Availability-aware substitution sounds straightforward: connect a recipe to inventory and suggest what is already on hand. In a restaurant, however, “on hand” can conceal several different states. An unopened case, a thawing product, a trimmed container and stock promised to another station cannot be treated as interchangeable resources. A system that ignores those differences can produce a plausible answer that the kitchen cannot execute.
Begin with the physical reality of the stockroom. Identify the exact product, pack format, preparation state and storage location. Record when the quantity was checked and who confirmed it. If an inventory feed is delayed, present it as delayed. An AI model should never convert a stale stock figure into a statement that an ingredient is definitely available now.
Separate gross stock from usable portions
Purchasing weight is not the same as edible yield. Trimming, draining, cooking and portioning can all change what remains available for the plate. Use your kitchen’s measured yields where possible, and identify estimates clearly when those measurements do not exist. A general reference can inform a calculation, but it does not know the condition of your delivery or the specification of your cut.
Keep units explicit. Kilograms of whole vegetables, litres of purée and portions of cooked garnish describe different things. A tool may assist with unit conversion, but it must not silently assume a density, yield or portion size. Show the conversion and its source so the chef can check it before making a service commitment.
| Field | Kitchen question | Why it matters |
|---|---|---|
| Product and state | Which exact item, prepared how? | Different formats behave differently. |
| Usable quantity | What remains after normal losses? | Gross stock may overstate portions. |
| Reservations | Who already needs this stock? | A local solution can create another shortage. |
| Last verification | When was it physically checked? | Inventory changes during service. |
| Handling status | Is its history documented? | Unknown handling cannot be solved by a ranking. |
Time is part of the ingredient
A preparation is only an option if the kitchen can complete it within its established controls and available capacity. Soaking, cooking, cooling, thawing and setting all require planning. Do not shorten a controlled process merely because a model describes a candidate as convenient. A substitution that adds substantial work can also displace attention from other critical tasks.
Equipment is another constraint. A purée that depends on a blender may be impractical when that equipment is occupied or requires a particular cleaning procedure. Ask the tool to distinguish preparation duration from hands-on labour and equipment occupancy. All three can affect whether an otherwise sensible candidate fits the evening’s service.
Use supplier data without surrendering judgement
Product identifiers and batch records help connect inventory to the right specifications. GS1 standards can support consistent identification, while supplier documentation provides product-specific detail. Neither a barcode nor a familiar brand name establishes that a formulation is unchanged. Match the actual product and current label, especially when a delivery has been substituted by the supplier.
Nutrition and composition datasets may add context, but keep them distinct from live inventory and product declarations. The UK’s CoFID dataset, for example, describes food composition; it is not a record of your stockroom or a guarantee about a packaged product’s allergens. Good systems label each source according to what it can actually establish.
Rank feasibility after excluding unsuitable stock
Remove candidates that fail a documented hard requirement before comparing cost, waste or convenience. Then make the trade-offs visible: usable portions, preparation burden, likely sensory change and uncertainty. Cost should include additional labour and probable yield loss, not only the invoice price. Waste reduction is worthwhile, but it cannot justify an unsuitable ingredient or an undocumented handling history.
The most helpful answer may be: this dish is not available tonight.
Close the loop after a trial or service decision. Update reservations, record actual yield and note why a candidate was accepted or rejected. This creates useful operational evidence for future decisions without pretending that one successful trial proves universal interchangeability. The next delivery, recipe or service condition may still change the answer.
Sources & further reading
- GS1 standards: product identification and data exchange.
- USDA Food Buying Guide: reference information on purchasing quantities and yields.
- UK Composition of Foods Integrated Dataset, CoFID: composition data with a different purpose from inventory.
- WRAP — food and drink: resources on measurement and food-waste reduction.
- Food Standards Agency — chilling food correctly: operational food-safety context; local requirements still apply.