Logistics ops · Production

DoorDash uses ML to dynamically optimize Dasher wait times via Auto Order Release

The problem

Long Dasher wait times hurt Dasher earnings and merchant experience because restaurants waited for Dashers to arrive before preparing food; the original AOR heuristic used hardcoded store-level geofences with no per-delivery flexibility.

First attempt

The original Auto Order Release (AOR) heuristic used hardcoded store-level geofences that could not capture actual variance in prep and Dasher arrival times, generated merchant complaints about Dashers waiting, and could not scale without manually defining geofences per store.

Workflow diagram · grounded in source
1
Order confirmed, dispatch kicked off
trigger
“right after the order is confirmed by the merchant, we kickstart the matching and dispatch process”
2
ML predicts prep and arrival times
ai_action
“we developed a set of new predictors to estimate the time needed for the restaurant to prepare the food against the time needed for a Dasher to arrive at the restaurant and pick up the food. We used a LightGBM model that reached both gre…”
3
Dynamic per-delivery release decision
routing
“Update our architecture to call ML models to decide if an order should be placed on a delayed release or an instant release”
4
Order released to kitchen at optimal time
output
“a new feature that leverages ML models to dynamically decide when an order should be released to the kitchen”
Reported outcome

The ML-based feature dynamically decides when each order should be released to the kitchen, reducing Dasher wait times and improving core delivery metrics.

Reported metrics
Dasher wait timereduction in the wait times that the Dashers were experiencing
Core delivery metricsImprovement in our core metrics
Reported stack
LightGBM
Source
https://careersatdoordash.com/blog/lifecycle-of-a-successful-ml-product-reducing-dasher-wait-times/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The ML-based feature dynamically decides when each order should be released to the kitchen, reducing Dasher wait times and improving core delivery metrics.

What tools did this team use?

LightGBM.

What results were reported?

Dasher wait time: reduction in the wait times that the Dashers were experiencing; Core delivery metrics: Improvement in our core metrics (source-reported, not independently verified).

What failed first in this deployment?

The original Auto Order Release (AOR) heuristic used hardcoded store-level geofences that could not capture actual variance in prep and Dasher arrival times, generated merchant complaints about Dashers waiting, and co…

How is this logistics ops AI workflow structured?

Order confirmed, dispatch kicked off → ML predicts prep and arrival times → Dynamic per-delivery release decision → Order released to kitchen at optimal time.

Related logistics ops cases