Paper · journal · under review
Offline reinforcement learning for a utility billing assistant
Kitchen table
A billing helper has to pick what to say to a customer. I tested careful planners and a cautious learner against one simple rule that adapts. Every bill in the test is made up.
Whiteboard
I compare support-constrained planners and a conservative Q-learner with an adaptive rule on synthetic bill assistance. The title states the result. The data is synthetic. The paper is under review.
Boardroom
A billing assistant must choose actions that are safe to take. This paper asks whether sophisticated learners beat a simple adaptive rule. The test uses synthetic data. The paper is under review.
Keywords