Seventy-five percent of algorithm-driven ad spend by 2028. That's the number dentsu's midyear forecast puts on the table, and it should reframe every budget conversation you're having this quarter. The implication is blunt: if your measurement stack can't explain what those algorithms actually caused, you're flying blind into a trillion-dollar market.
Global ad spend is projected to hit $1.26 trillion in 2026, with U.S. buyers expecting 9.5% year-over-year growth according to the IAB's January outlook. The tailwinds are real: FIFA World Cup, Winter Olympics, U.S. midterms, and an AI investment cycle that WPP Media calls the "21st century's Gold Rush." But growth without measurement clarity is just expensive noise. That's where the second headline matters more than the first.
The Measurement Gap Finance Keeps Asking About
CMOs have been promising attribution clarity for a decade. CFOs stopped believing around year three. The disconnect isn't philosophical; it's operational.
Multi-touch attribution coverage has collapsed to 30–60% of user journeys after iOS App Tracking Transparency, Safari ITP, and GDPR consent flows gutted identity resolution. Meanwhile, 46.9% of U.S. marketers plan to increase investment in marketing mix modeling over the next year, and 27.6% now rate MMM as the most reliable methodology, ahead of MTA's 19.4%.
The shift isn't nostalgia for 1970s econometrics. It's a pragmatic response to signal loss. MMM uses aggregate data, needs no cookies or device IDs, and can incorporate offline channels, pricing, and macroeconomic factors that MTA ignores entirely. For B2B companies with 90-day sales cycles and multi-stakeholder buying committees, that holistic view is the only one that maps to how revenue actually closes.
Google's Meridian: Open-Source MMM With Incrementality Baked In
Google's Meridian launch in January 2025 removed the six-figure consulting engagement that once gated MMM to enterprises. The framework is open-source, Bayesian, and designed to integrate with Google's ecosystem: YouTube reach and frequency data, Google Query Volume as a control variable, and direct BigQuery pipelines. Over 20 measurement partners are now certified on the platform.
The more consequential release came at Google Marketing Live 2026: Meridian GeoX. This is Google's answer to the calibration problem that has plagued MMM for decades. Traditional models measure correlation; they can't prove causation without external validation. GeoX runs geo-holdback experiments, compares treatment regions against control, and converts those causal results into Bayesian priors that anchor your MMM coefficients.
The integration matters because it closes the loop between strategic budget allocation (MMM's strength) and tactical proof of incrementality (what experiments deliver). Google's documentation describes native multi-cell execution that tests multiple treatments against a common control in a single study, reducing both cost and time-to-insight. For teams that have been running MMM and incrementality tests as separate workstreams with separate vendors, this is a consolidation play worth modeling.
What This Means for Your 2027 Planning Cycle
The practical question isn't whether to adopt MMM. It's how to structure the measurement stack so Finance trusts the outputs. Here's the framework I've seen work across PE-backed and public companies:
Layer one: MMM for portfolio-level allocation. Use Meridian, Meta's Robyn, or a commercial vendor to model channel-level ROI, saturation curves, and diminishing returns. This layer answers the CFO's question: "Where should the next marginal dollar go?" Update quarterly, not annually. Measured's 2026 guide reports that brands implementing causally-calibrated MMM see 10–30% efficiency gains within year one.
Layer two: Incrementality experiments for causal calibration. Run geo-holdbacks or time-series tests on your highest-spend channels. Feed those results back into MMM as priors. This is what Meridian GeoX automates, but the principle applies regardless of tooling. Without experimental calibration, your MMM is a sophisticated guess.
Layer three: Platform attribution for tactical optimization. MTA and platform-reported ROAS still have a role, but it's narrower than it was five years ago. Use them for in-flight creative rotation, bid adjustments, and audience refinement. Do not use them to justify budget allocation to the board.
The CFO Conversation You Should Be Having
Gartner's 2026 CMO Spend Survey pegs marketing budgets at 7.7% of revenue, flat from 2024 and down from the 11% peak in 2021. The implication: every dollar is under scrutiny, and "we think this channel works" no longer survives the budget review.
The winning posture is to present three scenarios with explicit assumptions, sensitivities, and risks. Show what happens if you cut 20% from paid social and reallocate to retail media. Show the CAC payback delta. Show the confidence interval on your MMM coefficients and explain which channels have experimental validation versus which are model-only estimates.
Finance doesn't need persuasion. Finance needs a model they can stress-test. If your measurement stack can't produce that model, you're not ready for the conversation.
Where the Puck Is Going
Google's AI-powered ad formats, announced at Marketing Live 2026, are already reshaping how ads appear in AI Mode and Search. Conversational Discovery ads, Highlighted Answers, and AI-powered Shopping ads all use Gemini to generate creative tailored to user queries. The measurement implication: attribution will get harder, not easier, as ad surfaces multiply and user journeys fragment further.
The response isn't to abandon measurement. It's to anchor it in causal inference. MMM calibrated with incrementality experiments gives you a defensible answer when the CFO asks why you're spending $4 million on a channel whose platform-reported ROAS looks soft. The answer is: "Because the geo-holdback showed 2.1x incremental lift, and the MMM coefficient is within the 95% credible interval of that experimental result."
That's the language that survives the board room. Model or it didn't happen.
