How long does the nightly close-out take?

4 minutes once the habit is set. The POS syncs at midnight; the brief compares actual vs yesterday's forecast and surfaces the top 5 dish-level deltas. The manager taps one reason code per flag (weather, event, staffing, promo, unknown).

Who does the close-out?

Closing manager or owner. Not the morning prep team — they receive the adjusted forecast, they don't tune the model.

What changed in prep accuracy?

Next-week prep error (MAPE on top 15 dishes) dropped from 14.2% to 9.4% over 21 days. Biggest gain: Friday and Saturday dinner batches, where event and walk-in variance was previously unlogged.

Do you need Forkcast for this?

The loop works with any system that compares actual vs prior-day forecast. Forkcast automates the sync and reason-code capture; a spreadsheet version works but breaks when the manager skips a night.

Nightly actual vs forecast close-out: how one habit fixed next-week prep

A forecast without a feedback loop is a guess that never gets smarter. This fictional-but-realistic case study follows a Khar, Mumbai casual-dining outlet that added a 4-minute nightly actual-vs-forecast close-out. Three weeks later, next-week prep error dropped 34%. Here's what they log, what they ignore, and what changed in the kitchen.

The outlet

52 covers, coastal-Maharashtrian and north-Indian mix, Petpooja POS, 38% aggregator share. Forecasting had been 'set and forget' — the owner looked at covers once a week. Prep teams learned from yesterday's waste, not from a structured delta review.

The problem: forecasts that don't learn

Week one of the pilot showed 88% dish-level accuracy — good on paper. But Friday dinners kept missing by 15-22% because a nearby corporate park ran irregular town halls. Saturday lunches under-forecasted when it didn't rain (walk-ins up). The model couldn't learn what nobody logged.

The nightly close-out ritual (4 minutes)

Open the brief — lands at 11:05pm after POS sync. Shows actual vs yesterday's forecast for covers and top 15 dishes.
Flag deltas >8% — any dish or cover count that missed by more than 8% gets a reason code.
Tap one reason — weather, local event, staffing, promo, aggregator outage, unknown. One tap per flag, not an essay.
Done — reason codes feed the next week's forecast adjustment. Manager doesn't edit numbers manually.

What they log vs what they ignore

Log (reason code)	Ignore
Corporate event within 500m	Single-table walk-in variance
Heavy rain / no rain when forecast	One dish off by 6%
Aggregator app outage >30 min	Normal Tuesday dip
BOGO or Zomato deal running	Long-tail SKUs (<2% of revenue)
Short-staffed service (tickets delayed)	Ingredient substitution on one order

Outcomes at day 21

Metric	Week 1	Week 3	Delta
Next-week prep error (MAPE, top 15)	14.2%	9.4%	−34%
Friday dinner forecast error	18.6%	10.1%	−46%
Daily waste from over-prep	₹2,100-2,800	₹1,100-1,500	−47%
Close-out completion rate	62%	94%	+32 pts
Manager time per close-out	—	4 min avg	—

The compounding effect on prep

Prep teams don't read forecasts — they read quantities. When Friday's fish thali batch dropped from 42 to 36 units (adjusted after two weeks of 'corporate event' flags), waste on that SKU fell from ₹680/day to ₹120/day. The kitchen didn't change behaviour; the numbers they received got sharper.

Salary-cycle and weather signals were already in the model. What close-out added was hyperlocal event memory — the corporate park's town halls, a weekly farmers' market two streets away, a school PTA dinner that spikes dessert orders. None of these are in a public calendar; all of them showed up in reason codes within 10 days.

We always knew Fridays were weird. Now the system knows why — and next Friday's fish count is right.
— Closing manager, Khar casual dining (composite pilot)

The close-out fails when it's optional. Tie it to the cash-drop ritual: count the drawer, lock up, tap three reason codes. Four minutes, same time every night.

Start with forecast + feedback

Try the free break-even calculator →