Logistics ops · Production

DoorDash supercharges delivery logistics through causal ML and joint optimization of supply and assignment

The problem

DoorDash's supply and assignment systems operated independently, causing inefficiencies: static supply targets led the system to falsely conclude there were no more improvements once targets were hit, and peak pay causal effect estimates were inherently biased because they were not derived through a causal ML framework.

First attempt

Fitting a standard ML model on observational data produced biased causal estimates because peak pay correlates with undersupply conditions, and the double ML approach was also sub-optimal due to unmet assumptions about confounding variables.

Workflow diagram · grounded in source
1
Demand and supply forecasting
ai_action
“We have regular ML models that forecast demand and baseline supply hours for all geo and time units over a few days”
2
Peak pay causal impact estimation
ai_action
“we build a causal model to estimate the heterogeneous impacts of peak pay on incremental supply hours”
3
Delivery duration prediction
ai_action
“we have a regular ML model that predicts delivery duration as a function of utilization and other features”
4
Batch rate causal modeling
ai_action
“we built two causal models that estimate the heterogeneous impacts of batch rate on incremental duration and Dasher cost, conditional on different utilization levels”
5
Joint optimization solve
ai_action
“The joint optimization system formulates the problem as a constrained integer programming model with two-dimensional decision variables: Peak pay amount and batch rate level”
6
Peak pay and batch rate published
output
“we can proactively determine and publish peak pay, mobilizing Dashers more effectively to the correct geo and time units”
7
Multi-stage reactive adjustment
feedback_loop
“we have a multi-stage joint optimization with both proactive and reactive systems”
Reported outcome

Multiple A/B tests validated that the two-stage joint optimization system significantly improves delivery quality and reduces total cost.

Reported metrics
Delivery qualitysignificantly improves delivery quality
Total costreduces total cost
Reported stack
LightGBMR-learner
Source
https://careersatdoordash.com/blog/supercharging-doordash-logistics-through-causal-ml-and-joint-optimization/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Multiple A/B tests validated that the two-stage joint optimization system significantly improves delivery quality and reduces total cost.

What tools did this team use?

LightGBM, R-learner.

What results were reported?

Delivery quality: significantly improves delivery quality; Total cost: reduces total cost (source-reported, not independently verified).

What failed first in this deployment?

Fitting a standard ML model on observational data produced biased causal estimates because peak pay correlates with undersupply conditions, and the double ML approach was also sub-optimal due to unmet assumptions abou…

How is this logistics ops AI workflow structured?

Demand and supply forecasting → Peak pay causal impact estimation → Delivery duration prediction → Batch rate causal modeling → Joint optimization solve → Peak pay and batch rate published → Multi-stage reactive adjustment.

Related logistics ops cases