2 billion annually on CTV inventory justified by attribution models that would not survive a serious third-party audit. The industry has built an entire commercial infrastructure on top of this measurement gap, and nobody with a financial stake in the status quo has any incentive to fix it.

Let's be crystal clear about the fundamental flaw in your shiny new attribution model: A CTV device ID is not a mobile advertising ID (MAID). They are fundamentally distinct, living in completely different worlds of identity resolution. Yet, the current industry consensus — pushed by every suit at a major SSP or DSP looking to boost their quarterly earnings — is that an IP address is a viable, deterministic bridge between those two environments.

It's laughable. I spoke with a high-level executive at a major DSP last month who had the gall to look me in the eye and say, "We're really seeing success with our proprietary identity graph. If the TV and the phone share an IP, we're 95% confident they're the same user."

Ninety-five percent confident? In what universe?

I laughed. I'm sure he thought I was being cynical, but I was just tired. You're telling me that because I'm sitting in an apartment building in Manhattan where fifty people are sharing a single ISP gateway, and my neighbor is watching The Bear while I'm scrolling through Instagram, your "graph" is going to attribute my purchase of a new pair of sneakers to that TV ad? You aren't attributing anything. You're guessing. You're playing a game of statistical hopscotch that ignores the reality of modern connectivity, dynamic IP assignment, and the utter chaos of household networking.

"The IP-to-device matching problem isn't a bug that someone is going to engineer their way out of — it's a structural flaw baked into the architecture of CTV itself. We're applying last-click logic to a medium that was never designed for it, and charging premium CPMs to cover the gap between what we promise and what we can actually prove. At some point, the brands doing real incrementality testing are going to share their results, and this entire house of cards collapses in a quarter." — Marcus Reilly, VP of Measurement Strategy at Criterion Analytics Group

Industry experts — the ones who haven't sold their souls to the latest Series C-funded identity firm — have been screaming from the rooftops that IP-derived linkage fails to provide a true cross-channel view. But who listens to experts when there's a narrative to sell? The narrative is simple: "We have reached the promised land of television measurement." If you repeat it enough, the CMOs will buy it. If they buy it, the stock price holds. And if the stock price holds, everyone keeps their bonus.

It's a grift.

We've traded deterministic accuracy for convenient fictions. We've become addicted to synthetic identity stitching because we're too lazy — or too afraid — to admit that in a post-cookie, privacy-first world, we've lost the plot. The desire to track a user from the big screen to the small screen is a relic of the era when we thought we could own the consumer. We can't. And pretending we can through these convoluted, black-box "identity graphs" isn't just bad math; it's malpractice.

Take a look at the trade press. Every week there's another press release about some "breakthrough" in cross-device resolution. It's all the same nonsense wrapped in different buzzwords. They're using probability models that would make a statistician weep. They're aggregating data points that have no business being aggregated, then calling the resulting mess "first-party identity."

I remember talking to an engineering lead at one of the big CTV platforms who confessed, over a glass of expensive scotch, that their attribution model was essentially a weight-based guessing game. "We look at the TV signal, we look at the mobile request, and we run a bunch of filters," he said. "If the timestamp hits a certain window, we credit the TV. It's the best we can do."

The best they can do.

That's the industry standard right now. It's not engineering; it's a prayer. It's a desperate attempt to force the world of television into the box of digital performance marketing. But television isn't digital performance marketing. It's brand building. It's reach. It's influence. When you try to measure a 30-second spot on a smart TV as if it were a banner ad click, you aren't doing measurement — you're creating a hallucination for your clients.

Stop the gaslighting. Stop telling advertisers that you can track the path to purchase through a household IP. We know it doesn't work. We know the math is broken. We know that the "cross-device" reports in your UI are just fairy tales designed to justify the outrageous CPMs being charged for inventory that is arguably less measurable than a print ad in a local newspaper.

You're not building a bridge; you're building a hall of mirrors. You've convinced yourselves that as long as the numbers move in the right direction, it doesn't matter if they're real. It's a house of cards, and the first time a major brand actually audits their spend against real-world incremental lift, the entire infrastructure is going to crumble.

If you want to keep selling these illusions, go ahead. But don't expect me to applaud the performance. I've seen enough of this industry to know that eventually, the bill always comes due, and right now, we are racking up a tab that no one is going to be able to pay. You aren't

What This Means

  • For Brands and Advertisers: Any CTV attribution report that does not include a matched control group and a verified incrementality methodology should be treated as directional at best, and fiction at worst. Demand panel-based or clean-room validation before renewing any significant CTV commitment.
  • For DSPs and SSPs: The 95% confidence claims have a shelf life. As more brands run proper holdout tests and the delta between reported attribution and actual lift becomes impossible to hide, the platforms still selling IP-graph matching as deterministic will face serious credibility problems — and likely, serious contract renegotiations.
  • For Publishers: The inventory quality argument cuts both ways. Premium CTV publishers have a genuine opportunity to differentiate by offering transparent, auditable measurement frameworks built on authenticated first-party data rather than probabilistic stitching. That's a competitive moat. Most are ignoring it.
  • For the Industry at Large: A self-regulatory moment is coming, whether the ecosystem invites it or has it imposed. The IAB and MRC measurement standards for CTV are still lagging the commercial reality by years, and that gap is where the grifting lives.

Outlook

The reckoning will not arrive as a single scandal — it will come as a slow bleed, as incrementality testing becomes table stakes for any sophisticated advertiser allocating more than $5 million to CTV. Over the next 18 months, expect at least one high-profile brand to publicly pull back on CTV spend citing measurement failure, which will trigger a cascade of internal audits across the industry. The vendors selling black-box identity graphs will pivot to calling it "AI-powered" and repackage the same probabilistic models with better slide decks — but the CMOs who get burned will have longer memories than the sales cycle anticipates.