Client’s treasury function makes 12 month rolling predictions for hundreds of cash flow time series on a monthly basis. The predictions are done manually thus leading to about 350+ man hours of effort on a monthly basis as well as the accuracy of the predictions are lower than desired. Hence, the client’s treasury function wanted to implement a solution to generate these forecasts automatically with a high degree of accuracy.
INXITE OUT APPROACH
We collaborated with the treasury team to understand the business forecasting process in detail and draw up a holistic list of parameters which impact cash flows and hence the predictions.
Features were generated for the following class of factors:
- Historical cash flow pattern for each cost center in the organization
- Volume, value, and timing of sales and purchase invoices
- Macro-economic factors like GDP, GDP Growth, Wage Growth, Inflation etc.
- Treasury business process related factors like mechanism and timing of tax payments, reconciliation etc.
To predict the cash flow of cost center, two models were applied in sequence:
- Outlier Detection Model: A robust outlier detection model was implemented to detect unexplained / erratic cash flow patterns related to acquisitions, change in payment mechanisms, human error etc.
- Cash Flow Prediction Model: Multi-variate regression-based model to predict the future cash flows based on the features identified above
Model developed achieved the following:
- Project was awarded the “Excellence in Treasury” award from the European Association of Corporate Treasurers
- Enhanced forecasting accuracy by ~20-30% over the existing manual forecasting process
- Achieved 80% reduction in man hour requirement of 350 hours on a monthly basis