Machine learning is far more integrated into accounting, planning and forecasting operations than many professionals realize, finance technology consultants said in a CFO.com webinar.
A lot of the cloud-based enterprise resource planning (ERP) and other platforms finance teams use already have machine learning built in, and that will only increase, said Kris Murphy, principal of The Hackett Group.
"It’s going to get to the point where ... when you enter a forecast number, the machine learning algorithm will say, 'Last time you missed by 10%, you average a miss of 8%, maybe you should think about that number,'" said Murphy. "I believe that is where we’re going to be going with machine learning."
In an informal poll conducted among finance professionals attending the webinar, two-thirds said they either didn’t use or didn’t know whether they used machine learning in their decision tools, and a quarter said they were intending to build it into their process in the next two years. Almost 20% already had it built in at some level.
Steadily improved results
Machine learning is a subset of artificial intelligence and takes one of two forms. In supervised ML, the finance team trains the algorithm by inputting company performance data from the last 12 or 18 months, along with the expected outcome. In unsupervised ML, teams input the historical performance data without the expected outcome, leaving the algorithm to find patterns, clusters and correlations in the data autonomously.
Today it’s mainly used to set up touch-free processes for any rule-based operation that involves a lot of repetition, like the accounts payable and accounts receivable functions. The algorithm can also help reduce error by spotting patterns that humans easily overlook.
"It can take the load off of accountants to look at PDF or hand-written invoices and enter those into a system," said, Justin Gillespie, principal of The Hackett Group. "It can recognize handwriting even better than a human, looking for things human may not be looking for."
It’s also becoming increasingly useful in forecasting, not as a substitute for the input finance leaders receive from functional operations but as an add-on to help provide another perspective.
"Everybody’s very good at forecasting, but when you use ML algorithms, it goes back over historical data — let’s say 48 periods of data — and we find there’s actually more accuracy with ML algorithms predicting sales or costs," said Murphy. "It’s very powerful."
The accuracy comes from ML tweaking its results as new data is introduced.
"The [business] drivers are constantly being updated and getting better every time data comes in," said Murphy. "The more data you feed it, the better the algorithm becomes."
What’s more, with the cloud, virtually unlimited storage space and computing power enables businesses to add large quantities of unstructured data such as weather patterns, macroeconomic trends and government regulations to the traditional structured data in an ERP. Combining the two types of data magnifies the forecasting power.
"If you’re looking at product profitability and what’s driving costs up or down, you can bring in … key metrics of the business cycle, social media and other data to predict what revenues might be based on the launch of a new product and what people are saying," said Gillespie.