Real-world application: communicating marketing value to the board and CFO

In 2019 Adidas' global media director, Simon Peel, stood up at an industry conference and said out loud what most marketing organisations bury: the company had been running roughly 77% of its media money into performance and 23% into brand, and the split was wrong. The number he named as closer to correct was 60% brand, 40% performance. Nothing in the market had changed. Adidas had finally measured itself properly and did not like the answer, and then had to re-argue the investment case twice: once inside the building with finance, once in public. That single situation is what this lesson follows.

How a defensible number became the wrong number

Adidas had not been careless. The performance budget was the easiest thing in the company to justify. Last-click attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → produced a clean revenue figure per euro within days, tied to e-commerce transactions, sitting in systems finance already reconciled. Brand spend produced nothing of that shape. Given the decision criteria the foundations lesson sets out, the drift was close to mechanical: money moves to the channel that can show its work, quarter after quarter, until the split nobody chose becomes the split you have.

Two flaws compounded. The attribution model handed digital the credit for demand that brand activity had created. And Adidas was optimising against e-commerce revenue at a point when e-commerce was around €1.6bn of €21.9bn in group revenue for 2018, roughly 7%. The overwhelming majority of sales went through wholesale partners and own retail, where no click ever records the purchase. The company was tuning the measurable seven percent and assuming the other ninety-three would follow it.

Re-arguing it internally

Econometric modelling showed brand-building work, television included, driving sales that the last-click model had been giving to performance. That finding makes the finance conversation harder, not easier. You are asking a CFO to move money off a line with a fast, countable return and onto a line whose return is estimated over quarters, with a confidence interval instead of a receipt.

The argument that survives that room is not "brand matters". It is: the number you have been approving measures the wrong thing, here is the size of the error, and here is what the corrected number does to the returns you thought you were buying. Framed that way it is a reclassification of past spend, not a new ask, and it uses the payback and mix models the frameworks lesson builds rather than re-arguing them from scratch.

There is an edge case worth planning for. The first time you put a mix model in front of a board, you are asking them to accept a new measuring instrument at the same moment you ask them to accept an unwelcome result. The challenge will land on the model, not on the money. Directors will ask who built it, what it assumes about baseline sales, and why it disagrees with the dashboard they were shown last quarter. The way through is sequencing: agree the model, the data window and the assumptions with the CFO before the outputs exist. Once the method is signed off in the abstract, the answer is not negotiable on the day it becomes inconvenient.

The cost of saying it in public

Peel's remarks were picked up across the trade press within a day and are still cited years later. Going publicGoing publicThe first sale of a private company's shares to public investors on a stock exchange, converting private ownership into publicly traded stock.View full definition → with your own misallocation is a deliberate trade. It sets a benchmark you are now committed to moving against, it gives every agency in the market a pitch line, and it hands equity analysts a question for the next results call. Adidas took that trade because a correction of that size leaks anyway, and being the second party to describe your own mistake is far worse than being the first.

The second-order consequence is the one CMOs underestimate. Once your function has said publicly that its own numbers were wrong, every subsequent number carries an extra burden of proof. The next budget discussion opens with "how do you know this time?" Answer it in the same breath as the admission: name the model, state how often you will re-run it, and state the threshold that would make you reverse the decision again. A reallocation with a published reversal condition reads as governance. One without reads as a new opinion replacing an old one.

The counter-example: when the answer runs the other way

Uber, 2017. Kevin Frisch, then running performance marketing, switched off around $100m of annual digital advertising, much of it app-install spend, and installs barely moved. Uber went on to sue its mobile agency, Fetch Media, over spend it argued was fraudulent or unattributable.

Structurally this is the same defect as Adidas: an attribution model crediting the wrong cause. The conclusion is the opposite. Adidas's model overstated performance because brand had created the demand that digital then harvested. Uber's overstated performance because organic demand for a category-defining app already existed and would have converted regardless. One correction moved spend. The other deleted it.

A board that has read either story will ask which one you are, and "we think it's the first" is not an answer. The test is the same in both cases and cheap enough that finance will approve it: turn the spend off somewhere. Matched geographies, a dark market, a holdout period long enough to clear the purchase cycle. An experiment that costs a fraction of a quarter's media and produces a result nobody can argue with is worth more in a board meeting than any model output.

What the rebalance actually bought, and who collected

Adidas raised brand investment after 2019, and by the 2021 "Own the Game" plan marketing investment sat alongside direct-to-consumer as a named growth lever rather than a cost line to be defended. There is a timing problem buried in that. Brand money re-argued in year one shows up in years two and three, beyond the horizon most boards work to and often beyond the tenure of the executive who made the argument. Adidas changed chief executives at the start of 2023, in the middle of the period the rebalance was meant to pay out in. Whoever wins this argument frequently does not get to bank it.

The practical response is to commit to interim markers a board can check at six and twelve months even though the revenue case runs twenty-four to thirty-six: unaided awareness in priority markets, share of branded search, the base sales component in the mix model, wholesale sell-through in territories where brand spend went up first. Without those checkpoints the reallocation gets quietly reversed during the first bad quarter, because performance marketing is the only line that can be cut on a Friday and show a margin effect by Monday.

CMO action items

• Work out your own split. Take the last full year of media investment and divide it into demand capture and demand creationdemand creationCreating and stimulating demand for your offer, often upstream of the buying process to generate interest and awareness before prospects are ready to buy.View full definition →, then say out loud what the ratio is and whether anyone ever chose it. Most CMOs discover, as Adidas did, that the number is an accumulation of small defensible decisions rather than a decision.
• Check what proportion of company revenue your primary optimisation metric actually covers. If you are tuning against e-commerce and e-commerce is a tenth of the business, write that fraction on the slide before the CFO finds it.
• Design one switch-off test this quarter, agree the read-out method with finance before it runs, and put the result in front of the board whichever way it falls.

Common mistakes that kill results

• Presenting the corrected split as a request for more money. Adidas's argument worked because it was about where existing money sits and what past returns were really worth. Bundle a rebalance with a budget increase and the board will refuse both.
• Letting the model be new information on the day the result is. Get the methodology approved while the answer is still unknown.
• Admitting the error without a monitoring commitment attached. Boards forgive a wrong allocation caught by better measurement. They do not forgive a function that keeps revising its own view with no stated way of knowing when it is wrong again.
• Treating a switch-off test as an admission that the spend might be worthless. Refusing to run one tells the CFO you already suspect the answer.