Insights
Why you can't fully prove the ROI of brand marketing
You can't fully prove the ROI of brand marketing — not causally, and not completely. Why geo tests, platform lift studies, brand lift studies, MTA and MMM each fall short, and how to build a credible case anyway.
Sundar Swaminathan
Marketing Science Advisor
Why you can't fully prove the ROI of brand marketing
The hardest measurement challenge in Marketing is proving the ROI of Brand Marketing. After building Uber's Brand Marketing Science team, I've spent years arguing that you can prove it.
Over time, though, something started bothering me. I think I finally figured out what it is:
I’m a liar.
The truth is, you can't fully measure the ROI of Brand Marketing.
At least not causally.
And definitely not completely.
Below, I'll share why Brand Marketing is hard to measure, and then I'll look at the tools we often have at our disposal and why they fall short.
The structural challenges
Brand Marketing isn't difficult to measure because measurement has failed. It's difficult because Brand itself isn't structurally designed to be measured cleanly. Unlike Performance Marketing, Brand accumulates over time, spills across channels and markets, and changes the very baseline you're trying to measure.
1. There's no true counterfactual
With Performance Marketing, you can always run a holdout. Turn off spend in a market, measure the difference, and compare it against a control. Often, that difference represents a short window of impact. How your performance marketing did 5 years ago doesn’t impact how it does today.
Brand doesn't have a natural baseline.
If you've been investing in Brand Marketing for years, you can't observe what would have happened if you'd never started. There's no clean pre-Brand period to compare against. Any test you run today is only measuring the marginal lift on top of an already elevated baseline, not the true causal effect.
You're not answering: "What did Brand Marketing do?"
You're answering: "What did the last dollar of Brand Marketing do, on top of a decade of Brand investment?"
Very different question.
Another complexity is that time actually makes the problem worse. The stronger your Brand becomes, the harder it is to measure. Your "off" state is already contaminated by years of previous Brand investment. There is no longer a world in which your Brand doesn't exist.
2. Brand doesn't respect experiment boundaries
Even if you design a great experiment, keeping treatment and control groups isolated becomes incredibly difficult. Brand activity inevitably spills into your control group. The national media doesn't respect test boundaries. People travel. Social sharing doesn't stop at the county line.
The more famous your Brand becomes, the worse this spillover gets. Your control group is no longer truly "unexposed." Your lift estimate becomes biased downward, and you can never fully correct for it.
3. Brand accumulates over time
There's a third challenge that's even harder to solve.
Even if you could design a perfect causal experiment with perfectly isolated treatment and control groups, you still wouldn't know when to measure the results.
Brand exposure today might influence a purchase in three months. Or eighteen months. Or years later.
Sure, you could estimate, but the truth is, a dollar today does influence behavior over time. No causal test will be able to completely capture that, so any causal measurement creates a narrower view of the impact and, thus, undervalues it.
The lag between impression and transaction is real, variable, and ultimately unknowable. Measure too early, and you'll underestimate Brand's impact. Measure too late, and too many other variables have changed to confidently attribute the result to Brand Marketing.
In other words, the challenge isn't just proving causality, it's knowing how long Brand continues to influence future behaviour. That's a fundamental limitation of Brand measurement, not a limitation of any specific measurement tool.
What would the ideal brand measurement system look like?
If Brand Marketing has no true counterfactual, spills across markets, and accumulates over time, then the ideal measurement system would need to have three characteristics:
Long-term: able to measure Brand over many years.
Complete: able to capture the impact of Brand across every channel.
Collective: able to measure the total effect of years of Brand investment, not just the marginal impact of your latest campaign.
No causal measurement methodology can satisfy all three requirements. Each one solves part of the problem while introducing its own limitations. Let's look at the best tools we have today.
Every causal measurement method falls short
Geo tests
Geo tests are the closest thing we have to a randomized controlled trial for Brand Marketing. But they're notoriously hard to maintain because both your treatment and control groups have to hold all assumptions steady throughout the test. However:
Promotions land.
Pricing changes.
Competitors enter.
Conversion rates shift.
Regulatory changes happen.
Any of these can break the relationship between treatment and control.
That's why most geo tests run for less than six months. Beyond that, the assumptions required for a clean experiment become increasingly difficult to maintain.
Maybe that's enough for your business case, maybe not. But even in the best-case scenario, you've never proven what Brand contributes over two or three years.
And even if you could design an 18-month geo test, there's another practical constraint: you can't realistically hold an entire market off Brand spend for that long. Leadership won't tolerate it, and neither will the business.
Geo tests are excellent for proving short-term causal lift. But they're structurally unable to prove the long-term cumulative value of Brand.
Platform lift tests
Platform-level Brand experiments, such as Meta or YouTube conversion lift studies, are among the cleanest experiments you can run. You get properly randomized treatment and control groups, you don't have to manage geographic holdouts, and you can often get results within a few weeks.
The limitation is completeness.
A Meta Brand lift study only measures Brand activity on Meta. A YouTube experiment only measures YouTube. What you're measuring is the impact of Brand exposure on one platform, at one point in time, for one audience inside one ecosystem. Extrapolating it across every other marketing channel quickly becomes difficult because no two channels operate the same.
In addition, Brand works through repetition. The whole premise is that seeing your message across multiple channels, over time, compounds in the consumer's mind. A single-channel test captures a sliver of that and by design, it can't see the complete effect.
Brand lift studies
Brand lift studies measure causal changes in attitude, such as awareness, consideration, and purchase intent.
They're useful because they can be run repeatedly over long periods of time. Companies often run one study before a campaign and another afterwards to estimate campaign lift. Larger organisations may continuously track these metrics through ongoing Brand health studies.
But brand lift studies have two limitations.
The first is that you're comparing snapshots, not measuring a continuous trend.
A viral moment, a PR story (or crisis), a competitor's mistake. Any of these can influence Brand metrics independently of your Brand Marketing. The individual Brand lift study may be causal, but the comparison across multiple studies over time is not.
Good old correlation, not causation.
The second limitation is that Brand lift studies measure attitude, not revenue. Finance teams can't book a six-point increase in aided awareness on the P&L. And if you've ever watched a CFO's face while trying to explain how a lift in consideration translates into next quarter's revenue, you know how that goes.
What about tools that estimate, rather than prove?
So far, we've looked at causal measurement methods. But what about the tools most marketing teams use day to day? Unlike geo tests or platform experiments, these tools don't aim to prove causality directly. Instead, they estimate the contribution of Brand using historical data and statistical models.
MTA
Multi-Touch Attribution (MTA) is the easiest one to discuss because it is such a terrible tool for measuring Brand Marketing. It systematically underestimates Brand's contribution because it looks at click-based customer journeys, while Brand Marketing is almost always designed not to generate immediate clicks.
Brand creates demand long before someone clicks on an ad. MTA only sees the click.
Marketing mix modeling (MMM)
MMM is the most credible estimation tool that most Marketing leaders reach for when trying to understand Brand ROI. And for good reason. Unlike MTA, MMM looks across channels and attempts to estimate how marketing contributes to business outcomes over time.
But it's still solving a fundamentally difficult problem.
MMM is trained on historical data, which means it can't directly observe the cumulative Brand equity that's been built over many years. Instead, it estimates that contribution based on historical relationships. Like every statistical model, MMM is also subject to modelling challenges such as endogeneity, multicollinearity, and other algorithmic constraints.
Here's a good example:
You run a TV campaign, and branded search increases. Now you have multicollinearity because TV and branded search are moving together. But you also have endogeneity because higher demand causes more branded searches, which increases branded search spend. These modelling challenges aren't unique to Brand Marketing, but they become much more pronounced when you're trying to estimate long-term Brand effects.
MMM also has practical time limitations. Like any model used to support business decisions, it’s unreliable to predict far into the future. But, as we know, Brand compounds over time. Even in a best-case scenario, you're really only going to be able to estimate the impact of sustained Brand investments over a year, maybe two. After that, there are too many assumptions that break.
So what should marketing leaders do?
None of this means Brand measurement is useless.
It means we need to be honest about what we’re measuring and why.
Rather than searching for one perfect measurement system, build confidence by combining multiple sources of evidence:
Use geo tests to prove six-month ROI and make the case for always-on spend
Use BLS to show that Brand is moving upper funnel metrics and earn the right to connect that to lower funnel over time (studies + MMM)
Use MMM as a directional signal, not a causal answer. This is particularly useful to explain the “baseline” and how it continues to trend higher than expected
Use proxy metrics such as branded search and direct traffic to help build a broader narrative around Brand performance over time.
Each method answers a different question. Together, they provide a much more complete picture than any one methodology alone.
What Marketing leaders should stop doing is overpromising. Telling your CFO you've "proven the ROI of Brand" when what you've actually proven is six months of geo lift in one market is one of the reasons Brand budgets get cut.
The more honest conversation sounds like this:
We have strong evidence from multiple complementary measurement approaches. None of them fully prove Brand ROI on their own, but together they tell a consistent and credible story.
That's ultimately how Brand should be measured. Not through one perfect number, but through multiple methods that converge on the same conclusion.
While you're here
Find your edge
Book a demo now and you'll get a live, expert-run walk through of how Paramark can help you with incrementality measurement and experimentation.
