The Number CTV Platforms Won't Show You

$0 is the media cost a net-reporting streaming platform shows on its own income statement. That’s not the price it paid for your ad. It’s the price it never has to show. Connected TV sells two promises: precision down to the household, and accountability down to the dollar. The numbers that would let a buyer check either promise are the ones the category won’t show.

CTV pitches itself as the accountable alternative to broadcast. Self-serve, software-driven, measurable to the household. The pitch is good. The disclosure behind it is thin. A buyer often can’t verify the take rate, can’t audit the household match, and can’t trace the dollar through the auction. This study walks the four numbers a buyer keeps asking for and rarely gets.

We didn’t audit any platform’s books. We read the public filings, the independent match-accuracy testing, and the supply-chain research. Then we checked what the category discloses against what a buyer needs to verify a buy. Every figure here traces to a named primary source. Where a number is a vendor’s own claim, we say so.

The Take Rate You Can’t See

Ad-tech platforms usually book revenue net. They report gross billings minus what they pay suppliers for the media, not the full amount an advertiser spends. Under ASC 606, a platform that acts as an agent, not a principal, reports only its net cut. The media cost then never appears on the income statement. Not as revenue, not as an expense. That’s a standard, auditor-reviewed method. It’s also why a buyer can’t read a platform’s take rate off its financials.

The Trade Desk is the exception that proves the point. It discloses gross platform spend, so anyone can compute its take. For fiscal 2025 it reported about $2.9 billion in revenue on roughly $13.4 billion in gross spend. That’s a take rate near 21.6%. Most self-serve platforms report only the net figure, so the same math is impossible from the outside.

The Precision You Can’t Check

CTV’s second promise is the household. Buy the household most likely to need you, the pitch goes, and skip the waste. Independent testing says the match is looser than the pitch. In November 2025, CIMM, Go Addressable, and Truthset tested six major data providers against real ISP and MVPD household files. IP-to-postal-address links were correct 13% of the time. IP-to-email links were correct 16% of the time.

The providers don’t even agree with each other. On IP-to-postal, any two providers matched the same way just 6.4% of the time. On IP-to-email, 2.8%. Most “matches” are unique to one vendor’s model, not a fact three vendors confirm.

Not every match is a guess. “Household-level” targeting mixes two methods. Deterministic matching uses a login or a hashed email, and vendors claim 90%-plus accuracy on the slice it covers. That figure is the vendor’s own, not independently audited. Probabilistic matching infers the household from an IP address or a device graph. A buyer is rarely told, impression by impression, which kind they bought.

Even the deterministic path has limits. Matching a first-party customer list to households commonly connects 45% to 65% of the list, by vendor and practitioner accounts. And ACR data, the content-recognition signal behind many CTV segments, carries no independent accreditation of that targeting on its own. The measurement bodies grade ACR only inside a full measurement-service audit, not as a standalone targeting input.

Too Many Hands on the Dollar

A CTV dollar also loses ground on its way to the screen. Across the top 10,000 auction-traded properties, publisher supply files now list more than 450 authorized supply paths. That’s about three times the 2020 count. Roughly 27 are direct sell-side platforms. Fourteen more are authorized to resell inventory downstream. Every resale hop is a business that takes a cut before the ad plays.

How much runs through those extra hops varies. Among the top 10 demand-side platforms, resold “rebroadcast” bid requests range from 22% to 79% by platform. A buyer usually can’t see, from a media plan, which end of that range their spend sits on.

What the Whole Market Loses

Zoom out to the open web, the wider programmatic ecosystem CTV sits inside, and the leakage is measured. The ANA studied the $88 billion US open-web programmatic market in 2023. It found $22 billion, one dollar in four, is wasteful or unproductive. No more than 36% of ad dollars reach the intended audience. In a log-level cut, 15% of spend went to made-for-advertising sites built to farm ad impressions.

A separate end-to-end audit tells the same story. The ISBA, the AOP, and PwC traced advertiser spend through the UK programmatic chain. Only 51% reached publishers as working media. Another 15% landed in an “unknown delta” no one could trace. That figure barely moved between the 2020 study and its 2023 follow-up. The UK market and a premium slice aren’t a perfect stand-in for US CTV. They’re the closest end-to-end trace the industry has published.

What This Means for a Buyer

None of this proves any one platform overcharges or mismatches. It proves something a buyer can act on. The category asks for trust on two numbers, the take and the match, and hands over neither. So a buyer sets the terms. Ask for the gross spend figure before the campaign starts, the way The Trade Desk already publishes it. Ask which matches are deterministic and which are inferred. Then hold the buy to conversions you can trace yourself, not a platform’s self-reported margin or match rate.

A firm running CTV advertising is paying for household precision it usually can’t independently verify. Set the terms before the buy, not after.

A platform that won’t show its take can’t prove its price.