A delivery-time promise is easy to display and hard to keep. The map pin that shows a rider "3 minutes away" is a UI element; behind it has to sit a routing and allocation system that's actually making good decisions in real time, or the promise becomes a source of complaints rather than trust.
The system behind Zuma Food's delivery estimates weighs live rider location against current order load, restaurant preparation-time history for that specific item (not a flat estimate per restaurant), and route conditions, then re-evaluates the estimate continuously rather than committing to a number at order time and hoping it holds. Riders are allocated based on who can complete the order fastest given their current position and existing assignments, not simply who's geographically nearest.
The operational lesson generalises to any logistics-adjacent product: a delivery-time estimate is only as good as the weakest signal feeding it. Restaurants with inconsistent prep times, or a rider-allocation model that ignores existing workload, will produce technically-computed estimates that are functionally guesses. Getting this right is unglamorous plumbing work, but it's the difference between a delivery app people trust and one they stop using after the third late order.