Development of a probabilistic model to determine the likelihood of success in a bank card dispute cycle.
The probabilistic model was built on a neural network trained on the results of previously conducted dispute cycles, including parameterization based on bank constraints or self-learning elements upon receiving new results.
As part of building the probabilistic model, a process was established for collecting initial dispute data and outcomes, as well as collaboration with the client's legal department.
The implementation of the probabilistic model led to the formulation and execution of a series of tasks to automate all dispute management processes — reducing requirements for new employees and shortening their onboarding time.
Economic effects: Improved customer satisfaction. Reduced staffing needed to service issued bank cards. Higher quality execution of dispute cycle procedures.
