A word before Frank starts commenting: he's spent forty years inside measurement, at just about every company that ever counted a viewer, and now he shows up in the margins here with what he actually thinks. Get used to him.

On Monday's ADOTAT Show, Howard Shimmel described a connected TV campaign in Bill Harvey's data. It reached 3 percent of its target at an average frequency of 200. The episode's argument was that the data, the analytics and the planning software to prevent that outcome already exist in production.

The argument stopped short of one question. If the tools exist, which party in the transaction is obligated to use them before the money is spent? This issue goes looking for that party. The first place it looks is the television set itself, and it turns out the industry cannot agree on which set it is counting.

Two hundred. I've seen smaller numbers on a jury summons.

On September 9, the ARF released new findings from its DASH TV Universe Estimate study:

Roughly two-thirds of US TV households have a bedroom television.

In 63 percent of homes with both a living room set and a bedroom set, the two are different brands. That figure is 56 percent even for the largest manufacturer.

In 15 percent of TV households, about one in seven, the bedroom set rather than the living room set is the most-watched television in the home.

DASH is not a side study. Nielsen began using DASH as its source of universe estimates in February 2026, and DASH received its own MRC accreditation earlier this year.

The finding matters for any household-level frequency cap built on one manufacturer's sets or one connection path. If the most-watched set in the home is a different brand, that cap may not see it at all.

It also matters for Nielsen's own inputs. Nielsen says its US national TV big data set includes roughly 45 million households and 75 million devices from Comcast, Dish, DIRECTV, Roku and Vizio. That comes alongside a panel of about 101,000 people in about 42,000 households. Divide the devices by the households and the big data set sees about 1.67 devices per home.

Nielsen is direct about what those devices cannot tell it. On its own site, the company says big data captures viewing at the device level, not the individual level, which makes it "impossible to determine whether anyone is actually watching TV when the device is on." Nielsen's answer is the panel. That is why the product is called Big Data + Panel.

Unmeasured sent Nielsen four questions on September 9:

How Big Data + Panel detects uncaptured sets inside an instrumented household.

Whether it holds an estimate of viewing that happens outside the primary set.

Whether any resulting weighting error is random or directional.

The scope and date of its DASH adoption.

Nielsen confirmed receipt the same morning. It had not answered by the noon ET Friday deadline.

The bedroom TV. The one you fall asleep to. Fifteen percent of homes, and it's the set that counts. I spent six years on the TV side at Nielsen Media Research and nobody ever once asked me what was upstairs. Now their own website says the boxes can't tell if anybody's watching. I'd have been walked out for typing that sentence in 2003. Today it's the brochure.

Ad-supported streaming services increased their commercial load substantially in 2026. According to Ampere Analysis data cited by Business Insider, the average number of ad minutes per hour rose 18 percent between January and August:

Paramount+ carries the heaviest load of the services tracked, at around nine minutes per hour.

Disney+ and Hulu run roughly 7.5 to 8.5 minutes.

Netflix remains under 2.5 minutes, despite posting the largest increase of the group.

The relevance to frequency is direct. More ad minutes means more inventory to fill. Without a coordinated cap, more inventory raises the odds that the same household sees the same campaign again. Earlier research cited by eMarketer found that 62 percent of US streaming subscribers had seen the same ad multiple times in a single session.

Eighteen percent more commercial time and not one press release about 18 percent more viewers. Funny how that works.

When Shimmel recorded with ADOTAT in May, he described high single digit declines in linear advertising at every major media company except Fox. The second quarter did not improve on that.

Paramount: linear TV ad revenue fell 14 percent year over year, while streaming ad revenue grew 8 percent.

Warner Bros. Discovery: linear TV ad revenue dropped 27 percent against 9 percent streaming ad growth, with the loss of NBA playoff rights driving much of the decline.

Fox: advertising revenue in its Television segment rose 108 percent in the quarter ended June 30, driven by the World Cup, Tubi and political advertising at its owned stations.

For measurement, the pattern matters because measurement contracts are not written as a share of a declining quarter. A company losing a quarter of its linear ad revenue still pays for measurement, now out of a smaller pool. Shimmel made that point Monday about contracts priced in growth years.

Frank Kowalczyk
Fox is up 108 percent because of soccer. Everybody else is down because of everybody else. Write that on the Nielsen invoice and see if they take it.

Samsung Ads sells a product called Optimal Reach. The pitch rests on a duplication figure of 41 percent. That figure has appeared in public in at least three formulations, and they describe three different quantities.

First, on Samsung's own blog, the number is an audience overlap. In a Samsung Ads analysis of 18 campaigns, 41 percent of audience reach overlapped between CTV and linear. That is a share of households reached.

Second, in sponsored content Samsung placed through AdExchanger's Content Studio, the number becomes a claim about impressions. That piece says that in the US, 41 percent of streaming impressions reach viewers who have already seen the same ad. That is a share of impressions that are repeats. It is a different denominator and a different numerator.

Third, in trade coverage of Samsung's "Missed reach, found revenue" report, the number becomes duplicated impressions across two platforms, converted into dollars. Samsung Ads said 41 percent of ad impressions were duplicated across CTV and linear. It valued that duplicated time at about $9.3 billion. It then projected up to $141 billion in additional sales across six sectors, assuming the money was reinvested at an average return of $15.16 per ad dollar.

The arithmetic of the headline number is worth stating plainly. $9.3 billion multiplied by $15.16 is $141 billion. The larger figure is not a measured loss. It is a duplication estimate multiplied by an assumed return on ad spend. The same coverage reports Samsung's claim that Optimal Reach can lift incremental cross-platform reach by as much as 47 percent.

Samsung's framing of duplication may well be directionally correct. Shimmel said much the same on Monday, without selling anything. The point here is narrower. A market cannot write a frequency rule into a contract when the leading public statistic on duplication changes meaning between documents. A buyer who hears "41 percent" cannot tell whether it refers to households, impressions or dollars. The seller controls which one the buyer is thinking about.

Kowalczyk Ratio for the week: number of ways a statistic gets described, divided by number of times it got measured. Samsung's 41 percent comes in at three to one. At Arbitron a three to one got you a meeting with legal and a box for your desk plant. I sat in that meeting once in 2008, and it wasn't even my number. Sanded the transom on the boat Saturday. Unrelated. Also, a TV maker measuring duplication across other people's TVs. In more than half the two-set houses, the bedroom set is somebody else's brand, pal.

The ANA's Aquila project exists because large advertisers concluded the existing currencies were not giving them deduplicated reach and frequency. Its own estimates of what that is worth have varied widely.

March 2025: an Aquila release said the initiative was estimated to eliminate $50 billion in waste due to excess frequency over three years.

October 2025: trade coverage reported Aquila's estimate that the reach deduplication component would improve big advertisers' buying efficiency by 10 percent and yield $50 million in improved productivity over its first three years.

As worded, the two figures describe different scopes: one covers the whole endeavor, the other covers one component. Neither published statement explains how the two relate. The gap between them is a factor of one thousand. Aquila is an advertiser-funded project designed to replace estimates with measurement, and both of these numbers are estimates.

Aquila is live. In March 2026 it announced a first rollout phase within 30 days, using impression-level data from Meta, YouTube and linear TV. A second phase adding TikTok, Amazon and iHeartRadio was slated for the second half of the year, and 27 advertisers had signed on.

The most instructive data point from that rollout cuts against the premise of this issue. Aquila presented a case study from beta tester Uber, which found its ads were reaching a wider audience than planned, at less than half the planned frequency. Uber went back to review its frequency strategy. Excess frequency is not the only failure a better count reveals. Sometimes the count shows the plan never delivered the frequency the buyer paid for.

Aquila also describes its own role carefully. It was built so advertisers could measure the actual reach and frequency of their campaigns across TV and streaming. It was not built to serve as a currency that buyers and sellers transact on. A measurement service that is not a currency reports what happened. By design, it is not the mechanism that stops delivery.

Shimmel has been skeptical of Aquila's prospects. In an on-record call with ADOTAT in August, he said he expects it to "follow the same path" as Project Origin in the UK, where advertisers reportedly did not see enough value to fund it.

I'm typing this from a client's offsite. They think I'm in the fraud-prevention breakout. Fifty billion, then fifty million. When I was at Rentrak, a regional sales VP dropped three zeros off a forecast once. He got to keep the car but not the parking space. And Uber paid for frequency it didn't get and found out from the ANA. The ANA. I'll let that sit.

How we reported this: This issue draws on company releases, Fox's SEC filing, Nielsen's own published materials, the ARF's September 9 DASH release, and trade coverage in MediaPost, Digiday, AdExchanger, StreamTV Insider, TV Tech and Business Insider. Shimmel's quotes come from Monday's episode and an on-record August call. The 3 reach, 200 frequency campaign is Shimmel's account of Bill Harvey's data; ADOTAT has not seen that data. Samsung Ads and Aquila figures are self-published and not independently audited. The multiplication of Samsung's $9.3 billion by its $15.16 return assumption is ADOTAT's arithmetic, as is the devices-per-household figure. ADOTAT has asked the ARF for the exact question wording behind "most-watched." Nielsen received questions on September 9, confirmed receipt, and had not answered by the September 11 deadline. Howard Shimmel is a recurring ADOTAT source, was Monday's guest on The ADOTAT Show, and is head of strategy and a board member at DataFuelX. Pesach Lattin consults for clients on ad tech analysis as a standing engagement and holds no individual equities.

The five links between a rendered impression and a human being, and why a perfectly enforced household cap can still be capping furniture.

What the MRC flagged about Nielsen's panel in March, and why it lands on the exact variable DASH measured last week.

The billable event map: every party in the chain that gets paid for the two hundredth impression, and the two that don't.

The five contract terms that would make a stop button exist, and who in the chain loses money on each one.

Above the wall, the industry cannot agree what it is counting. Below it, nobody is obligated to act on the count, and almost everybody is paid not to.