What is scenario planning?

A scenario is a question put to a trained model: what happens to sales if we spend this much, split this way, over these weeks? The model answers with a forecast and its range, channel by channel. Scenario planning is running several such questions on the same model and comparing them: last year's plan repeated, the plan at the budget the return target supports, a plan with a fifth less, a plan that tests a new channel.

Every scenario runs on the saturation curves, carryover and seasonality the model has estimated, so the forecast for a changed budget is grounded in how each channel has responded before, with the range reflecting how sure the model is. A scenario can hold the total fixed and let the optimizer find the split, or set a return threshold and let it find the total; constraints such as a locked TV level are respected and their cost shown.

Two questions come up every budget cycle, and both are scenarios. A cut: run it and read what it costs in forecast sales, and which channels lose most. A target that has to be hit with confidence: find the budget that puts the target at the cautious end of the forecast range, and compare it with the budget that puts it in the middle. The gap is the price of certainty, and it is better discussed in September than in December.

Scenarios and forecasts are self-serve. In Odins they run any time, in the platform or from your own AI assistant through the MCP, and return the same numbers the platform computes. Only the model retrain and the reviewed recommendation are monthly. The outputs are shaped for each reader: total budget against forecast outcome and marginal return for the CFO, channel allocation and constraint indicators for the CMO, period-level plans the media team can buy.

Every scenario is stored, so when the quarter closes the forecast can be checked against what happened. That is what makes scenario planning a track record rather than a slide.

• Budget optimization: Finding the split and the total where the next unit works hardest, under the constraints the business has.
• Confidence range: The range a Bayesian model reports around every number. A recommendation without one is hiding something.
• Forecast verification: Forecast first, compare with actuals afterwards. The only honest way to know whether a model is right.
• Marginal ROAS: What the next unit of spend in a channel earns at today's level. The number a budget change actually asks for.
• Seasonality: The yearly pattern in demand that the model separates from marketing, so a December spike is not credited to December's ads.

Go one level deeper: Scenario planning and budget optimization. Board-ready scenarios and forecasts. Or go back to all terms in the glossary.