The insight-led solution was built around two requirements:

• producing a list of claims, prioritised based on significance and resolution probability, updated daily
• classifying rejected claims into categories, aligned to consistent workflows

The BSG team opted for a lean, agile approach, creating a minimal viable product (MVP) to show benefit early. The machine learning model developed could accurately prioritise incoming claims, based on significance (value, age, type, etc.) and the probability of resolution.

Timing of claims emerged as a firm driver of priority. In many cases, claims in a suspended state in workflow due to lead times in the process, were incorrectly actioned by debtor controllers, resulting in time being spent on cases that could not be resolved. By appropriately prioritising cases, the model directed debtor controllers to spend time on cases with a higher likelihood of resolution (i.e., effort was prioritised where it could most benefit the bottom line).

A second model was built to classify rejected claims into categories aligned to specific workflows. This provided clarity to all debtor controllers on the procedures to follow when actioning claims.

During the data-discovery phase, it emerged that debtor controllers relied on reports drawn from several separate systems to investigate and resolve claims. A significant amount of time was spent switching between systems to extract these reports. Consolidating all the necessary information into one dashboard, resulted in a 50% improvement in efficiency of debt collections, directly impacting revenue collected.

The combined effect of the model and dashboard was to provide debtor controllers with a prioritised list of claims each day, as well as all the information they needed, and a recommended action to resolve the claims.