accessing event data from a storage module which contain historical dated records of events including collection actions taken against each debtor and transactions from each debtor including payment;

estimating using a computer a value function specifying an expected value of each of one or more collection actions for each of one or more states of a debtor, said step of estimating using a constrained reinforcement learning process to approximately estimate said value function with respect to a constrained Markov Decision Process formulation, to maximize long term expected return throughout the course of a collection process, and to optimize collection resources within one or more organizations for maximum expected return subject to one or more given resource constraints for said organizations;

wherein said constrained reinforcement learning process uses a segmented linear regression method to approximately estimate said value function and allows resource optimization within a constrained reinforcement learning module to be performed with respect to one or more segments output by said segmented linear regression method, wherein an objective function is approximated using one or more regression coefficients on one or more segment action pairs and one or more estimated sizes of said segmented action pairs; and

wherein one or more existing business or legal rules, expressed as action constraints, are provided as input and used as additional constraints in a resource, optimization process within said constrained reinforcement learning process; and using a computer to provide an optimized debt collection policy as output in a machine readable formal wherein said optimized debt collection policy is expressed as one or more rules consisting of one or more segments and corresponding action allocations; and providing output on a peripheral device based on said optimized debt collection policy in a human-readable format.