Using innovative Microsoft technologies like machine learning, PowerBI and Power Platform the collections team introduced an algorithm that forecasts a customer’s likely payment pattern and reliability based on historical data, along with cash flows and potential write offs. A suite of various machine learning predictive algorithms allows the team to forecast “risk of loss” customers, driving more collections efforts on those risky accounts rather than focusing the collections process on customers that pay regularly.

The team identified that a single predictive modelling technique (machine learning) would not address the problem and instead explored the capabilities of multiple predictive modelling techniques. They built a bespoke forecast model using advanced machine learning techniques like Deep Neural networks, LSTM, time series forecasts and boosted trees.

The model has been built by the AI team taking data from the company’s main subledger system, and then feeding it into their own model. The model consumes the data and generates its own data forecasts every week, highlighting the key details.

Based on the historic payment behaviour, the collections team have categorised customers into distinct categories called Multi Entity Profiling. This in turn categorises outstanding payments in distinct categories like “forecasted at risk” and “forecasted to receive” based on customer profiling.

“Forecasting the amount against each outstanding payment helps us identify the overall expected AR at a specific point of time, whilst also forecasting bad debt write off for which Microsoft needs to plan,” says Apra Bothra, Sr Finance Program Manager.

The findings assist the collections team in prioritising the collection effort for these high value payments which are likely to get delayed, in turn impacting the company’s DSO and operating cash. Additionally, it has assisted in enhancing the organisation’s overall cash forecasts, which has led to better cash planning.