Microsoft Treasury and Microsoft’s Cortana intelligence team partnered together to build a machine learning forecasting solution for accounts receivable. This addresses a key exposure for the company and exceeded expectations by improving forecasting accuracy and operational efficiency for the team.

Cortana, how much cash do we need today?

The challenge

Microsoft has substantial exposure to foreign exchange (FX) risk, owing to its operations in over 150 countries.

It is the job of its treasury team to continuously evaluate and manage this risk. In the past, the team used forecasts generated by a series of Excel spreadsheets, using inputs from disparate groups across the company that needed to be aggregated and rationalised in a very manual and tedious process.

These exposures were then hedged by the team using FX forwards and options, effectively reducing the risk of the forecasted exposure to zero.

However, the uncertainty in these forecasts often led to excessive variance between actual and forecasted FX exposures, resulting in increased P&L volatility in other income (FAS 52).

Past challenges include:

• Inconsistent definitions, processes and reports.
• 80% of analysts’ time spent collecting and compiling data.
• Less than 500 subsidiaries that need to be included in consolidation of exposures.
• Accuracy – poor forecasting leads to P/L volatility.

The solution

Microsoft Treasury and Microsoft’s Cortana Intelligence Suite team partnered together to build a machine learning forecasting solution for accounts receivable. This addresses a key exposure for the company and exceeded expectations by improving forecasting accuracy and operational efficiency for the team.

The teams worked jointly on obtaining and analysing the historical data, defining the data processing pipeline and cleaning the data for modelling. The machine learning team then applied their techniques to create an automated forecasting solution using the coding language R.

Since the solution was created in Azure Machine Learning it is accessible by web where any user with access can run the forecasting or consume its data without any need for any knowledge of coding in R or a separate installation.

Best practice and innovation

The treasury team has successfully leveraged machine learning solutions to reduce FX volatility, which has helped to increase shareholder value. “Azure Machine Learning has significantly helped improve our hedging forecasts, saving real dollars during a period of volatile currency markets,” says Oystein Harsvik, Director at Microsoft.

This solution has helped improve treasury efficiency, as FX forecasts are now auto-generated using Cortana Intelligence Suite. Accuracy has also improved, as it removes potential human bias from accounts receivable projections and improving overall accuracy. Together these have resulted in cost savings through improved forecasting a reduced FX impact.