Tax Collectors in Their Office
Source: Wiki Commons
Tax collectors. Not a popular bunch. Still, the job must be done, and it’s best to do it effectively and efficiently. For years, IBM has been providing governmental tax offices with technology that helps collectors spot cheats and delinquents. Our latest bit of technical wizardry, the IBM Tax Collections Optimizer (nickname: TACO), helps them collect the money that’s owed by individuals and businesses.
This project got underway a couple of years ago when the folks at the New York state tax office began asking for an improved system for going after scofflaws–something that would be more effective but also require less effort. Up until now, the Office of Tax Enforcement followed a standard set of rules and a step-by-step process for going after cheats, but it wasn’t satisfied with the results it was getting.
TACO optimizes the actions of agents by taking into account the complex dependencies between the office’s resources, the potential for recovering money, and legal constraints. It uses a variety of taxpayer data, such as filing status, the amount owed, and past payment history. Then it comes up with an overall plan for collecting from the entire population of delinquent taxpayers.
Rather than building something from scratch, scientists at IBM Research cast around for techniques being used for other purposes that could be applied to the tax office situation. They found them far afield, bringing together methods used commonly in game theory and robotics. Using the Markov Decision Process, or MDP, they mapped the collections process and segmented taxpayers based on their characteristics. Then they applied a method called Reinforcement Learning to the map. Based on historical information, they were able to learn the optimal sequence of actions for each type of taxpayer. “We came up with the optimal approach to maximizing the chances of success while minimizing the expenses,” says Chid Apte, director of analytics research at IBM Research.
Because of a crackdown on tax delinquents, New York has 1.2 million active cases pending. One of its goals is to resolve more cases without having to issue warrants and initiate a formal legal processes that can drag on for years. The State expects to recover an additional $100 million over the next three years with the help of TACO.