Reviewing assumptions before accepting a budget recommendation

An MMM budget recommendation is an allocation under stated model and scenario assumptions. Before approving it, establish which outcome it optimises, what spend changes it permits, what future costs and activity it assumes, and how uncertain the estimated benefit is. The proposed allocation must also be feasible to buy.

Reconstruct the proposed change

Request current and proposed channel amounts in AUD for the same period and cost boundary. Confirm whether the total budget is fixed. Record each channel's proposed movement, permitted range, assumed media price, market and flight. Check that the outcome is the business measure the decision requires.

Ask whether the result is an estimated incremental outcome under stated assumptions or a different measure. Meridian's future budget optimisation estimates the incremental portion of the outcome under its assumptions; it does not predict the future outcome value itself. If the scenario values each outcome in revenue, check that value against the agreed business definition.

Inspect the model behind the allocation

Review input quality, convergence where applicable, baseline plausibility and the sensitivity of channel estimates to priors. A strong fit to historical outcomes can reveal some problems when absent, but does not settle whether estimated effects are causal.

Inspect response curves at current and proposed spend. Many MMMs allow delayed effects and diminishing marginal returns. A large move may rely on activity levels the model has scarcely observed.

Request a clear account of that extrapolation and uncertainty. Historical average return is not the expected return from the next increment of spend.

Check the buying plan

Re-run the scenario with plausible changes to media price, flight timing, outcome value and channel limits. Record when those changes reverse the preferred allocation. Check booked inventory, minimum commitments, creative capacity and offer availability. A model's channel constraint does not verify any of those commercial facts.

Approve a dated scenario with its data version, outcome, assumptions, uncertainty and decision owner. If it suggests a large untested shift, consider a staged change with an agreed review point. Compare what actually ran with the approved conditions before judging the recommendation.

Pros and Cons of Large Untested Shifts in Media Spend