Podcast episode

MMM, Unplugged

by Ken Rona ·June 24, 2026 ·Aperiam

incrementality marketing-mix-model measurement

TL;DR

LiftLab CEO John Wallace joins the Aperiam podcast to explain why marketing mix modeling (MMM) — statistical analysis tying ad spend to total revenue — needs to be rebuilt around incrementality and real-time platform signals rather than used as an annual planning exercise. The episode introduces LiftLab's new product, PlatformSense, which injects live platform data into an incrementality-based MMM foundation. Moderate relevance for ad-tech operators; most useful for measurement, agency, and brand-side audiences.

What was covered

LiftLab's origin and positioning: John Wallace founded LiftLab after selling a prior multi-touch attribution (MTA) startup roughly 10 years ago. LiftLab is built on two pillars: an automated experimentation platform and a next-generation MMM, with incrementality — measuring the marginal revenue actually caused by paid media — as the core metric.
The problem with platform ROAS (return on ad spend): Wallace argues that the ROAS figures reported inside Meta Ads Manager or Google Ads are not predictive of total company revenue. He contends that optimizing to in-platform or last-click ROAS can actively harm a brand's growth and that "incremental ROAS" (iROAS — how much total revenue is generated per dollar spent) is the only number worth bringing to a CFO.
Incrementality experiments as "step functions": LiftLab's platform can spin up geo- or audience-based holdout experiments on a single channel within a day. Wallace describes these as producing large, discrete jumps in understanding — a brand thinking it has a ROAS of 5 discovers it is actually 3, freeing budget to redeploy elsewhere.
Skims / TikTok case study: LiftLab ran an incrementality experiment for Skims on TikTok. Results showed TikTok was driving more revenue than platform metrics implied, leading Skims to triple its TikTok spend. Wallace cites this as a model for how experiment-validated measurement translates directly to budget reallocation.
PlatformSense — new product announcement: LiftLab announced PlatformSense at a recent customer event in New York. It injects real-time, in-platform signals (creative performance, delivery changes, CPC fluctuations) into the incrementality-based MMM so the model updates daily rather than being "blind" to what changed on Monday versus Tuesday. Wallace positions this as superior to the common in-house shortcut of applying a flat "haircut" factor (e.g., multiplying Meta numbers by 0.7) to estimate incrementality.
Human-in-the-loop stance on automation: LiftLab can technically write code to push budget changes directly to ad platforms, but Wallace says clients spending seven-to-eight figures monthly per month still want human approval before changes execute. He notes automated bidding is already happening at the SMB (small and medium business) level.
Upper-funnel measurement as the persistent industry gap: Wallace describes a recurring CMO failure cycle — brand campaign launched, metrics not visible in Google Analytics, surveys show a 2-point unaided recall lift that can't tie to the P&L, budget gets cut, CMO exits. Incrementality applied to top-of-funnel spend breaks this cycle by giving brand spend a revenue-linked defense.

Notable claims & predictions

Wallace: "The more they optimize the media plan to a last-click or in-platform ROAS number, eventually the more harm they're going to do the firm. You're going to stunt the growth of the firm if the measurement's leading in the wrong direction."
Wallace on walled-garden measurement: "The measurement that [ad platforms] produce is certainly going to shed them, put them in the best light possible… there's no real incentive to share bad news about a campaign."
Wallace on MMMs' structural blind spot: "MMMs from their design have one flaw — they're treating all impressions as created equal… The ads I bought on Monday were not identical to the ads I bought on Tuesday. Something changed in the delivery… that should change the answer."
Wallace on full automation of budget decisions: "The idea that you can replace corporate marketing purely with an algorithm — I don't think we're anywhere close to that [on large brands]. Nobody wants to be a Wall Street Journal headline about how they wasted $10 million by accident."
Wallace's practitioner advice (summer 2026): "Be the marketer that's on the bleeding edge… consider BAU [business as usual] as the enemy." He frames continuous experimentation — testing assumptions made a year ago, since platform dynamics shift — as the defining trait of top-performing marketing organizations.

Why this matters for ad-tech operators

Measurement monetization risk for DSPs and SSPs: LiftLab's explicit pitch is that self-reported platform ROAS is structurally unreliable. As incrementality-based tools gain adoption among mid-market and enterprise brands, ad platforms (including programmatic pipes) face pressure to justify spend against a stricter revenue-causation standard — not just efficiency metrics they control.
PlatformSense signals a new integration layer: By pulling real-time signals from ad platforms into an independent MMM, LiftLab is building a data integration surface that sits on top of The Trade Desk, Meta, Google, and others. At scale, this type of layer could influence how agencies and brands allocate across channels week-to-week without going back to each platform's native optimization tools.
Upper-funnel spend reallocation opportunity: Wallace's description of the "CMO brand-campaign death cycle" is a structural issue well known to CTV and streaming publishers competing for brand dollars. The argument that incrementality measurement can give top-of-funnel spend a defensible P&L tie-in is a direct tailwind for publishers trying to shift budgets away from performance-only channels.
Human-in-the-loop as a temporary condition: Wallace acknowledges the code to auto-push budget changes to platforms already exists, and that automation is already live for SMBs. The implication for agencies and trading desks is that the timeline for AI-driven budget reallocation at enterprise scale is a matter of organizational risk tolerance, not technical readiness — meaning competitive pressure to automate is building.

Full analysis

Decision Council: MMM, Unplugged

Step 1 — Frame

The story: A measurement vendor (LiftLab) is making the case that the ROAS numbers ad platforms report on themselves are unreliable, and that incrementality — measuring the revenue a dollar of ad spend actually caused versus what would have happened anyway — should become the standard. Their new product injects live platform data into a marketing-mix model so it updates daily instead of yearly.

What's actually being decided (for the reader): Not "should I buy LiftLab." It's "how fast is the measurement standard shifting from platform-reported efficiency to independent incrementality, and what does that do to my business?"

Reversibility: Type 2 for any single operator — you can pilot incrementality measurement and walk away cheaply. The industry-level shift is closer to Type 1 once big brands re-anchor their budget logic.

Forcing function: None hard. This is a slow-burn directional story, not an event. The honest read: moderate-to-low impact this quarter, real direction over years.

No clarifying questions needed.

Step 2 — The Council

The Market Analyst Incrementality is not a new pitch — it's the oldest religion in measurement, and it has never crossed the chasm to default status. The case study tells you why it stays niche: Skims tripled TikTok spend after one test. Vendors only publish the wins. For every "you're actually a 3 not a 5" that frees budget, there's a quiet result that just shrinks someone's plan. The interesting tell is the partnership geometry: an independent layer sitting on top of Meta, Google, and The Trade Desk. Plain version: someone's trying to become the neutral referee between brands and the platforms grading their own homework — lucrative if it works, structurally resisted by everyone being graded.

The Skeptic The load-bearing assumption is that brands want the truth. They don't — they want a defensible number. A CMO whose ROAS just dropped from 5 to 3 now has to explain to the CFO why the old number was fiction. That's a career risk, not a feature. Second: geo holdout tests are noisy, expensive to run continuously, and brittle when you have ten channels interacting. "Update the model daily" sounds great until the daily answer wobbles and nobody trusts it. Plain version: the product solves a problem most buyers are quietly incentivized not to solve too loudly.

The Operator Tuesday morning reality: someone has to own these experiments, read them, and fight the channel teams whose budgets they threaten. The Meta lead does not want a holdout test proving Meta is a 3. What breaks first is organizational, not technical. The "flat 0.7 haircut" Wallace mocks exists because it's cheap, legible, and nobody fights about it. At 90 days, the realistic outcome isn't transformation — it's a parallel reporting layer that the planning team consults and then largely ignores when it contradicts the platform dashboards everyone's bonus runs on.

The Customer / End User — the publisher / CTV seller This is the one genuinely useful thread for sellers. The "CMO brand-campaign death cycle" — launch brand spend, can't tie it to revenue, get cut, get fired — is exactly why upper-funnel and CTV dollars are fragile. If incrementality gives brand spend a revenue-linked defense, it's a tailwind for anyone selling reach and attention rather than last-click. Plain version: if a streaming or brand-media seller can show its impressions caused sales, it stops competing on the terms set by Google and Meta's own scoreboards.

Step 3 — The Tensions

Truth vs. defensibility. The Analyst and the Customer see a real shift toward causation-based measurement. The Skeptic and Operator say the buyer's actual incentive is a defensible number, not an accurate one — and incrementality often delivers an uncomfortable one. This is the whole ballgame.
Independent layer vs. captured layer. Is an over-the-top measurement layer durable, or does it get squeezed the moment Meta/Google decide to degrade data access or ship "good enough" incrementality natively? Walled gardens have killed neutral referees before.
How fast does enterprise automation arrive? Wallace says the code to auto-push budgets exists; only risk tolerance holds it back. The Operator says org friction is the moat, and it doesn't melt on a tech timeline.

Step 4 — Synthesis

What this hinges on: whether brand-side demand for incrementality is pull or push. If CFOs start demanding causation proof for media budgets, the standard shifts and independent measurement layers win. If it stays a vendor push against buyers who prefer comfortable numbers, incrementality stays a respected niche — as it has for a decade.

The council leans modest. This episode is a well-argued restatement of a true thing the industry already half-believes and mostly doesn't act on. The one durable signal for operators: upper-funnel and CTV sellers should lean into incrementality framing — it's the most credible defense brand budgets have, and it reframes the contest away from walled-garden scoreboards.

What to verify before betting on it: are CFOs (not measurement teams) citing iROAS in budget reviews? That's the tell that pull has arrived. Until then, treat it as direction, not event.

Step 5 — The Prediction

The honest confidence here is low. Incrementality's adoption curve has been "slowly, then slowly" for ten years, and nothing in this episode is a forcing function that breaks that pattern on a checkable timeline. The interesting claims (independent measurement layers winning, enterprise budget automation arriving) are real but un-datable from this source.

No high-conviction prediction this week.

Listen to the episode →