AI & Intelligence
Your restaurant has an intelligence layer.
It reads your own history — orders, tickets, recipes, invoices, customers — and turns it into forecasts, anomalies and recommended actions.
01
Data
every event, captured once
02
Analytics
what happened, reconciled
03
Machine learning
patterns in your history
04
Prediction
what is coming, with a band
05
Recommendation
what to do about it
06
Automation
done inside the system
AI Assistant
Ask in plain language. Get numbers, drivers and a next step.
The copilot queries your live operational data. Every answer shows what moved, by how much, and what to do about it.
GullyBite CopilotBombay Chowk · 3 outletsHow a forecast is made
History in. Prep list out.
Not a black box: the signals are yours, and the output lands in the work your team already does.
Signals
2 years of order history
Daypart and weekday shape
Menu and price changes
Local events and fixtures
Weather and holidays
Recipe and stock movements
ML model
Trained per outlet, retrained weekly
Gradient-boosted demand models with seasonality, event and daypart features — evaluated against your last four weeks before anything is shown.
Lands in
Kitchen prep list
Purchase orders
Shift and staffing plan
Reorder points
Campaign timing
Forecast accuracy · last 8 weeks
ActualPredictedMean absolute error 5.8%
Capabilities
Fourteen ways it earns its place.
Assistant
AI Assistant
Plain-language answers over your live data, with drivers attached.
"Why was Outlet 2 slow on Tuesday?"
Assistant
AI Agents
Scoped agents that watch one domain and raise what needs a decision.
Inventory agent flags a reorder before the shift starts.
Assistant
Automation
Approved recommendations become prep lists, POs and campaigns.
Forecast +18% → prep list and PO drafted.
Learning
Machine Learning
Models trained on your outlets, menu and calendar.
Retrained weekly, validated on your last four weeks.
Learning
Predictive Analytics
Every forecast carries a confidence band, not a single number.
Saturday covers 198–232, most likely 214.
Forecasting
Demand Forecasting
Covers and orders per outlet, day and daypart.
Friday dinner +18% at Indiranagar.
Forecasting
Sales Forecasting
Revenue projection by channel and outlet.
Next 30 days within ±6% historically.
Forecasting
Inventory Forecasting
Ingredient-level requirements from forecast demand and recipes.
22kg chicken, 14kg rice for Friday.
Forecasting
Waste Prediction
Over-prep risk per item before the prep is done.
Veg prep over-prepped 2.1% this week.
Forecasting
Staff Demand
Shift-level staffing suggestions from forecast covers.
One extra rider, 7–10pm, Outlet 2.
Business
Customer Intelligence
Segments, churn risk and lifetime value from visit behaviour.
12 high-value regulars going quiet.
Business
Menu Intelligence
Popularity against true plate cost — menu decisions as margin decisions.
Two variants 9pt below category margin.
Business
Margin Intelligence
Where cost leaks: ingredients, wastage, discounts, channels.
Aggregator discounts cost 1.4pt.
Business
Anomaly Detection
Learned thresholds per outlet and daypart, not fixed rules.
Void rate 3x normal on one till.
Anomaly feed · today
Void rate 3x normal on till 2
Eleven voids after billing between 21:10 and 22:40 — all by one operator.
Veg prep wastage above learned range
Paneer prep exceeded forecast consumption by 2.1kg for the third day.
Cash variance ₹1,240 at shift close
Outside this outlet’s normal ±₹300 band for a Tuesday.
Prep time improvement held for 7 days
Hot station average down 1m 40s since the routing change.
Anomaly detection
It knows your normal.
Thresholds are learned per outlet and per daypart, so a busy Friday isn't an alert and a quiet Friday is. Voids, discounts, wastage and cash variance are watched the same way.
Ask it a question about your own restaurant.
Bring one week of your data to the call and we'll run the copilot against it live.