7 marketing measurement myths costing companies millions

This is the marketing-science leg of Brandformance, co-hosted by Pranav Piyush and Sundar Swaminathan. Every two weeks they break down the concepts behind modern marketing measurement — incrementality, marketing mix modeling, brand versus performance — in plain, practical terms, and bust a few myths along the way.

Pranav Piyush is the co-founder and CEO of Paramark, where he’s building modern marketing measurement — incrementality testing and marketing mix modeling — for growth teams. A long-time marketing leader, he hosts Brandformance.

Sundar Swaminathan is a marketing data science and experimentation advisor to consumer-tech scaleups and the creator of the ExperiMENTAL newsletter and podcast. He previously built and led brand data science at Uber — measuring over a billion dollars of brand spend — and earlier worked on the debt desk at the US Treasury.

The gist

• Most marketing-measurement “rules” are limiting beliefs — they’re the reason teams don’t experiment, not statements of fact.

• You don’t need a big budget or fancy tools to test. Turning one channel down is a real pre/post experiment; SQL and a spreadsheet go a long way.

• A channel is never “not incremental.” Your execution on it isn’t — yet. Reframe the question accordingly.

• MMM, attribution and incrementality aren’t competing for “right.” They’re different lenses you triangulate to be “less wrong.”

• Measurement is a muscle, not a finish line — and brand decays, so treat it as a painkiller you keep taking, not a vitamin.

Testing isn’t for big budgets or fancy tools

The biggest limiting belief, Sundar argues, is that experimentation requires scale, budget and complicated tooling. Humans are natural experimenters — an ancestor tasting an unfamiliar plant ran a sample-size-of-one test and updated instantly. If 75–80% of companies have one dominant channel (often Meta), turning it down and watching the trend line is, by definition, a pre/post experiment. You might land on “cut to 50%” when the true answer was 42% — but going from 100% to 50% is the win. For most of his career Sundar has gotten far on SQL and Google Sheets; if you’re hiding behind complicated tools, that’s the tell.

Pranav adds the real unlock: courage. He points to a startup selling to local governments — run radio in seven counties and measure the lift — and to True Classic publicly zeroing out spend every few months and rebuilding it. Sundar’s blunt version: a leader’s willingness to feel the pain is about 80% of why experimentation does or doesn’t happen. If you believe Meta is incremental but you’re too scared to turn it off and prove it, that’s a red flag about your culture, not your math.

Incrementality is bigger than holdouts

A head of growth told Pranav that incrementality testing didn’t fit a scale-fast mode — because he equated it with holdouts. But you can do the opposite: a pulse-up in the specific geographies your model recommends. You can test budget levels, creative, a new channel, or new bidding logic. As Sundar puts it, incrementality testing is just a subset of A/B testing — all “A and B” means is that the two aren’t the same.

The trap is all-or-nothing: if you can’t run the extreme holdout, you run nothing — even though you’d never treat email or product that way. The caveat for marketing is that signals are noisier, so your pulse-ups need to be meaningful — not a 1% nudge but more like 10–20% — because the line items are bigger.

A channel is never the thing that “doesn’t work”

“Is TikTok incremental?” is the wrong question, and so is treating a flat test as proof a channel is dead. With platforms that have a billion-plus users, your audience is there; the honest conclusion isn’t “the channel isn’t incremental” but “our efforts on this channel aren’t incremental — yet.” Getting it right means stacking variables: the creative, the audience, the frequency, a long enough window to catch people ready to buy — on top of macro and competition.

Sundar’s line: it’s impossible for a channel to not work; it’s very possible for it not to work for you. Even a consumer brand can crack LinkedIn with the right angle — Pranav’s example is the Rolex ads that kept showing up in his LinkedIn feed. If a channel isn’t working, you haven’t found the alpha in it.

MMM vs. MTA: triangulation, not a contest

The belief that MMM and MTA disagreeing means one is wrong assumes both are supposed to be perfect. They’re not. MTA tracks only digital touchpoints you can see; MMM includes offline channels, brand and halo effects — so comparing them is apples to oranges, and the disagreement is the point. Both are models you configure and feed; each returns its own version of reality.

Pranav’s analogy: a model airplane and a real airplane. A model reflects reality but is wrong in fundamental ways because it can’t hold every variable — and in marketing you can never see inside someone’s head. The answer is triangulation: MMM plus an attribution model plus incrementality testing, working together. No single piece is right; their job isn’t to be right, it’s to make you, in Sundar’s phrase, “less wrong.”

More data isn’t better — variance is

There’s no magic threshold of years or touchpoints for an MMM. What actually feeds a model is variance and volatility — spend moving up and down so the model has something to learn from; a flat spend line gives it nothing. Relevance and recency matter too: five years of data is useless if three of them predate who you are now.

Pranav’s team aims for about two years — enough to capture seasonality and hit a reasonable confidence level — but he’ll happily start a hypergrowth brand at six months if they accept wider uncertainty, because starting now beats waiting and running on pure gut. The bonus Sundar highlights: starting early also builds the organizational muscle of reading and acting on an MMM, which matters as much as the model itself.

An MMM forecasts — it doesn’t predict (and you’re never done)

An accurate MMM won’t tell you exactly where to spend your next dollar and mint money. It’s a forecasting tool, not a prediction machine — useful when the forecast period resembles the recent past, shakier in hypergrowth or disrupted categories, and only as good as the assumptions you document. Better to run three two-month forecasts with pulse-ups than one blind six-month bet.

And measurement is never “in the bank.” Marketing runs on curves: incrementality at $5M of spend isn’t incrementality at $10M, even holding everything else constant, because you slide to a less efficient part of the curve. The best teams re-run incrementality roughly every six months and recalibrate MMMs at least quarterly. It’s a muscle — what was incremental six months ago is often arbitraged away — which is why Pranav floats the idea of an “incrementality manager” whose job is to keep the team putting in the reps.

Brand decays: a painkiller, not a vitamin

Both hosts reject “brand compounds” as a free lunch. Pranav’s framing: brand is a garden, not a savings account — stop watering and it dies. Sundar wants to retire the “vitamin” framing entirely and call brand a painkiller: it prevents CAC from skyrocketing as you push for more volume, which is the language a CFO actually responds to.

His proof is a UK brand-lift study at Uber: awareness rose from ~50 to ~65 during a campaign, the budget was cut mid-flight, and within two to four weeks awareness was back to 50 — as if the campaign never happened. The half-life of brand is brutal. It does compound in one sense — show up consistently and people form associations faster, and you get better at it over time — but it decays just as surely, so the job is to keep investing, not to bank it.

Quote snacks

• “It’s impossible for a channel to not work. It’s very possible for it not to work for you.” — Sundar

• “If a channel doesn’t work, it’s because you haven’t found the alpha in it.” — Sundar

• “Let’s learn to be a little less wrong, instead of trying to be more right.” — Sundar

• “An MMM is a forecasting tool, not a prediction tool.” — Pranav

• “Every scaled marketing team should have an incrementality manager.” — Pranav

• “Stop treating brand as a vitamin and start treating it as a painkiller.” — Sundar

Why it matters

Almost every barrier the hosts name is psychological, not technical — a limiting belief that gives teams permission not to experiment. Naming them is the unlock: testing is cheap, channels aren’t verdicts, models aren’t oracles, and “done” isn’t a state measurement ever reaches.

It also reframes the brand-versus-performance debate the show keeps circling. Measured honestly, brand is a CAC painkiller with a short half-life, and performance is a curve that shifts as you scale — both demand continuous, triangulated measurement rather than a one-time answer. The throughline: be willing to feel the pain, measure to be less wrong, and keep putting in the reps.

Practical next steps

Run the simplest test you can. Turn one channel down — ideally in a few geos — and read the trend line. Don’t wait for budget or tooling.

Reframe the channel question. Instead of “is this channel incremental,” ask “have we actually cracked it” — creative, audience, frequency, window.

Triangulate, and expect disagreement. Use MMM, attribution and incrementality together; treat conflicting signals as information, not as proof one is broken.

Prioritize variance and recency over years of history. Start measuring now, accept wider uncertainty early, and set a baseline you can improve.

Treat the MMM as a forecast. Document your assumptions, prefer short horizons with pulse-ups, and re-test incrementality every ~6 months while recalibrating quarterly.

Keep watering the brand. Invest continuously to offset decay, and sell it internally as protection against rising CAC at higher volumes.