Demand forecast and preparation plan are not the same number.
The forecast estimates what guests are likely to buy. The preparation plan decides how much the kitchen should make available. It may include a buffer for uncertainty, but that buffer should vary by item rather than applying one percentage to the whole menu.
A made-to-order item may require ingredient staging rather than finished portions. A long-hold sauce may be prepared in batches. A short-life fried item may need an opening quantity plus replenishment triggers during service.
Choose buffers from service risk and holding life.
Under-preparation creates stockouts, delayed service and substitutions. Over-preparation creates leftover risk. The relative cost differs by item. Use historical forecast error as evidence, then let operations define acceptable service risk.
Higher buffer
Critical, fast-moving items with long replenishment time and acceptable holding life.
Lower buffer
Short-life items, expensive ingredients or items that can be prepared quickly during service.
Respect batch size and service timing.
A mathematically optimal quantity may be impossible to produce. Round to practical batch sizes, observe equipment capacity and split the plan across opening and replenishment batches. The plan should state when a batch is needed, not only the total for the day.
Known bookings and events should be represented explicitly. A manager may reasonably override the statistical baseline when a confirmed group booking is absent from POS history.
Worked example: tomorrow’s biryani demand.
For the next Saturday in the synthetic sample, the same-weekday baseline suggests roughly 40 Chicken Dum Biryani portions at Jubilee Hills, 38 at Gachibowli and 37 at Madhapur before buffers.
If Jubilee Hills uses a 10% buffer and batches in multiples of five, the plan may round to 45 portions. That is a business-rule outcome, not the raw forecast.
Overrides are data, not failure.
Capture the original forecast, recommended plan, final operator quantity and reason. Reasons might include a booking, local event, ingredient constraint, equipment issue or manager judgement. Avoid free-text alone; combine structured reason codes with optional notes.
Repeated overrides can reveal missing features or an unsuitable policy. An operator who always reduces one item may be compensating for known spoilage, a menu change or an inflated recipe mapping.
Measure the complete preparation loop.
At minimum, record forecast demand, recommended prep, actual prep and sales. To establish waste outcomes, add remaining usable preparation and waste by reason. Do not assume that recommended minus sold equals waste.
| Measure | What it reveals |
|---|---|
| Forecast vs sales | Demand-model error, subject to stockouts. |
| Recommendation vs actual prep | Operator adoption and constraints. |
| Actual prep vs sales | Potential leftover, not automatically waste. |
| Recorded waste by reason | Where intervention may be possible. |