We talk so much about cash forecasting and it’s importance, however how we approach this critical task has barely changed in the last 10 years.
We may have added more data and reference points to help the analysis, but fundamentally we are looking at what has happened, adding in some general ledger data, the FP&A forecast and hoping for the best. The treasurer may go a step further and add their own view of the business progression to give an extra flavour. But the real science behind the numbers is rarely understood or utilised.
…And we wonder why we some months we are so accurate, and others we are so wildly inaccurate it’s comical.
In this blog I explore how to approach the task of cash forecasting differently. How to use the latest technology to break through into the new era of looking-ahead. I spoke to the expert developers we have here at Palm to help me fully understand the power of the tools we have at our finger tips.
The Opportunity: Where we can add real value by using the latest tools?
Data
The data we use in our forecasts doesn’t have to be purely financial data relating to your internal operations. Think outside of the box about other external factors that effect the success of the business. Is it the weather forecast, the performance of the economy or the marketing of competitor products? Whatever it is, you can use this data to improve the quality of your cash flow forecast.
Details
Imagine you had unlimited time to analyse every piece of data you feed into your forecast, for example, you could build in logic to understand the payment behaviour of each of your customers knowing some always pay between 3-5 days late depending one which day the invoice falls due and when their weekly payment run is. Or, a model that understands that before year end, some invoices due for payment are brought forward to limit open balance sheet positions. However, instead of spending hours analysing and creating this logic, the model understands these trends and does it for you.
Direction
Unlike your best Treasury Analyst, your machine learning forecast model will never leave you. Therefore spending time teaching it all that you know, speaking to it directly to explain why somethings are correct and others aren’t, is a time investment that will pay off in dividends over the years. You can tell your forecast, for example, that it has misunderstood the seasonality of your business due to strange one-off behaviour and to exclude this from the its trend analysis in future. Giving this feedback regularly will also help the model learn more about the business and the trends it detects.
The Technology: What do we need to achieve this?
Although not new, it is surprising how few treasuries are using AI and machine learning in their day to day operations. According to the 2024 Deloitte survey, it is less than 2%. Is this because the technology is not properly understood? Or it’s application hasn’t been facilitated in a way that is accessible to treasurers? My suspicion is that it is a mix of both. What are the technologies you can use to realise the benefits we’ve outlined in the above sections?
Predictive AI
Using different data sets, AI can predict what will happen with the future of your cash flows. The difference between this technology and your treasury team itself, is that you can supply many different and varying data sets in different formats to feed the model. This will have a limited impact on the amount of time the forecast takes to produce however will significantly improve the accuracy.
The difference between predictive AI and generative AI is that predictive AI uses existing data sets to predict what will happen in the future and is therefore most relevant for producing you cash flow forecast. Generative AI however creates new and original content, such as the commentary to support the forecast and understand the cash flows.
Machine Learning
Mathematical models are the foundation of machine learning, often using several at once and testing each of them until you find a blend that works. These can be refined over time to achieve the optimal solution. To do this within your team you would need access to a data analyst or machine learning expert, however Palm uses a suite of models which are uniquely blended to your cash flows to give you the best possible outcome in forecasting each of your cash categories.
The Reality- What will happen when you bring these technologies into your treasury?
When applying these systems and models you will be required a change your expectations of your forecast model. The advantage of using mathematical statistical models to produce your cash forecast means it will predict your cash flow for you, however you must relinquish your power to the model. You need to accept that you will not be able to click into a cell in excel and see how a number is calculated. You will not be able to “follow it through”. As many of the trends of patterns it detects are not necessarily explainable. Of course you can adjust the forecast and amend it to include the additional information you have, you could also feed it with a set of assumptions you believe to be true as guidance for the model to follow, for example, headcount is set to grow by 2% each month from now until the end of 2025 however on the most part you will not be able to see its inner workings.
To become comfortable with this will take time and adjustment, building trust in the model over time and conducting regular variance analysis to pinpoint the route cause of any differences will help both you and the model. However, it is in embracing these innovations and new ways of looking at data that we can truly unlock our cash and bring efficiencies into our treasury teams.
