Application of machine learning for forecasting receivables

Objective

The client is a major Russian oil&gas holding company. When shipping petroleum products under the credit payment scheme, there were often situations when a counterparty made a payment later than the date specified in the contract giving rise to overdue receivables. 

Effect from the solution

  • 450 m in receivables at the risk of being significantly overdue identified on a monthly basis
  • 70% overall reduction in the level of overdue receivables, minimization of losses from a debtor’s default (for example, in the event of bankruptcy)

KPMG approach to the project

  • Development of a machine learning model that, by analyzing data on the payment discipline of counterparty debtors and information from available internal and external sources, provides the most accurate and complete forecast of the counterparty’s behavior
  • Implementation and pilot operation of the developed machine learning model
  • Analysis of forecasting results becomes part of the credit risk management business process. The credit controller receives a signal that overdue receivables may arise and can proactively manage the interaction with the counterparty at the level of individual deliveries and adjust the cooperation strategy (e.g. through clarification of contractual terms, deferral of payments).