Full analysis

Decision Council — Briefing Mode: "Modern MMM" (Aperiam / Prescient AI)

Step 1 — Frame

The episode is a vendor-led argument that media mix modeling (MMM) — a statistical method that estimates each channel's sales contribution without tracking individual users — is back, and that when you measure correctly, search is over-credited and upper-funnel video (especially YouTube) is under-credited. The implied decision for operators: do I rethink how I measure and allocate upper-funnel budget, and what does the shift toward modeled, modeled-daily measurement mean for my business?

• Reversibility: Mostly Type 2 (easy to reverse). Shifting 5–15% of search budget into a test, or trialing an MMM vendor, is cheap to undo. The strategic positioning questions (is your value prop threatened by modeled measurement?) are more Type 1.
• What's actually being decided: Two different things get conflated. (1) A budget-allocation question buyers can test next quarter. (2) A structural question — measurement is moving from deterministic tracking to modeled inference, and from quarterly to near-real-time. That second one reshapes whose numbers get trusted.
• Forcing function: None acute. Cookie deprecation drift, channel fragmentation, and a fresh measurement gap in AI/LLM ad surfaces create slow pressure, not a deadline.

Net impact read: medium for measurement and buy-side operators; low as hard news. This is a well-argued vendor thesis, not an industry event. The signal worth your attention is directional, not the specific claimed numbers.

Step 2 — The Council

The Market Analyst Strip out the vendor spin and there's a real macro signal: budget authority is migrating toward whoever controls the modeled truth. If MMM-style measurement keeps gaining, the deterministic measurement crowd (last-click attribution, platform-reported ROAS) loses pricing power, and incrementality/MMM vendors gain it. In plain terms: the scoreboard is being rewritten, and whoever owns the new scoreboard gets paid. YouTube being "under-credited" is convenient for Google but also plausible — Google itself published a June 2025 paper pushing neural-net measurement. Watch for the big platforms to co-opt this: own the methodology, own the credit. Public measurement names (DoubleVerify, IAS, Comscore, Nielsen, VideoAmp, iSpot) should care — the center of gravity is shifting from verification toward causal allocation.

The Skeptic The load-bearing assumption is that Prescient's model is right about the counterfactual — that those YouTube impressions actually caused the later searches. Every MMM claims this; the honest ones admit you can't fully prove it without holdouts. "70% lower CAC and 2.2–2.5 points higher ROAS than YouTube reports" is a marketing line, not a validated benchmark — and notice the recommendation conveniently favors the cheapest-to-defend, hardest-to-disprove channel. Plainly: a vendor whose model says "spend on the channel platforms under-report" is selling exactly the story buyers want to hear. Also note the self-serving symmetry: search over-credits, so trust our model instead. The 96.3%-accurate Cardi B anecdote is unfalsifiable. Healthy thesis, motivated narrator.

The Operator Tuesday morning this is genuinely usable — and that's the real story. "15 minutes to connect data, live in 4–8 hours" plus weekly (not quarterly) reoptimization is a workflow change buyers will feel. The land-and-expand play — make one channel dynamic, then spread — is how this actually enters an org without a measurement-team turf war. Plainly: it gets in the door by being useful on one channel before anyone has to bet the whole budget on it. What breaks at 90 days: weekly budget swings create attribution whiplash, finance asks why ROAS numbers move every Monday, and the platform reps whose channels lose share start fighting back with their numbers. Whoever owns measurement now owns a political fight.

The Customer / End User (the brand / advertiser) From the brand's seat, the appeal is real: fragmentation across CTV, TikTok Shop, 7–8 retailers, and now LLM surfaces has made the old measurement stack useless for upper-funnel. A modeled, single-pane view is what they're actually begging for. Plainly: brands can't see across their channels anymore, and someone promising one honest scoreboard is very attractive. But the buyer should be careful what they wish for — they're trading platform-reported numbers (biased toward the platform) for vendor-modeled numbers (biased toward the vendor's recommendations). The right posture is "trust but verify with holdouts," which True himself endorses. The AEO/GEO measurement vacuum is the genuinely unaddressed pain.

Step 3 — The Tensions

1. Is YouTube actually under-credited, or is that the most sellable conclusion? The Market Analyst sees a plausible structural truth (impressions seed searches; search hoards the credit). The Skeptic sees a vendor optimizing for the recommendation that's cheapest to defend and impossible to disprove. Both can be partly right — the direction (search over-credited) is well-supported by years of incrementality work; the magnitude is marketing.

2. Does modeled measurement empower buyers or just relocate the bias? The Customer wants one honest scoreboard. The Skeptic notes you've swapped platform bias for model bias. The resolution is governance: holdout tests are the only referee neither the platform nor the vendor controls.

3. Who captures the value of this shift? The Operator sees a nimble vendor winning on workflow. The Market Analyst suspects the walled gardens (Google publishing neural-net measurement papers) will absorb the methodology and keep the credit inside their own gardens — leaving independent measurement firms squeezed.

Step 4 — Synthesize

What it hinges on: Whether you believe (a) upper-funnel video is systematically under-credited — well-supported in direction, oversold in magnitude; and (b) the industry's measurement center of gravity is shifting from deterministic tracking toward modeled, near-real-time allocation — yes, and this is the durable signal.

Which way the council leans: The direction is real and worth acting on; the specific numbers are a sales deck. Treat the episode as confirmation of a trend, not as a benchmark.

What this means by stakeholder:

• Media buyers / agencies: Run the test. Pull 5–15% of search into an upper-funnel channel with a proper holdout. It's a cheap Type 2 bet with real upside. Don't adopt any vendor's ROAS uplift number as fact — make holdouts your referee.
• Independent measurement firms (DV, IAS, Comscore, Nielsen, VideoAmp, iSpot): This is the strategic watch item. Causal/MMM allocation is encroaching on verification's turf, and Google is publishing in the space. If your roadmap is still mostly deterministic verification, the modeled-allocation layer is where budget authority is migrating.
• Publishers / sellers, especially CTV and video: A measurement narrative that credits upper-funnel halo is tailwind for your inventory's story — but only if you can plug into the modeled frameworks buyers start trusting. Get your inventory legible to MMM vendors.
• Walled gardens: Google looks positioned to own both the methodology and the credit. Watch whether they keep the modeled truth inside their garden.
• Everyone: The AEO/GEO / LLM-ad measurement gap is the genuinely open frontier — OpenAI selling ChatGPT ads with no measurement is a vacuum someone will fill, and that's a bigger eventual story than YouTube allocation.

To de-risk before acting: Demand holdout-validated results, not modeled-vs-self-reported deltas. Test in one channel before reallocating at scale. And separate the trend (modeled, real-time measurement is rising) from the pitch (this specific vendor's specific numbers).

What did we miss? Is there a persona we should add for this specific decision? A General Counsel / Privacy lens could be worth adding — MMM's selling point is that it avoids user-level tracking, which in a post-cookie, privacy-regulated world is a structural advantage worth weighing explicitly.