Inside the Blurb · The MarTech Matrix

Why Your Marketing Attribution Numbers Never Agree. Marketing Measurement Triangulation Fixes It.

The big picture: Ask three different attribution tools how your marketing is performing and you’ll get three different answers. Last-click says search is your biggest driver. Your CFO doesn’t buy the number. Your incrementality test says something else entirely. None of those tools are lying — they’re all measuring something real, just not the same thing.

Madan Bharadwaj, founder and CEO of M-Squared, has built his career around marketing measurement triangulation: reconciling attribution, MMM, and incrementality numbers that never agree. He’s built four companies inside the measurement space, including Visual IQ, one of the first multi-touch attribution platforms, and Measured, the industry’s first pure-play incrementality company. In this episode of Inside the Blurb, he explains why neither approach was ever the full answer on its own, and why AI can’t manufacture the causal data brands actually need.

Why it matters: Most brands pick one measurement method and trust it completely, or stack dashboards with no framework for reconciling what they say. Marketing measurement triangulation isn’t about finding the one true number. It’s about knowing which number to trust for which decision.

---

Why Your Attribution Numbers Don’t Agree

Nearly everything brands use to measure marketing today is what Bharadwaj calls correlational. Last-click, multi-touch attribution, platform reporting, UTM tracking — each one tries to connect an impression to a conversion after the fact. That comes with built-in bias. Search looks artificially strong because people search right before they buy. Upper-funnel channels like podcasts or CTV look artificially weak because the payoff can take 45 days to show up, by which point the correlation is long gone.

“There is correlational and there is causal. All the classic attribution data sets — last click, first click, platform attribution, UTM attribution — those are all correlational techniques. It’s clearly very correlational, so you have all the biases attached to it.”

— Madan Bharadwaj

On the other side of the line sit the two causal techniques: incrementality testing, a test-and-control experiment that isolates a channel’s true lift, and marketing mix modeling (MMM), a longer-term statistical read on the same question. Neither one asks what a user clicked. Both measure whether the outcome would have happened anyway.

---

Why Incrementality Testing Is Harder Than It Looks

The concept explains itself in about fifteen seconds. The execution doesn’t. Incrementality testing means holding out a market or audience, running the campaign everywhere else, and comparing outcomes. But you’re testing on real human beings, and human behavior is noisy. A test designed to catch a 1-2% lift will drown in that noise. Bharadwaj’s teams design for a 10-15% swing instead, because anything smaller can’t be told apart from randomness.

Before a test even runs, his team works backward from the decision it needs to support: what result would change the budget, and by how much. Roughly half the time, that exercise alone answers the question and the test never runs at all. “Time is the most precious resource for everybody,” Bharadwaj said. “Spending eight weeks, twelve weeks, and finding that you got nothing — that hurts more than the actual money. People can get that back. Time is what is very hard to get back.”

---

Marketing Measurement Triangulation: Putting Every Measurement “Truth” in One Line

This is the idea marketing measurement triangulation is built around, and it starts from a blunt premise: there is no single source of truth in marketing measurement, because measurement is measuring human behavior. Incrementality testing captures a point in time. MMM captures a longer average effect. Last-click captures the very last touch. None of them is false. They’re all true in their own way, describing the same channel from a different angle.

“Triangulation allows you to put all of these truths about a particular channel, against a particular outcome, in one line. You can look at UTM attribution, platform attribution, MMM attribution, and incrementality-estimated attribution all together — and now you can use that to triangulate to an investment decision.”

— Madan Bharadwaj

In practice, that means using incrementality testing to generate a multiplier. If brand search gets credited with 40% of conversions through standard attribution, but an incrementality test shows the real causal lift is closer to 5%, that 5-over-40 ratio becomes a multiplier. Applied to daily reporting, it lets a brand estimate the causal impact of every channel continuously, without running a fresh test every day.

---

Why AI Can’t Manufacture the Data You Actually Need

Bharadwaj is direct about where AI fits into this and where it doesn’t. An AI model is only as good as the data it’s given, and causal data doesn’t already exist anywhere — it has to be built through the kind of testing and modeling described above. Point an AI model at biased, correlational data, like raw platform reporting, and it will hand back a confident answer that’s confidently wrong.

“All AI is doing is working with the data you have. It cannot create causal data — that data doesn’t exist anywhere. You have to build it. That’s the hard part.”

— Madan Bharadwaj

Bharadwaj described a recent case: an AI-native operating partner ran an automated analysis on roughly $25 million in Amazon ad spend, and the model’s recommendation came back in seconds — cut the spend entirely, because it found no measurable incrementality. Instead of accepting that, M-Squared built causal models at the individual product (ASIN) level. The real picture was far more nuanced: roughly a quarter of the products were genuinely incremental, another quarter weren’t working at all, and the rest depended heavily on context. The brand reallocated the $25 million toward what was actually working, rather than cutting it — a decision the single AI-generated answer would never have supported.

---

Why This Has to Be a Program, Not a One-Time Test

The biggest mindset shift Bharadwaj pushes brands toward isn’t a methodology — it’s a cadence. One test or one model produces one answer. A “causal insights program,” modeled after the internal customer-insights functions at companies like Amazon, treats measurement as a continuous loop: every result raises new questions, those questions get tested, and the answers compound.

“It’s not a one and done. It’s not one model, it’s not one test — it’s about building a program where you continuously learn and compound learnings on each other. You do that for three, four, five cycles, and you’ll understand your business significantly better than you do right now.”

— Madan Bharadwaj

---

Who Actually Needs This Level of Measurement

This isn’t a fit for every brand, and Bharadwaj is upfront about that. Brands spending only on Facebook and Google, at a few million dollars a year, are usually well served by standard last-click and UTM attribution — the added complexity of triangulation isn’t worth it yet. The fit changes once a brand has expanded past those two channels and hit a ceiling: rising CPMs, plateauing ROAS, and a real, expensive decision to make about where the next dollar of media spend should go.

“As you get to the point where all the low-hanging fruit is gone, you need a more deliberate, repeatable, scientific approach for driving growth consistently over many quarters. That’s where I become an option.”

— Madan Bharadwaj

---

The Bottom Line

Marketing measurement triangulation isn’t a search for one perfect number. It’s a way to get a CMO, a media buyer, a finance lead, and a CEO looking at the same evidence and making a decision together, instead of each defending a different dashboard. That kind of organizational alignment, more than any single model, is what actually moves the growth needle.

Bharadwaj also runs an open, in-person workshop on advanced attribution called Squared Collective — the next one is September 10 in Boston — where brands and publishers work through causal measurement together rather than sit through a sales pitch.

If you’re evaluating a measurement or attribution partner, check out Blurbs for vendor profiles, case studies, and the questions worth asking before you take a demo call.

Subscribe to Inside the Blurb on Spotify, Apple Podcasts, or wherever you listen. New episodes weekly.

---

More Conversations on The MarTech Matrix