Forecasting, simulation and optimisation: which does your decision need?

A forecast, a simulation and an optimisation model answer different questions. A decision about inventory, capacity or capital may need all three. Start with what the accountable person needs to choose, then select the methods that help them compare the consequences.

Forecasting: what is likely to happen?

A demand forecast estimates future demand from available evidence. Its usefulness depends on the horizon, level of detail and decision it supports. A weekly SKU forecast and an annual capacity forecast serve different choices.

Show a range as well as a central estimate. Explain the main drivers, data freshness and conditions in which the model has been tested. Confidence bands are conditional on the model and its assumptions; they do not capture every possible disruption.

The cost of forecast error is rarely symmetrical. Underestimating demand can lose sales or disrupt production. Overestimating it can tie up cash and create waste. Measure those consequences alongside forecast accuracy.

Simulation: what could happen under different assumptions?

Simulation explores alternatives. What if demand rises while a supplier is late? What if an asset is taken offline now instead of next month? What if a capital project is delayed?

Use explicit assumptions and plausible ranges. Compare a baseline with alternatives and test sensitivity to the variables that could change the decision. Do not present a hypothetical scenario as a prediction.

Follow second-order effects. An extra production shift may improve throughput while consuming maintenance capacity, increasing overtime and building inventory that customers have not ordered. A useful scenario makes those dependencies visible.

Optimisation: which feasible option best meets the objective?

Optimisation compares choices within defined constraints. The objective might be cost, service level or a balance of both. Constraints might include labour, storage, maintenance windows or minimum safety stock.

The result is only as useful as those definitions. A mathematically optimal plan can be operationally unsuitable if an important constraint is missing. Show the trade-offs and let the operator challenge the assumptions before approving a plan.

Use the methods together

For replenishment, forecast demand with a range. Simulate supplier delays and demand shocks. Compare inventory policies against service levels, cash tied up and stockout costs. The planner reviews the options, records any override and approves the replenishment plan.

For capital, compare investment timing under several demand and cost scenarios. Show downside exposure and the assumptions that change the preferred option. The investment owner remains accountable for the choice.

Close the loop

Record the evidence, options, assumptions and rationale at decision time. Compare actual outcomes with expectations. Review whether forecast errors, missing constraints or an operational intervention explain the difference. An outcome alone does not establish causation.

Explore forecasting, simulation and optimisation, or see how demand and inventory forecasting keeps confidence bands and human overrides in the planning process.