Marketing ops · Production

DoorDash applies causal machine learning to optimize promotion targeting and personalize discount depth

The problem

DoorDash's blanket promotional campaigns eroded margins by discounting orders that would have happened regardless — the non-incremental promotion problem — with no mechanism to identify which customers genuinely needed an incentive to order.

First attempt

Uniform blanket campaigns looked successful on surface metrics — order volume rose — but masked systematic margin waste by discounting customers who would have ordered regardless of any promotion.

Workflow diagram · grounded in source
1
Campaign constraints defined
trigger
“Each campaign comes with its own constraints, such as total budget, per-offer limits, or category restrictions.”
2
Causal lift estimation via DML
ai_action
“Using a Double Machine Learning (DML) model, we measured the incremental lift for each user under different promotion types. The model helped us identify who was most likely to respond — the users for whom the promotion would truly make …”
3
Optimization assigns best offer
ai_action
“the optimizer assigns each customer the promotion that yields the highest expected return, all while staying within the campaign's overall budget”
4
Personalized promotion delivered
output
“customers received just the right amount of incentive needed to act”
Reported outcome

The causal ML framework achieved the same incremental orders at roughly half the cost per incremental order in promotion targeting, and delivered a higher order rate lift with improved cost efficiency in personalized discounting.

Reported metrics
Cost per incremental orderroughly half the cost per incremental order
Order rate lifthigher order rate lift
Campaign cost efficiencyimproving cost efficiency
Reported stack
Double Machine Learning (DML)
Source
https://careersatdoordash.com/blog/doordash-smarter-promotions-with-causal-machine-learning/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The causal ML framework achieved the same incremental orders at roughly half the cost per incremental order in promotion targeting, and delivered a higher order rate lift with improved cost efficiency in personalized…

What tools did this team use?

Double Machine Learning (DML).

What results were reported?

Cost per incremental order: roughly half the cost per incremental order; Order rate lift: higher order rate lift; Campaign cost efficiency: improving cost efficiency (source-reported, not independently verified).

What failed first in this deployment?

Uniform blanket campaigns looked successful on surface metrics — order volume rose — but masked systematic margin waste by discounting customers who would have ordered regardless of any promotion.

How is this marketing ops AI workflow structured?

Campaign constraints defined → Causal lift estimation via DML → Optimization assigns best offer → Personalized promotion delivered.

Related marketing ops cases