Worked scenarios
Concrete customer, kitchen, corporate, partner, and crisis examples.
Dina needs a 550-calorie high-protein lunch
Her account profile excludes nuts and favors high protein. TL-AI composes a chicken rice bowl with vegetables and a compatible sauce from live ingredients. Validation checks category structure, portions, calories, allergen flags, availability, and price. Dina replaces one vegetable, applies points partially, schedules delivery, and tracks the driver. The completed order updates repeat history and earns new points.
Value: relevance + conversion + retained identity.One cart, different needs
The family combines a salad, pizza, baguette, and protein dessert. Each item has its own required base/structure; nutrition is summarized per item and total. The signed-in customer’s delivery details prefill, an exact pin is required, and the kitchen receives one deduplicated order with clear item detail. Push messages show preparation and delivery status.
Value: broader occasion and larger basket.Forty employees across two office days
The employer buys meal credits without seeing employees’ wellness data. Employees select or approve meals individually. Tossed Leaf aggregates production by day, ingredient, and delivery cluster while preserving account privacy. Substitutions occur against live inventory; consolidated delivery improves density.
Value: recurring volume + delivery efficiency.Touchscreen and website ordering
The gym receives a secure kiosk plus an approved-domain Web Component. Orders use the shared menu, live availability, nutrition, and payment methods. Partner attribution is automatic; only completed/eligible sales mature into commission. The gym sees its history and payout state while Tossed Leaf retains centralized kitchen and safety control.
Value: asset-light distribution + auditable CAC.Capacity prevents overpromising
Operations reaches the configured slot limit. The next customer sees a later local-time slot rather than an impossible promise. A sold-out ingredient disappears across direct ordering and kiosks. Kitchen staff move tickets through safe states; a driver starts sharing only after assignment and readiness.
Value: service recovery before failure.Personalization degrades safely
The model request fails. The customer’s TL-AI credit is restored, cached valid results remain available, and the deterministic nutrition engine supplies a controlled fallback where possible. Normal ordering, Chef’s Picks, checkout, and kitchen operations continue without the model.
Value: AI is additive, not a single point of failure.Payment is not assumed complete
The private payment service detects an underpayment and reports an exception through a signed callback. The order remains in the correct payment-aware state; the kitchen and partner commission do not treat it as paid. Reconciliation is visible to management.
Value: financial integrity.Digital filter plus human process
The profile removes flagged ingredients and the order carries the customer’s note to the kitchen. The interface repeats the cross-contact warning and does not claim allergen-free food. Staff follow the location SOP, document the decision, and contact the customer if the request cannot be served safely.
Value: safer expectations and escalation.