Podcast episode

Picking An MMM

by Ken Rona ·July 20, 2026 ·AdExchanger: The Big Story

attribution big-tech measurement publisher-economics

TL;DR

AdExchanger's editorial team breaks down the resurgence of marketing mix modeling (MMM) — a decades-old measurement method that aggregates channel-level data to gauge advertising effectiveness — and scrutinizes Google's Meridian and Meta's Robyn as open-source tools that conveniently grade their own homework. The episode is useful for any marketer or ad-tech operator evaluating MMM vendors, but it is essentially an extended conversation about a single editorial piece rather than a news-heavy briefing.

What was covered

Why MMM is back: Multi-touch attribution (MTA) — which tracks individual user journeys across touchpoints — has become largely infeasible as third-party data signals disappear. James Hirscher argues this has pushed marketers back toward MMM, which works on aggregated, channel-level inputs rather than user-level tracking.
Google Meridian's positioning: Google's open-source MMM framework is backed by real sales KPIs — reps are reportedly commissioned on Meridian adoption, even when the advertiser doesn't know they're running on it. It integrates with Google Analytics and Google Cloud, giving Google a quiet infrastructure layer around measurement.
Default settings favor Google: Multiple sources told Hirscher that Meridian's out-of-the-box configuration favors Google channels. Customization requires data scientists; most adopters don't do it, which means they receive a "Google-friendly" read of their media.
Meta pulling back from Robyn: Hirscher says Meta is disinvesting from Robyn. The tool originated as an internal science project and never got the dedicated sales and engineering push Meridian received. Meta is now focused on how it shows up in third-party MMM reports rather than controlling the model itself.
Amazon's MMM: honest self-interest: Amazon's MMM data feed is explicitly designed to make Amazon look good as both a media seller and a retailer — no open-source pretense. Hirscher suggests this makes it arguably more honest than Google's approach.
Meta's parallel moves to influence attribution: Meta has expanded the "hot zone" (the pixel area on-screen that registers as an ad click on Instagram/Facebook feeds), extended conversion windows, and adjusted how it passes data to Google Analytics — all designed to appear more favorably in MMM and attribution reports that historically default to search.
Vendor landscape: Lightweight MMM startups (Northbeam, Mutton Next cited) are compressing cycle times from annual to near-campaign cadence. Some sophisticated agencies use Meridian as a white-labeled base. Earlier adopters of Robyn — including one airline described by agency Adswerve as building a "Frankenstein's monster" of Robyn, Meridian, and other tools — migrated to Meridian.

Notable claims & predictions

James Hirscher: "The default off-the-shelf settings [of Meridian] are very much made to absorb Google data and favor Google — it is made to be customized, but probably a lot of people do not customize it." The implication: widespread Meridian adoption quietly tilts measurement in Google's favor at scale.
Hirscher on Amazon: Amazon's MMM product is "totally upfront about the fact that the whole point is having it make Amazon look as good as possible" — framing this candor as ironically more trustworthy than Google's transparency narrative.
Henry N. (CEO, Mutton Next), as quoted by Allison Schiff: Using Meridian for free is "like getting a free puppy, not a free beer" — you have to invest significantly in maintenance and customization or you get nothing useful out of it.
Hirscher on Google's commercial incentive: "A win for [Google's salespeople] is an advertiser using [Meridian] — and they don't even have to know it's Google Meridian." Google reportedly has real KPIs and commissioned sales staff behind an ostensibly open-source, "give it away free" product.
Hirscher on Meta's strategy shift: Meta is "focusing more now on shaping how it shows up in Google Analytics and MMM reports" rather than building its own modeling infrastructure — a retreat from owning the measurement layer to gaming whatever layer others control.

Fact check

Claim (Hirscher): Meta's Robyn was "the only open-source MMM model on the market" for roughly the first two years after its release, which he implies was around 20 years ago.
Misleading on timing. Meta (then Facebook) open-sourced Robyn in 2021 — roughly four years ago, not twenty. Google released Meridian publicly in 2024. The "only open-source model for 20 years" framing conflates MMM as a discipline (which is decades old) with Robyn as a product (which is recent). The substance of the point — that Robyn had an early-mover advantage as an open-source option — is broadly accurate, but the timeline language in the transcript is garbled and could mislead listeners into thinking the open-source MMM category is older than it is.
Claim (Hirscher): Google's Meridian sales team is "commissioned" on Meridian adoption, including cases where the advertiser doesn't know they're using it.
Unverified. This is attributed to unnamed sources Hirscher spoke to while reporting. Plausible given Google's pattern with other products (e.g., Google Analytics, Google Cloud), but it cannot be independently confirmed from the transcript and has not been publicly disclosed by Google. Readers should treat it as reported but unconfirmed.
Claim (Hirscher/Schiff): Meta expanded the clickable "hot zone" on Facebook/Instagram feed ads, deliberately inflating click counts fed into attribution and MMM systems.
True but context omitted. Meta did expand the interactive tap/click zone on mobile ads. However, the company described this as a UX improvement to reduce accidental non-clicks, not as a measurement manipulation. The effect on attribution data — and whether it is intentional gaming of MMM systems — is the editorial interpretation of Hirscher and Schiff, not a confirmed Meta admission. Worth flagging the incentive: both speakers have editorial reasons to frame platform behavior skeptically, and Meta has an obvious financial incentive to improve how it scores in attribution, but calling the hot-zone change deliberate measurement manipulation goes beyond what is publicly established.

Why this matters for ad-tech operators

Measurement neutrality is structurally compromised at scale. If Meridian becomes the de facto MMM substrate — white-labeled inside agencies and measurement vendors — Google gains quiet influence over how all cross-channel media is valued, without ever touching a DSP bid. Publishers and non-Google media sellers should be asking whether their channels are adequately represented in clients' Meridian configurations, or whether they are systematically undervalued by default model priors.
Meta's attribution moves are a defensive response to search's structural advantage. Expanding click zones and conversion windows, and adjusting Google Analytics data handoffs, are signals that Meta views MMM adoption as a threat to how social media credit is counted — and is engineering around it. Agencies building measurement stacks should audit how Meta signals flow into their models before presenting MMM outputs to clients.
The "free and open-source" framing is a growth strategy, not a public service. Amazon's forthright self-interest in its MMM product may actually be less operationally dangerous than Meridian's open-source brand equity masking default Google-favorable settings. Operators and agencies that haven't stress-tested their Meridian configurations against neutral priors should do so before presenting results to clients.
Direct impact on this episode is relatively low for buy-side trading desks and SSPs/DSPs focused on programmatic mechanics — MMM operates above the plumbing layer. The episode is most relevant to measurement leads, agency planning teams, and CMO-level decision makers choosing or auditing their attribution stack.

Full analysis

The story: measurement is quietly becoming a battleground where the two biggest ad sellers — Google and Meta — control or influence the "free" tools that grade every channel's performance. Google's open-source MMM (marketing mix modeling — a statistical method that estimates how much each ad channel drove sales, using aggregated data instead of tracking individual users) ships with default settings that flatter Google, and reps are reportedly paid on adoption. Meta is retreating from building its own tool and instead engineering how it looks inside everyone else's reports.

What's actually being decided: whether agencies, publishers, and measurement vendors let the sellers own the scorekeeping — or invest to keep it neutral.

Reversibility: Type 1 for the market (once Meridian becomes the default substrate inside agencies, it's sticky and hard to dislodge). Type 2 for any single operator (you can audit or swap your config tomorrow).

Forcing function: none acute. This is a slow-moving structural shift, not a dated event. Impact is genuinely low for programmatic plumbing (DSPs, SSPs, trading desks) and high for measurement leads, agency planners, and non-Google/Meta media sellers.

The Market Analyst — Follow the incentive, not the press release. Google giving away Meridian isn't philanthropy; it's distribution. If Meridian becomes the base layer inside agencies and measurement vendors, Google shapes how every dollar of cross-channel spend gets valued — without touching a single bid. That's a cheaper, more durable moat than winning any auction. The tell is the commissioned sales force behind a "free" product. Meta's retreat from Robyn is the revealing move: they've conceded they can't win the modeling layer, so they're gaming it instead. Plain version: the referees are now sponsored by two of the teams, and one team just fired its own referee to go lobby the other's.

The Skeptic — The load-bearing assumption is that default settings meaningfully swing outcomes at scale. Maybe — but MMM has always been analyst-dependent and priors-driven; that's not new corruption, it's how the method works. Henry N.'s "free puppy, not free beer" line cuts both ways: a tool nobody customizes produces garbage regardless of whose thumb is on the scale, and sophisticated agencies do customize. Also note both AdExchanger speakers have editorial reasons to frame platform behavior as sinister — the hot-zone change is documented as a UX tweak, and "deliberate manipulation" is interpretation, not admission. The scandal here is thinner than the framing suggests.

The Operator — Tuesday morning, this is a config-audit problem, not a strategy crisis. If your agency runs Meridian white-labeled, someone needs to actually inspect the priors before results hit a client deck — and most teams don't have a data scientist to do it. That's where it breaks first: you're presenting a Google-flattering read and calling it neutral. Second-order effect at 90 days: a client's own Google rep and your MMM output agree suspiciously often, and eventually a non-Google seller (a CTV publisher, a retail media network) challenges why they're undervalued. Have the audit trail ready before that meeting, not after.

The Customer / End User — the non-Google/Meta media seller — This is where the real loss lands. If you're a publisher, an independent CTV platform, or a retail media network, and your channels get systematically discounted by default model priors you never see, you lose budget you earned — invisibly. Nobody sends you a rejection email; you just don't get renewed. The defensive play: demand to know what MMM your agency clients use, and push for your channel to be represented with proper priors. Plain version: if the test that decides whether your ads "worked" is written by your biggest competitor, you're going to fail it a lot.

Where they part ways:

Is this corruption or just how MMM works? The Market Analyst sees a structural land-grab; the Skeptic sees a decades-old analyst-dependent method being sensationalized. Both can't be fully right — the truth hinges on how many adopters actually leave defaults untouched.
Who bears the cost? The Operator frames it as a manageable audit task; the non-Google seller frames it as existential and invisible. The gap is who has the leverage to demand the audit — big agencies do, small publishers don't.

What it hinges on: one measurable fact — what share of Meridian deployments run on or near default settings. If most customize (Skeptic's world), the "Google-friendly priors" problem is contained to the unsophisticated tail. If most don't (the reporting's implication, backed by the "free puppy" economics), then the default is the market, and Google has quietly bought the scorekeeping function.

The council leans toward the reporting being directionally right but overcooked: Meridian adoption is real and self-serving, but "manipulation" is stronger than the evidence. The move for operators isn't panic — it's to treat MMM outputs as adversarial by default and audit the priors, especially where a seller supplied the tool.

What to verify before acting: ask your measurement vendor or agency, in writing, (1) which MMM they run, (2) whether it's customized off defaults, and (3) how non-Google/non-Meta channels are represented in the priors. If they can't answer cleanly, that's your answer.

Prediction: Neither Google nor Meta will publicly disclose the default channel priors inside Meridian (or their MMM data feeds) by the end of Q1 2027 earnings season, and no major industry body (IAB, MRC, ANA) will have issued an MMM neutrality/auditing standard by then.

Confidence: Medium — Silence serves both sellers; standards bodies move slowly.

Why: The entire commercial value of a "free" MMM as a distribution moat depends on the priors staying opaque — disclosing them would surrender exactly the advantage the reporting describes, so Google has a direct incentive not to publish them. Meta's shift from owning Robyn to influencing third-party reports shows even the second-largest seller has given up on transparency as a strategy. Industry standards bodies have talked about measurement neutrality for years without shipping enforceable audit frameworks (see the multi-year MRC/currency debates in CTV), so expecting a binding MMM standard within ~two quarters cuts against the track record. The opposite outcome — voluntary disclosure or a fast new standard — would require the incumbents to act against their own economics, which is the less likely path.

Revisit by 2027-04-30: We're right if neither platform has published default-prior documentation and no major body has issued an MMM auditing/neutrality standard. We're wrong if Google or Meta publicly discloses Meridian/feed default priors, or if the IAB/MRC/ANA ships an MMM neutrality standard, before that date.

Listen to the episode →