What Causal Measurement Actually Says About Brand Spend

There is a moment in most measurement migrations, usually somewhere in the second quarter, when the model says something the team did not expect.

It is almost always the same thing.

The brand campaign that nobody could justify is doing more than anyone thought. The bottom-funnel channel that has been the pride of the quarterly review is doing considerably less. And the recommended reallocation is to move money toward the thing with the weakest apparent evidence and away from the thing everyone has been celebrating.

I want to write about that finding, because I think it is the most counterintuitive thing in this field and the most consistently misunderstood.

The assumption almost every marketer arrives with is that rigorous measurement is a weapon pointed at brand. That the more precisely you measure, the more money flows to whatever can be proven fastest, and the more the unprovable slow work gets squeezed. That assumption is completely reasonable. It is also, in my experience, backwards — and the reason it is backwards tells you something important about what went wrong in the first place.

The expectation is not irrational. It is a memory of what actually happened.

For fifteen years, marketing measurement in practice meant attribution, and attribution genuinely did defund brand. I have written about the mechanism before and I will restate it briefly because it is the load-bearing point of this issue.

Attribution works by allocating credit for a conversion among the touches that preceded it. It is an instrument built to answer a specific question: which interactions were near the purchase?

Now consider what a brand campaign does. It raises baseline demand across an entire market over months. Its effects are diffuse, delayed, and largely invisible in the days before any individual purchase. The person who buys in November because of something they absorbed in July does not arrive carrying a token that says so.

So attribution looks at the brand campaign and returns approximately nothing. Not because the effect is absent. Because the instrument is answering a question the effect does not fit.

Fifteen years of budget meetings ran on that number. The result was exactly what Rory Sutherland describes when he argues that measurement culture systematically defunds whatever it cannot see — a steady migration of budget toward the measurable and away from the effective, dressed up as accountability. Les Binet and Peter Field spent years documenting the same drift from the effectiveness side, and their The Long and the Short of It remains the clearest account of what it cost.

So when a CMO assumes that better measurement means less brand spend, they are not being unsophisticated. They are extrapolating from the only measurement they have ever been shown.

Here is what changes when the instrument changes, and it happens in two directions at once — which is the part people miss.

Brand becomes visible. A marketing mix model with properly specified carryover does not require an effect to appear near the conversion. It can express the idea that spend in July continues to produce outcomes in September and October at a decaying rate. That is not an exotic technique; it is the standard structure of the method. Once that structure exists, the effect that attribution was returning as zero appears with a real coefficient attached.

Bottom-funnel channels deflate. At the same time, the channels sitting closest to the conversion — branded search, retargeting, parts of affiliate — get their first honest read, and it is frequently well below the attributed number. These channels were the primary beneficiaries of the old instrument, because proximity to purchase is exactly what the old instrument rewarded.

Both corrections land in the same quarter. The reallocation that falls out of them is therefore consistently directional: more brand, more mid-funnel, less bottom-funnel duplication. In our customer base I have watched that direction hold with a consistency I did not anticipate when we started building this.

And once you see it, the mechanism underneath is almost embarrassingly simple.

Brand creates demand. Bottom-funnel harvests it.

A great deal of what gets called performance marketing is not creating anything. It is collecting demand that already existed, at the moment it was going to convert anyway, and then reporting the collection as creation. Branded search is the purest case — the customer searched your name, which means the work was already done, and the ad intercepted a purchase that was going to happen. Retargeting frequently does the same at scale.

This is why the two corrections are not independent. The performance numbers look excellent because the brand work is functioning. Cutting brand to fund performance produces a very good quarter and a materially worse year, because you have reduced the thing that generates the demand while increasing the thing that collects it. Eventually there is less to collect, and the harvesting channels start showing declining returns that nobody can explain from inside the harvesting channels.

I have now watched several companies discover this in the wrong order — first the great quarter, then the unexplained decay, then the migration, then the realisation of what the earlier decision cost.

If I stopped there this would be an advertisement, so I want to be precise about what this evidence does and does not support.

The direction is much more reliable than the magnitude. That a well-run migration finds brand under-invested is something I have seen consistently. How much under-invested is a far less confident number in most businesses. Brand effects are diffuse by nature, which means the confidence interval around a brand coefficient is usually wide — often wide enough that the honest statement is "somewhere between meaningfully positive and very positive," which is a real answer and an uncomfortable one to take into a budget meeting.

Experiments on brand are hard and expensive. You can run a geographic test on a brand campaign, and it is worth doing, but it requires long windows and enough spend to move a measurable outcome against a noisy baseline. This is not a two-week read. Anyone suggesting otherwise is describing something other than a brand effect.

Some of it we cannot resolve at all. Distinctive brand assets — the logo, the colour, the sonic signature, the accumulated recognisability that makes an advert land in three frames — compound over a decade. No responsible test window captures that arc. Neither does creative quality, in most businesses at most spend levels. Neither does category-creation spending, where the honest comparison world is one in which the category does not exist.

For that residue, my position has not changed: fund it on theory. Byron Sharp's How Brands Grow and the Ehrenberg-Bass body of work behind it are a stronger evidence base than most individual company measurements will ever produce, and that theory does not become weak because our instruments cannot resolve it inside a quarter. A measurement system that pressures a CMO into cutting distinctive-asset investment because the interval is wide is not delivering rigour. It is laundering ignorance as discipline.

The correct behaviour when a system cannot bound something is to say so loudly, rather than return a confident zero. That is precisely the failure attribution had, and the failure this whole category has to design against permanently.

This is the practical question, and it is genuinely hard. A CMO walks out of a modelling review holding a result that says increase brand investment, with an interval wide enough that a sceptical CFO could drive a truck through it. What now?

Four things work, in my experience.

Lead with the mechanism, not the number. A CFO who understands why attribution structurally undercounts delayed effects will accept a wide interval on the corrected number. A CFO who is handed only the corrected number will interrogate the number. The explanation is more persuasive than the estimate, which is not what most marketers expect and is consistently true.

Bound the downside instead of promising the upside. Do not walk in saying brand investment will return some multiple. Walk in saying: here is the range, here is the worst case in that range, here is what that worst case costs us, and here is why I am willing to accept it. Finance people are considerably more comfortable with a bounded downside than an asserted upside, because bounded downside is how they think about every other investment in the company.

Stage it with checkpoints. Do not request a large permanent reallocation on a wide interval. Request a defined increase for two quarters with a stated measurement plan and a pre-committed review. This converts an argument about belief into an agreement about a test, and arguments about beliefs do not resolve while tests do.

Write down the kill criteria first. State in advance what result would cause you to reverse the increase. A CMO who has pre-committed to the conditions under which she would back down is enormously more credible than one who has not, and — this is the part people underestimate — she is also much more likely to get the increase approved, because the CFO now knows the request has a defined end rather than being the first step of a permanent drift.

One organisational observation that surprised me.

The CFO is usually not the obstacle.

Finance thinks natively in terms of investments that build assets versus expenses that produce immediate returns, and brand-building maps onto the first category cleanly. A CFO can hear "we have been under-investing in an asset that compounds and over-investing in a channel that harvests" and find it entirely familiar, because it is the same argument they have made about maintenance capital expenditure or research spending in every other part of the business.

The harder conversation is frequently inside the marketing team.

Performance marketing is a discipline whose entire professional culture is built on measurable weekly wins. The people in those roles are good at their jobs, they have been rewarded for a decade on numbers that this migration reveals to be inflated, and the reallocation moves budget out of their remit toward work whose results they cannot see on a Monday dashboard. That is not resistance to truth. That is a rational response to a change that reduces their apparent contribution using a method they did not choose.

I wrote in Issue #6 about how roles change in a migrated function, and this is the sharpest instance of it. The CMOs who handle this well move quickly to make the performance team owners of the reallocation rather than its victims — putting them in charge of designing the tests that validate the brand investment, which is genuinely interesting work and converts the most sceptical group in the building into the group with the strongest incentive to get the answer right.

Rajeev has written in The Incrementalist about the methodology of measuring diffuse effects, and it is the piece I would put in front of an analyst who wants to interrogate rather than accept any of the above.

One. Whether the sixty-forty split gets revisited with causal evidence. Binet and Field's brand-versus-activation ratio has been the industry's rule of thumb for over a decade, derived from effectiveness data rather than from company-level causal models. As more companies run properly calibrated measurement, we should start to see whether the ratio holds, and whether it varies systematically by category, growth stage, or market position. My expectation is that the direction survives and the single number does not. I would like to be part of producing that evidence rather than just citing the original.

Two. Whether any company publishes a brand-spend increase justified by causal measurement. The disclosure I most want to see is a CMO stating publicly that they raised brand investment on the basis of a modelled result, showing the interval, and reporting the outcome eighteen months later — including if it did not work. It would be the single most useful document in this field. Nobody has published one that I am aware of.

Three. Whether agents make this worse before they make it better. An agent optimising against a measurement layer that undercounts brand will cut brand relentlessly and report improving efficiency the entire time. The guardrail question — what the agent is not permitted to touch — is most acute exactly here, because brand is the category where the gap between measured value and real value is widest. Any company deploying marketing agents this year should decide that boundary before deployment rather than after.

The most important thing I have to say about this issue is that the finding is commercially convenient for me, and the reader should weight that heavily.

I sell measurement. An argument that concludes proper measurement tells you to spend more on marketing is an argument that makes my product easier to buy and easier to love. If the consistent finding had been that rigorous measurement justifies large cuts to marketing budgets, I would have a harder business and this would be a less pleasant letter to write. I do not think that has biased the finding — the mechanism is well understood and does not depend on my preferences — but I am not a neutral party and it would be dishonest to present myself as one.

Two further limits.

The direction I have described is drawn from the kind of company that becomes our customer: mid-market and enterprise brands with material budgets across several channels. It should not be assumed to hold universally. A young company with no brand equity to compound, a business in a pure-intent category, or a marketplace whose demand is genuinely captured rather than created may find something quite different, and should.

And this finding does not mean every brand campaign is good. It means the category of brand investment has been systematically undervalued by the previous instrument. A specific bad campaign remains a bad campaign, and better measurement will tell you that too — which is, in the end, the actual point of all of this.

If you have taken a brand-spend increase to a CFO on the strength of a modelled result, I would like to hear how the conversation went — particularly if it went badly. The failures would be more useful to publish than the successes and I have collected far fewer of them.

Thanks for reading the ninth one.

Tobin Co-Founder & CEO, Lifesight August 2026

Coming in Issue #10 (in two weeks) Creative Is the Last Unmeasured Variable

Every serious study of marketing effectiveness lands on the same conclusion — creative quality is among the largest drivers of return, and possibly the largest. It is also the variable our entire industry measures worst. What can actually be established about creative, what cannot, what the emerging attempts get wrong, and why I think this is the most important open problem in the field.

Forward this issue to whoever has been losing the brand-budget argument in your company.