Next Best Action: Causal Constrained Offline RL for Business Action Targeting
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…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.