In this two-part article series, we discuss technology-driven solutions for improving cash forecasting accuracy. Part One focuses on developments in machine learning tools and how those can help.
Treasurers recognise the opportunity to improve cash forecasting accuracy to drive working capital efficiencies. Cash forecasting accuracy has long been a challenge, but in the past decade, the ease of borrowing at low interest rates had helped cushion the impact of uncertainty. Companies could rely on abundant liquidity to blunt the effects of forecasting inaccuracy and unanticipated events. Then the COVID-19 pandemic hit in early 2020, with dramatic effects on revenue, production, and supply chains. Businesses tapped credit lines and stockpiled cash to cover funding requirements. In one week in March 2020, banks provided more than USD240 billion in new loans to businesses—double the amount normally dispensed over the course of a full year.1
Today it remains difficult to forecast the continued, reverberating impacts of the pandemic. Additionally, market liquidity, inflation, interest rates, and foreign exchange volatility are among top-of-mind treasury concerns, all variables that are hard to predict but that can significantly impact cash flow.
Amid a period of heightened uncertainty, it becomes even more imperative to understand cash, liquidity, and working capital requirements not only with accuracy but with speed, and, depending on the business model, to be able to forecast more frequently.
Persisting manual constraints
Automation drives forecasting accuracy and timeliness, and, through process simplification, enables a shorter forecasting cycle. However, many companies still have a highly manual forecasting process, from data collection across internal and external sources from a patchwork of systems to the use of notoriously error-prone spreadsheets. While 91 per cent of treasury professionals employ cash flow forecasting in some form, 72 per cent manually collect and categorise data to generate a forecast.2
Poor cash forecasting links to inadequate visibility over cash flows and the risk of fraud. These longstanding challenges, despite the plethora of forecasting tools available today, make the end-to-end forecasting process cumbersome for many companies. The result impedes treasury’s ability to unlock the full value of the forecasting function.
Applying machine learning for accelerated analysis and accuracy
While treasury workstations can help in aggregating fragmented data, they have fallen short in unlocking insights and predictions from the data. Machine learning (ML) capabilities potentially could help to improve the use of treasury workstations for cash forecasting and is particularly useful for detecting historical patterns and identifying trends that would otherwise require years of direct experience to grasp.
For a treasury, investing in ML can be a resource-intensive, multi-year project. In the immediate term, consider leveraging the expertise of partner banks which have also invested in ML-driven forecasting tools.
For example, Standard Chartered leverages unsupervised ML algorithms such as ARIMA, Holt-Winters, and ETS to iteratively analyse and learn from balance movement trends and other inflows and outflows of data as well as variances until achieving the expected level of accuracy. This supports areas such as balance forecasting, trends analysis, and even investment decisions.
Learn more about simplifying access to data sources, connecting forecasting insights to working capital levers, and more in Part Two of this article series.
1 “As Virus Hobbles Economy, Companies Race to Tap Credit and Raise Cash,” by Kate Kelly and Peter Eavis, March 31, 2020, The New York Times.
2 “72% of Treasurers Still Do Cash Flow Forecasts Manually”, by PYMNTS, February 25, 2022.
Visibility, control, and optimisation are unwavering liquidity needs. Yet demand for efficiency is escalating. With automation central to our solutions, we can help you achieve it all.