Same budget, more results.

Most budget arguments between marketing and finance are about evidence, not marketing. The CMO brings platform ROAS, which finance discounts because every platform claims credit for the same sale. Finance brings last year's budget and a growth target. Neither side can say what a change in spend would do to revenue, so the number that wins is usually the one that was there before.

A marketing mix model changes the evidence. It starts from the sales finance already reports, separates what would have sold anyway from what each channel added, and gives every channel a response curve with a confidence range. Three numbers come out of it that finance recognizes:

• Contribution, the revenue each channel added over the period beyond the baseline.
• Marginal return, what the next pound, euro or krone in each channel brings back at today's spend.
• A forecast for any budget, with a range, stored and checked against actual sales when the month closes.

The same model answers the questions that come from the top. When leadership wants to cut the brand budget, run the cut as a scenario and read what it costs in forecast sales before anyone decides. For brand channels, count the sales that land later too; brand effects explains how the model credits them. When marketing wants more, show the marginal return at the current budget against the return the business requires. If the next unit still clears the bar, the case for more is arithmetic. An owner reviewing a portfolio company can run the same check: is the marginal return at today's spend above or below what the business needs?

Finance will also ask whether the model can be trusted. Show the verification chart: forecasts stored before the actuals arrived, next to what happened. In steady state ours land within 10 to 15 percent of actual sales. The data it takes is modest. Two years of weekly spend and sales is the comfortable floor, three is better, and six months can work when well-set priors carry more of the load. Our guide to MMM data covers what to collect.

Every channel follows the same pattern. The first unit of spend reaches the easiest customers. Later units reach harder ones, so each extra unit earns a little less, until the curve flattens. Rising cost per acquisition in paid search or paid social is often, at least in part, this curve showing up in the platform numbers: you are buying further along it.

The curve separates two returns that get mixed up constantly. Average return is what a channel earned per unit over the whole period, and it is what the platforms report. Marginal return is what the next unit earns at today's spend. A channel with a strong average and a weak margin is the one to stop increasing, even when it looks like your best performer. Budget decisions belong on the margin.

Reallocation follows from that. Move spend from channels where the next unit earns little to channels where it still earns a lot, until the marginal returns are roughly equal. The total does not change. The result does: typically 5 to 15 percent more effect from the same budget. Because digital and offline channels sit in one model, the comparison covers TV, radio and out-of-home as well as search and social, on the same terms.

The same curves answer the question finance asks first. Set the return the business requires on its last unit of marketing, as a marginal ROAS or a maximum marginal cost per customer, and the model finds the total at which every extra unit still clears it. Above that total, money is spent below the bar. Below it, money that would have cleared the bar is left unspent.

That turns a political conversation into a technical one. The question stops being "marketing wants more" and becomes "at what return do we stop?" A CFO can answer that, and defend the answer to a board. Set the right marketing budget goes further into how the total is set.

A plan for next year is a forecast, so it should come with a range. Build three or four scenarios on the same model and compare them over the same weeks: last year's plan repeated, the plan at the budget your return target supports, a plan with a fifth less, a plan that tests a new channel. Each shows forecast revenue with its range, the split by channel and the marginal return at the end point. Leadership chooses between futures instead of approving a single number.

Ranges also make revenue targets plannable. If the fourth-quarter target has to be hit with confidence, plan for the budget that puts the target at the cautious end of the forecast range, not the middle. That budget is usually higher, and the gap between the two is the price of certainty. Better to have that conversation in September than in December.

Seasonality belongs in the plan, not in a flat monthly split. Demand swings through the year, and in peaks like Black Friday and Christmas marketing often converts more easily, so the same spend buys more. The model separates the season from the marketing effect and lets each channel's curve move with the season, so it can say which weeks to lean into and which to hold back. Time-varying saturation shows how.

Two more planning questions come up every year. Brand against performance: count the sales that land months later as well as those that land in the period, and compare the plan on both horizons. A new country: one model per market, with the first market's channel learnings carried over as priors, scaled to the new market's size and corrected by its own data as it arrives. Where a channel or a market has no data yet, say so plainly and treat the first months as the test.

An allocation is only useful if the media team can buy it. The optimizer spreads the chosen total across channels and weeks, inside the constraints you set: a TV contract already signed, a floor on brand, a channel held steady for a test. Each constraint gets a price, so "we must spend this on TV" becomes "and that costs this much forecast revenue". Many teams lock TV at the planned level and let the model optimize the rest. That is a decision with a known cost, and a perfectly reasonable one.

The result is a plan per channel per week, ready for wherever the buying happens. Then the cycle runs. Data flows in daily, the model retrains and is checked against actual sales every month, and our team reviews every recommendation before it reaches you. Where the model is unsure about a channel, it does not move money on a weak signal. It proposes a structured test with a budget and a success criterion, and the result goes into the next model. For the method behind all of this, read our complete guide to marketing mix modeling.