Why your delivery app never shows the average

transcript

The time your delivery app shows? It's not the average. On purpose. Show the average, and you break the promise about half the time.

Prep, two rides, a handover. All guesses. Say they add up to 31 minutes on average. But one time in five, orders like it take over 38. So predict the whole spread. Promise near the 90th percentile. Maybe 40.

Why not just pad it? Here's the catch. Told 25 minutes, waited 40? Trust drops. Told 50 for a 30-minute delivery? You order somewhere else. And that never shows up as a late order. Only as lost demand.

And one rule for the whole city? Picture an example evening. The city keeps 91 point 2 percent of its promises. Looks great. One rainy zone sits at 84 point 7. The average hid it.

So the kitchen's queue, traffic and rain feed a model of the spread. Then pick the quantile per zone and per hour. Every arrival is a result. Kept too few promises? Nudge that zone up. Kept far more? Nudge it down.

Now it rains. The spread shifts right, and the shown time grows. The promise grows. Never the rider's speed.

Arrival time is a distribution, not a number. Promise a quantile. Not an average.

Full breakdown in the lecture.