Next Best Action: Causal Constrained Offline RL for Business Action Targeting

Paper recorded by Signals 4 on 2026-10-19 in cs.AI. Abstract reproduced from arXiv; link to the original below.

Published 2026-10-19 on arXiv · recorded by Signals 4 on 2026-10-20

Category: cs.AI · 人工智能 · first seen 2026-10-20

Abstract

Next Best Action (NBA) systems must select among customer interventions (e.g., discounts, promotions, and lifecycle actions), maximize long-term economic value, satisfy complex operational constraints, and remain causally grounded. Existing approaches lack a rigorous framework for meeting these requirements jointly. We address this gap by formulating NBA as a Constrained Markov Decision Process (C

Read on arXiv →