Brand building is the part of the budget that is hardest to defend and easiest to cut. The sales it drives arrive over months and years, long after any attribution window has closed, so the channels that build brand look expensive on every short-term report, and the channels that harvest demand look cheap. When leadership asks what a cut to the brand budget would cost, most measurement has no answer, because it never counted the sales in the first place. Marketing mix modelling can count them, if the model is built to. This guide covers how the long-term effect of brand building is measured in an MMM, where the numbers come from, and how to turn the result into the case for or against a cut.
Measure brand's effect on sales directly in the marketing mix model, with a long-term component. Each channel's effect is split in two: sales that arrive within weeks, and a share that goes into one shared stock of brand value, which pays out sales over time and leaks without new investment. The split per channel and the decay rate start as priors from the published research and are moved by your data where it has something to say. The result is the same budget read on three horizons, and a forecast of what a cut would cost in year one and in years two and three. Brand trackers are a cross-check on the level, not the thing the model is built against.
Attribution counts conversions inside a window of days. A plain MMM carries each channel's effect forward for a few weeks through adstock and then stops. Both methods book whatever arrives later as baseline: the sales that would have happened anyway. For a brand campaign that is most of what it did. The published evidence is consistent on this. Thinkbox's Profit Ability 2 study of 141 brands found that 58 percent of advertising's profit arrives after the first 13 weeks. A model that stops at week four has not measured brand; it has decided brand does not exist.
The symptom is familiar: every model run says cut TV and fund search, the team does, and sales drift down a year or two later for reasons nobody can connect to the cut. The full guide to marketing mix modelling covers how a model credits sales in general; this guide is about the long tail.
Odins models brand as one leaky bucket. Each channel pours some of its effect into the bucket; the bucket pays out sales over time; and the level falls without new investment. Three settings make it work.
The figure uses illustrative numbers for two channels with identical spend. The performance channel is credited about the same on every horizon: 216 sales on three weeks, 221 on a year, 228 on three years. The brand channel is credited 155, then 225, then 331, because most of its sales arrive later. Neither number is wrong. They answer different questions, and the budget meeting has to agree which question it is asking.
Two years of weekly data cannot identify a three-year decay on its own, and a vendor who says otherwise is reading noise. The method that holds up combines three sources.
| Source | What it supplies | How it is used |
|---|---|---|
| Published research | The shape and size of the long-term effect: Thinkbox's Profit Ability 2 on when profit arrives, Binet and Field on the balance of brand and activation by category, brand size and life stage, Ehrenberg-Bass on how brands decay when support stops | Priors with bounds for the brand share per channel and the decay rate |
| Your data | How sales moved when brand spend changed level: a campaign that started, paused or stepped up on a known date | Moves the priors inside their bounds, and widens or narrows the range |
| Brand trackers | Monthly awareness and consideration readings | A cross-check on the level of the brand stock, not a target the model is fitted to, because monthly survey readings are too noisy to carry that weight |
The data's contribution depends on one thing: whether brand spend has varied. A brand budget held flat for three years looks exactly like the baseline, and the model will return its priors for it. Where a campaign started, paused or changed level on a known date, that level shift is marked explicitly and the model learns from it. If the history has none, the result is a range from the research, and the range is the finding.
This is the conversation the method exists for. Leadership proposes cutting the brand budget by a third. The question is what that costs in sales, and when.
Argue about the horizon, not about whether brand works
Take both plans to finance with the three-horizon table. The debate then becomes which horizon the business plans on, which is a legitimate strategic question with an answer, instead of whether brand advertising "works", which is an argument nobody wins. The same model that shows the cost of a cut also shows where brand spend has stopped adding: a brand channel can be over-funded too, and the model says so with the same range.
Five limits, stated plainly, because the method is only credible if they are.
Odins builds and runs the model with the long-term component and shows the result on short and long horizons. The scenario planner forecasts any two plans against each other whenever you need it. Companies like CDON, Nettbil, Hyre and Megaflis run their marketing budgets on it.
With a marketing mix model that splits each channel's effect into short-term sales and a share that builds a brand stock, which pays out over years and decays without investment. The shares and the decay are priors from the published research, moved by your data where brand spend has varied. The result is the same budget read on short and long horizons.
An MMM with a long-term brand component and a scenario planner that forecasts the current plan and the cut side by side, on one-year and three-year horizons, with confidence ranges. Google's full-funnel guidance for Meridian and PyMC-Marketing's long-term brand examples describe how to add one in code; the work is in setting the priors, collecting the data and operating the model month after month. Odins does that as a managed service.
Two years cannot identify a three-year decay on their own. With priors from the research and at least one known change in brand spend, the model gives a credible range. Without any variation in brand spend, the answer is the prior, and a good vendor says so.
No. The model measures brand's effect on sales directly. A tracker is a useful cross-check on the level of the brand stock, but monthly survey readings are too noisy to build the model against.
At roughly two percent a month in the model's default prior, with a range of one to four. Half of what a campaign built is paid out after just under three years. The fall is gradual, which is why a cut looks free in the first year and expensive in the third.
By marginal return on the horizon the business plans on. On a short horizon the optimiser will move money to performance; on a three-year horizon it will keep more in brand. The split is a strategic choice about the horizon, and the model shows the cost of each choice in sales.
Read the full method and the numbers in brand effects in the model, see how the scenario planner forecasts two plans against each other, or book a demo and bring the proposed cut.