AI-based (artificial intelligence) cash forecasting tools greatly simplify the treasurer’s job – but striving to achieve 100 percent accuracy is a waste of valuable time, experts say
This article was published on 2 Feb 2022 on theglobaltreasurer.com and written by Selwyn Parker with contributions of our consultant Patrick Kunz
A flood of AI-based cash forecasting solutions holds great promise for treasurers – but don’t expect too much too soon.
That’s the advice of consultants as treasurers grapple with vast lakes of actionable, though largely dormant, data that could be tapped in a variety of ways if they only knew how.
As Patrick Kunz the founder and CEO of Netherlands-based consultancy Pecunia Treasury & Finance, tells The Global Treasurer, this is a long and arduous journey.
“It’s important to remember that cash forecasting has been in the top treasury priorities every year for the last ten years,” he says. “This basically means that it’s hard to get it right.”
The root of the problem is that accurate cash forecasting requires treasurers to perfectly predict the future.
“The future however depends on too many factors to fully predict,” he adds. “The same goes for corporate cash — too many factors, both internal and external, determine the future cash position. And when you factor in FX changes, it gets even more complex.”
Striving for perfection
The pandemic has complicated cash forecasting by triggering unforeseeable developments, for instance by changing the way payments are delivered into treasuries. According to Worldpay FIS’ latest Global Payments Report, there has been a behavioral shift towards digital and online buying that has required businesses of all sizes to become more digital. The report estimates that global e-commerce transaction value will exceed $3.9trn by 2024.
Yet promising solutions keep coming to the market which utilizes machine learning and AI capabilities to enhance the accuracy of cash forecasting systems.
And while AI and predictive analysis-based solutions certainly help, provided treasurers are systematic about implementing them, according to Kunz, a cost and effort analysis must be applied.
“The more time you spend on improving the model — and therefore the forecast – the more accurate the forecast will get,” he says. “This might also take much more time to achieve.”
But don’t expect perfection, he warns: “In my opinion, if a model is 80 percent accurate, it is already doing a good job. Spending time and money on getting the accuracy to maybe 90 percent might be lost time. The question is always whether this extra 5-10 percent accuracy will get you more information. Probably not.”
In other words, treasurers may be wasting their time by striving for perfect cash forecasting. “A weekly forecast should be ready by Monday before lunch. You also need time for other treasury work and have time to look back at the actual versus forecast,” says Kunz. “A smaller company should start with some predictive analysis tool based on seasonality and previous year results.”
The value of AI-based cash forecasting
If used judiciously, AI-based solutions are certain to make treasurers’ life easier though. Reconciliation is just one example, says Bob Stark, vice president of strategy for Kyriba. As he told an Association of Financial Professionals seminar just before the pandemic, AI and machine learning have a high potential for cash management and forecasting, particularly when reconciling prior-day bank files with yesterday’s expected cash position.
“This is one of the first cash management processes performed each day,” he said. “And for some organizations, the volume of transactions is so big that it can take hours and multiple people to do that reconciliation.”
Similarly, improving efficiencies in collections can improve cash flows.
“Because of the sheer volume of these organizations and the magnitude of the size – even if they’re bringing in an improvement of one day of sales outstanding it’s huge in terms of dollar values for these companies,” he said.
Post-pandemic, nothing has changed about the value of AI cash forecasting tools but if treasurers want to get the most out of AI-based solutions, they need to know what they are looking for, says Kunz.
“If I build a cashflow forecasting tool, I will start with the information at hand — accounts receivable and accounts payable information in the ERP. This will give you a 30-45 day forecasting window.”
For longer forecasting, he suggests treasurers use different information.
“For the 45-120 day window, you can use sales and purchase orders (that are not billed yet) and enrich that with expected sales from your budget or previous month or previous year’s month if seasonality is a big factor. For the longer-term forecast, full reliance on a budget or smart logic is needed as there just is no data yet.”
But don’t get too far ahead of your data, he warns: “The first three sound easy but these can be massive data sources with different formats and/or data quality. The Key is to integrate these into one model and use that properly for making future decisions.”
Then, and only then, should treasurers move on?
“When you have all these data sources in order — and you trust them to be correct and predictive, then you should start looking into more complex models which take into account future demand, price changes, competitor moves and future expected demand, and the economic situation. But this should really really step three or four – first get the basics in order.”
In short, AI is not a silver bullet – at least not yet.
“Usually companies already have a lot of information at hand, but it is just not used properly, or it’s too much to process.