Optimise Your Marketing Budget with Modern MMM

Most marketing budgets are set by inertia. Last year's plan, plus or minus 5%, redistributed by whichever channel made the loudest case in the Q4 review. Modern MMM gives you a defensible alternative - but only if the optimisation step is treated with the same rigour as the model itself.

The optimisation problem in one paragraph

Once an MMM has fitted a response curve to each channel, optimisation is straightforward maths: hold total spend constant, redistribute marginal pounds from channels at the flat top of their curve to channels still on the steep part, stop when marginal returns equalise. The recommendation is the new mix.

That is the easy half.

Where optimisation goes wrong

In practice we see four failure modes:

1. No constraints. An unconstrained optimiser will happily recommend cutting TV by 80% and quadrupling paid search. Real businesses cannot move that fast - sales teams, agency contracts, and creative pipelines all have inertia. Constraints (min/max per channel, week-on-week movement caps) make the answer actionable.
2. Single-KPI tunnel vision. Optimising for short-term sales kills brand investment every time. Modern practice optimises a blended objective: short-term revenue *and* a long-term brand metric (consideration, search volume, branded direct traffic).
3. Ignoring saturation uncertainty. The flat part of a response curve is exactly where the model is least confident. Treating a saturation point as a hard number rather than a range leads to "we cut paid social and revenue did not move" - followed by "actually it did, you just cut too far".
4. No experimentation loop. The optimiser's recommendation is a hypothesis. Test 30% of it, measure the lift, feed the result back into the next model. Anyone running optimisation without experimentation is gambling.

What good looks like

A useful budget optimisation output is not a single number. It is:

• A recommended mix with confidence intervals
• A list of explicit constraints applied
• A flagged set of "high-leverage, high-uncertainty" moves to test before fully committing
• A clear delta vs the current plan, in both spend and expected return

If your optimiser hands you a pie chart and walks away, you have a slide, not a decision.