“What should Jubilee Hills prepare tomorrow?”
The request identifies the decision, outlet and time horizon.
Restaurant Forecasting · Forecast Chat
Ask an operating question. See the evidence behind the answer.
Forecast Chat lets operators query revenue, demand, preparation, inventory and waste in plain language. Answers are built from the same forecast tables and business rules used across the product, not invented from an unconstrained model response.
Grounded in connected dataShows assumptions and scopeRefuses unsupported calculations
Forecast Chat
Answer anatomy
A good answer has more than a number.
Saturday · 1 August 2026
The answer states the selected outlet, date, available history and forecast version.
Item-level quantities
For example, the sample same-weekday baseline suggests about 40 Chicken Dum Biryani portions for Jubilee Hills before preparation buffers.
History and adjustments
The response explains seasonality, events, overrides, uncertainty and the rule that turned demand into a preparation quantity.
Accept, adjust or investigate
The user can move from the answer into an operational decision instead of ending with conversational text.
Grounded examples
Questions the current data can answer.
RevenueWhich outlet has the highest expected revenue next Saturday?
PreparationHow many Chicken 65 portions should Gachibowli prepare tomorrow?
InventoryWhich ingredients are below the next seven days of expected usage?
WasteWhich menu items recorded the highest waste share last week?
Sample answer
Next Saturday’s all-outlet baseline
Using the previous four Saturdays in the attached synthetic dataset, the baseline is approximately ₹1.69 lakh in revenue and 281 completed orders. Jubilee Hills contributes about ₹61,126, Gachibowli ₹59,089 and Madhapur ₹48,407 before event or promotion adjustments.
Missing-data policy
Unsupported questions should produce a data request, not a confident guess.
| Question | Available answer | Missing input |
|---|---|---|
| Which aggregator channel is most profitable? | Not supported by the attached sample. | Channel-level orders, commissions, discounts, refunds and fulfilment cost. |
| Which vendor increased prices? | Not supported by the attached sample. | Supplier, ingredient, invoice price, unit, purchase date and comparable pack size. |
| How much cash will we need next week? | Ingredient requirement can be estimated, but cash need cannot. | Purchase prices, payment terms, payroll, rent, tax and other cash commitments. |
| Did we save ₹20,000 in waste? | Potential excess can be estimated. | Actual prep, actual waste, recipe cost and an agreed baseline. |
Team workflow
Designed for decisions, not open-ended conversation.
Start with the number
Ask for a forecast, comparison or exception within the user’s permitted outlet and date scope.
Open the contributing rows
See which outlets, items, ingredients or days explain the result.
Apply operating context
Record a buffer, override, event or approval rather than silently changing the number.
Turn the decision into work
Export a prep list, purchase review or outlet action, with the originating evidence attached.
FAQ
Questions teams usually ask.
Is Forecast Chat the forecasting model?
No. Forecasting, calculations and business rules should run in governed services. The language model interprets the question, calls permitted capabilities and explains the result.
Can it answer across outlets?
Yes, subject to user permissions and consistent entity mapping. The response should show whether the result is consolidated or outlet-specific.
Does Openfluid train a model on every conversation?
The current prototype does not establish a production training or retention policy. Those controls must be defined in the deployed architecture and customer agreement.
How are numerical errors prevented?
Important totals should be computed outside the language model, returned with source identifiers, and checked for conservation between grouped and ungrouped views.
Bring your real operating questions.
A useful pilot begins with the questions managers repeatedly ask before prep, purchasing and service.