.svg)
Why "Trustworthy Measurement" Is the Real Marketing Foundation (Not Another Attribution Model)
The short version: Most marketing teams don't have a measurement problem because they picked the wrong attribution model — they have one because they inherited a model no one chose on purpose, never audited whether it could be trusted, and built years of budget decisions on top of it. The fix isn't a better single model; it's treating measurement as a trustworthy foundation you audit first, triangulate across multiple imperfect methods (multi-touch attribution, media mix modeling, incrementality testing), and keep maintaining. As Kelly Zeitlow puts it, the dream of one "single source of truth" is gone — the best teams triangulate imperfect sources and treat measurement as a continuous investment.
📺 This is a recap of our fireside chat with Kelly Zeitlow. Prefer to watch? See the full conversation here →
Most marketing teams don't have a measurement problem because they picked the wrong attribution model. They have one because they inherited a model nobody chose on purpose, never audited whether it could be trusted, and then built years of budget decisions on top of it.
That's the pattern Kelly Zeitlow — a marketing executive who has built and rebuilt measurement systems at Booking.com, Gorillas, and HeyJobs, and is now VP of Marketing at Spread Group — has seen repeat itself across nearly every company she's worked with. In a recent Adasight fireside chat, she walked through how measurement has evolved, why "one source of truth" quietly became a myth, and what it actually takes to build something a team can trust.
Start With an Audit, Not a New Model
When Zeitlow walks into a new company, she doesn't start by picking an attribution model. She starts by checking what's already there — what she calls a "data strength" audit.
It runs on two tracks. First, she looks at the data herself: are the KPIs accurate, visible to the right people, and actually available when they're supposed to be? Do the dashboards marketers and analysts use daily line up with the ones executives use for strategy, or do they quietly conflict? Does the picture cover the whole customer journey, or is a chunk of it structurally missing?
Second, she interviews the people using the data. What's the KPI they actually act on? Do they trust it? Is data supposed to arrive by 9am but really show up the next day? What have they tried before that didn't work, and why do they think it failed?
She's no longer surprised when two people at the same company give completely different answers to these questions. That gap — between what people believe about their measurement and what's actually happening — is itself the diagnosis.
Every Attribution Model Has a Bias Built In
Over her career, Zeitlow has worked with the full range of measurement approaches: first touch, last touch, multi-touch, the platforms' own attribution (Google's data-driven attribution, Meta's version), lift studies, media mix modeling, and incrementality testing. Her conclusion after seeing all of them: every model optimizes for something specific, and quietly discounts everything else.
Click-based models — first touch, last touch, multi-touch — are trackable, which is their appeal. But they systematically favor the channels that drive clicks, especially bottom-of-funnel channels like paid search. They undercount view-based channels like organic and paid social and influencer marketing, they miss retention and CRM programs, and they leave out offline entirely — out-of-home campaigns, flyers, anything that doesn't generate a click.
On top of the model bias, privacy regulation is shrinking what any click-based model can see in the first place. Apple blocks cookies by default across its browsers and devices. GDPR keeps getting stricter across the EU. If your measurement relies on cookies, you're structurally missing a growing share of your audience every year — not because your model is wrong, but because the data it needs is increasingly unavailable.
For years, Zeitlow says, the best digital measurement teams believed they could eventually reach a single source of truth — granular, individual-level tracking of every customer. That goal is gone. What's replaced it is a shift Google itself has acknowledged: instead of chasing one perfect source, you triangulate multiple imperfect ones.
Media Mix Modeling and Incrementality Testing Fill the Gaps
Two methods do most of the work of filling in what click-based attribution misses.
Media mix modeling (MMM) uses aggregate spend and outcome data instead of individual clicks, which means it can capture channels that click-based models structurally undercount — paid and organic social, influencer, CRM, and offline. It also gives marketers a simulation tool: model what happens if you move $20,000 from one channel to another before you actually do it. It's not perfect, but it shows directionally what will move.
Incrementality testing — geo-splits and lift studies that turn a channel off for a period, usually in a specific region — is, in Zeitlow's experience, the most underused and most powerful tool available. It's the only method that shows what you'd actually lose without a channel, rather than what a model credits to it. It's also costly, since you're deliberately sacrificing performance for a period to learn something, which is exactly why most companies skip it. She recommends only running incrementality tests once you're live with four or five channels — with just one or two, you'll simply lose the spend rather than learn anything useful from it.
You Don't Need Every Method on Day One
For a company just getting started, Zeitlow's advice is not to copy another company's stack wholesale — there's no one-size-fits-all setup, because businesses look different in what channels matter, how much offline plays into the mix, and how far out purchase decisions get made.
Instead, she recommends adding methods as the channel count grows. Start with multi-touch attribution rather than a single-source model like first or last touch — it's more work, but it captures more of the real path to conversion. Once there's real spend across a handful of channels, add media mix modeling, using open-source tools or an agency to get it running. Once the business is live with four or five channels, add incrementality testing.
Two more calibration questions matter before locking in a model. The first is the attribution window: set it close to how long it actually takes someone to decide to buy. If that's typically one to two weeks, a week-long window keeps data fast enough for marketers to act on, even if longer-tail behavior needs to be modeled separately. The second is metric focus — is the business actually optimizing for new customers, or for purchases? Whichever it is, the measurement setup needs to be built around that goal, not just whatever happens to be easiest to track.
"CPCs Are Up" Is a Symptom, Not a Cause
One pattern shows up again and again in Zeitlow's conversations with struggling marketing teams: leadership says, "we're struggling because CPCs are going up." She's heard some version of this line from multiple companies, and to her it's never actually the explanation — it's a symptom sitting on top of a deeper problem.
Dig one level down, and the real causes tend to be structural. Maybe the marketing mix isn't diversified enough, so spend is concentrated in one channel and that's exactly where CPCs are getting bid up. Maybe the business hasn't invested enough in brand and organic, so competitors are capturing a larger share of voice and paid costs are rising as a result.
There's often a measurement illusion mixed in as well. Under last-touch attribution, retargeting frequently looks like a top-performing channel, because it's the last thing a converting customer clicked before buying. Run an incrementality test on it, though, and it's common to discover that a large share of those customers would have converted anyway — organically, through CRM, or by going straight to the site. The ad never needed to run.
This is where measurement stops being a reporting exercise and starts shaping actual strategy. With media mix modeling in place, brand spend becomes something a team can build a real business case for, instead of defending on instinct alone — without it, brand investment is genuinely hard to justify, because it simply doesn't show up in a click-based dashboard. And when budget gets tight and leadership asks where to cut, teams with real measurement can say exactly what a 10% cut to a given channel will cost. Without it, as Zeitlow puts it, "it's only opinions" — and the loudest voice in the room tends to win.
Measurement Is Never Finished
Zeitlow's closing point is one that's easy to nod along to and hard to actually act on: measurement isn't a project you set up once and leave alone. New channels launch, new tactics work until they don't, new products come to market, and privacy rules keep tightening. Treating measurement as a one-time setup guarantees it goes stale.
Nobody has fully solved this — not Zeitlow, not the companies she's worked with, not the platforms themselves. But starting today, even imperfectly, beats waiting for a cleaner moment that never arrives. The companies that end up best-in-class aren't the ones with a perfect model. They're the ones treating measurement as a continuous investment, adding one puzzle piece at a time until the picture is clear enough to make real decisions on.
Sourced from the Adasight fireside chat with Kelly Zeitlow, marketing measurement expert and VP of Marketing at Spread Group, formerly at Booking.com, Gorillas, and HeyJobs. September 2026.
Watch the full fireside chat
This article captures the highlights, but Kelly walks through the whole evolution of marketing measurement — the audit, the model biases, MMM and incrementality, and the "CPCs are a symptom" insight — in the full conversation.
▶ Watch "Marketing Measurement: Yesterday, Today, and Tomorrow" →
Want to put Kelly's first move into practice on your own setup? Start with our free Growth Gap Assessment — three minutes, no email needed — or our Data Stack Audit to pressure-test where your measurement can and can't be trusted.
Frequently asked questions
What is trustworthy measurement in marketing?
It's measurement you've deliberately chosen, audited, and can actually rely on for decisions — as opposed to an inherited attribution model no one verified. Kelly Zeitlow argues it starts with a "data strength" audit (checking whether KPIs are accurate and trusted, and whether teams even agree on them) rather than with picking a model.
Why isn't there a single source of truth for marketing measurement anymore?
Because privacy changes (Apple blocking cookies by default, tightening GDPR) mean click-based tracking structurally misses a growing share of your audience, and every attribution model has built-in bias. The goal has shifted from one perfect, granular source to triangulating several imperfect methods — a change Google itself has acknowledged.
What's the difference between media mix modeling and incrementality testing?
Media mix modeling uses aggregate spend and outcome data to estimate each channel's contribution (and simulate budget shifts before you make them), capturing channels click-based models undercount. Incrementality testing turns a channel off — usually a geo-split — to reveal what you'd actually lose without it. MMM shows direction across everything; incrementality proves causation for one channel at a time.
When should you add incrementality testing?
Kelly Zeitlow recommends waiting until you're live across four or five channels. With only one or two, you'll just lose the spend you're sacrificing rather than learn anything useful — the test needs enough of a channel mix to produce a meaningful read.
Are rising CPCs really why marketing teams struggle?
Usually not — Zeitlow treats "CPCs are up" as a symptom, not a cause. The real drivers tend to be structural: an under-diversified channel mix, too little investment in brand and organic, and a measurement illusion where last-touch makes retargeting look better than it is. Incrementality testing often reveals many of those conversions would have happened anyway.
