A Cynopsis industry brief published September 29, 2026 put a name to a problem that's been building for a couple of years: marketing mix models can tell you whether your CTV spend worked, but they routinely can't tell you why - which creative, which platform, which daypart actually drove it (Cynopsis). That's not a flaw unique to any one vendor's model. It's a symptom of a deeper measurement problem: connected TV doesn't fit cleanly into either half of how most companies still organize their measurement stack - "TV," handled by offline/linear tools, and "digital," handled by pixel- and platform-based attribution.

CTV is the fastest-growing part of many media budgets, and it's also, by a wide margin, the hardest channel to measure honestly. That combination - rising spend, weak measurement - makes it the clearest current example of why offline and online media need to be modeled together instead of handed to two different tools.

The CTV paradox

CTV behaves like television: it's a lean-back, sound-on, non-clickable viewing experience, often on a shared household screen. But it's delivered like digital: programmatically bought, impression-logged, theoretically "addressable." Over 90% of CTV ad spend is now transacted programmatically, which is exactly the kind of fact that tempts organizations into measuring it with digital attribution tools built for click-based channels.

The problem is that CTV essentially never produces a click. A viewer sees a spot on the living-room screen, and if they convert, they do it minutes or days later, on a different device, with no identifier connecting the two moments. Streaming platforms also control their own authentication data inside a walled garden, so even probabilistic device-matching only gets you partial, platform-reported visibility - and that platform has every incentive to report the exposure favorably, not accurately.

Why last-click and platform reporting both fail CTV

Two separate measurement failures stack on top of each other here, and it's worth naming them separately because they push in the same direction but for different reasons.

Last-click attribution undercounts CTV by an estimated 40 to 60 percent in most accounts, precisely because the conversion event it's built to catch - a click followed shortly by a purchase - essentially doesn't exist for a TV-screen impression. The model isn't wrong about what it measures; it's measuring the wrong thing for this channel entirely.

At the same time, platform-reported CTV performance tends to run in the opposite direction for a different reason: walled-garden reporting has no visibility into, and no incentive to account for, what happened outside its own environment. The result is a channel that gets simultaneously undercounted by the attribution methods built for digital and over-credited by the platforms selling the inventory - two errors that don't cancel out, they compound, and they point budget in opposite directions depending on which number a planner happens to be looking at that week.

Why MMM alone doesn't fully close the gap either

This is the part the Cynopsis piece gets right, and it's worth taking seriously rather than treating MMM as an automatic fix. A conventional marketing mix model, run at the aggregate spend level, can confirm that a CTV flight correlated with incremental lift. What it often can't do is decompose that lift by creative, platform, or daypart, because that granularity was never in the model's channel structure - CTV went in as one line item, same as it would in a spreadsheet.

That's a resolution problem, not a methodology problem, and it's fixable - but only if the model is built to carry that granularity in the first place, and only if it's anchored to real experiments rather than historical spend-and-outcome correlation alone. A model that's calibrated against actual incrementality tests - geo holdouts, lift studies - on a meaningful share of its channels has a ground truth to check its CTV coefficient against. A model trained purely on historical co-movement of spend and revenue has no such anchor, and will happily produce a confident-looking number that's simply wrong, because nothing in its training data ever isolated CTV's effect from everything else moving at the same time.

CTV as a preview, not an edge case

It's tempting to treat CTV's measurement problem as a special case - a weird channel that doesn't fit the usual boxes. The more useful way to see it: CTV is what every channel's measurement problem looks like once you remove the assumption that "online" and "offline" are naturally separate categories with naturally separate tools.

Linear TV has long been measured with offline-style tools: panel data, reach-and-frequency models, MMM at the channel level. Paid social and search have been measured with online-style tools: pixels, click-through windows, platform dashboards. CTV sits exactly on the seam between those two traditions, and it breaks both of them, because it was never going to respect a boundary that was drawn for organizational convenience rather than because the underlying media behaves differently.

That's precisely why it's the channel most exposed by the industry's current wave of MMM adoption - and the best evidence that the fix has to be structural, not a better version of either the "TV tool" or the "digital tool," but one model that was never asked to draw that line in the first place.

What this means operationally

For a marketer deciding where to put the next incremental dollar, the practical implications are fairly direct:

  • Don't trust either number on its own. If your last-click attribution and your platform-reported dashboard disagree on CTV's performance by a wide margin, that's expected, not a sign one tool needs fixing - it's a sign neither tool was built to answer the question.
  • Push for channel-level granularity inside your MMM, not just a CTV line item. If your model can't distinguish creative or platform within CTV, you'll know if it worked but not what to do next, which is the exact gap the Cynopsis piece flagged.
  • Calibrate with real experiments wherever you can. Geo holdouts and lift tests on even a subset of CTV spend give your model a ground truth it otherwise has no way to get, and that discipline matters more for CTV than almost any other channel precisely because its platform-reported numbers are the least trustworthy starting point.
  • Expect this to be where your mixed-channel campaigns live. A CTV flight that's really working is likely also moving branded search, direct traffic, and in some categories offline response. A model that doesn't see TV, OOH, radio, print, and digital side by side has no way to attribute that movement back to the CTV flight that caused it.

The takeaway

CTV's measurement problem isn't a one-off quirk to patch around - it's the clearest current proof that treating "TV" and "digital" as separate measurement universes doesn't survive contact with how media actually works anymore. The channels that used to sit neatly on one side of that line are disappearing, and the ones that replace them, CTV first among them, sit directly on the seam.

This is the exact case Animo is built around: one always-on, causal MMM that models TV, OOH, radio, print, and digital - CTV included - as a single system, calibrated against real incrementality experiments across the mix, rather than a digital model and a TV model stitched together after the fact. If you're currently getting two different answers for what your CTV spend is doing depending on which dashboard you open, that's the problem worth fixing before the next flight goes out.

Sources

  • Streamlined 9/29/2026: MMMs Can Tell You If CTV Worked. They Can't Always Tell You Why (Cynopsis): cynopsis.com
  • Connected TV Attribution & Incrementality: Complete Measurement Guide for DTC Brands (ATTN): attnagency.com
  • CTV Measurement Guide 2026: Track & Optimize CTV Ads (Improvado): improvado.io
  • Over 90% of CTV Ad Spend Is Now Transacted Programmatically (MNTN Research): research.mountain.com
  • Nearly half of US marketers plan to invest in MMM over the next year (EMARKETER): emarketer.com
  • Incrementality Trends & Statistics (EMARKETER): emarketer.com