phonlamaiphoto - stock.adobe.com
Using AI for finance processes has been a compelling idea for years, and it's not hard to find companies that are really doing it.
Machine learning, natural language processing (NLP) and intelligent chatbots are taking over much of the tedious work of accounting, reporting and auditing. They're even performing basic financial analysis and decision-making that used to be unique to humans. But several experts agreed that AI for finance departments is still in the early adoption phase.
Companies are either piloting AI or using it for narrow purposes, said Adrian Tay, managing director of finance and CFO services at Deloitte Consulting. Most are working to identify use cases that will enable them to deploy AI more broadly.
"One CFO recently shared with me that he understood AI at a high level but lacked the time and expertise to really implement a full AI strategy," said Jack McCullough, president of the CFO Leadership Council, an association of senior financial executives, in an email. "[AI for finance] is certainly the love child of analysts, but CFOs in total are not thinking about it."
The long-term potential sounds promising, however.
"We're at the front end of a long and sustained rate of adoption that is going to build over the next three or four years," said Robert Kugel, senior vice president and research director at Ventana Research.
The best AI use cases
Anecdotal evidence suggests AI excels at financial processes that involve repetitive operations on large volumes of data.
Robert KugelSenior vice president and research director, Ventana Research
"It will eliminate the need for people to do a lot of the boring, repetitive work that they're doing today," Kugel said. "It will make it possible for systems to wrap themselves around the habits and requirements of the user, as opposed to the user having to adapt how they work within the limitations of technology."
Data quality will also improve and, with it, the quality of analytics as AI gets better at flagging errors for people to correct, Kugel said.
AI is also helping with tedious accounts payable tasks, such as confirming that goods were received and that an invoice contains the right items, Tay said. Companies that use automated payments are deploying machine learning to scan payment patterns for deviations.
"If the machine learning algorithm tells them that the probability of the goods having been received and everything being good with that specific invoice, they'll pay that immediately," Tay said.
The software can also look for outliers from the usual spending patterns by vendor, product and region, he said.
Risk sensing is another early use of AI for finance departments. Companies use it to scan news for threats to critical suppliers, he said.
Early deployments of AI for the finance department
A number of organizations have started using AI for finance, though it is challenging to get them to talk. Vendors and system integrators are willing to name some, though.
PwC, for example, reported that client Microsoft uses machine learning and intelligent process automation (IPA) in analytics that performs real-time reviews of sales to ensure resellers comply with the Foreign Corrupt Practices Act. The software can flag potential corruption risks by identifying relationships and anomalies in individual sales.
At financial company Citigroup, employees developed robotic process automation (RPA), data visualization and NLP tools to automate processes such as financial reporting and generate management commentary on financial results.
Some PwC clients are using AI to make the analytics produced by financial planning and analysis (FP&A) departments more predictive or prescriptive, said Bob Woods, partner at PwC.
"We're seeing a lot of work around particular sales forecasts, external forecasts [and] demand forecasts," he said.
Many clients start by using RPA to move information, transform it at the data layer and then add AI or some form of analytics to produce insights (see sidebar). The AI takes on much of the grunt work so FP&A workers can concentrate on analysis.
Meanwhile, the role of AI for finance workers in controlling and compliance is to automate risk prediction, controls and account reconciliation.
"Many companies have multiple environments that they're trying to bring together and reconcile to get comfort with the data before they start using it for predictive forecasting," Woods said.
Tax professionals, which need specialized skills to keep up with country regulatory requirements, are also early users of AI for finance, he said. AI quickly transforms the data they need and makes it more reliable.
Smart bot or not?
Some things that look like AI technically aren't. Chatbots -- and RPA generally -- are increasingly used to automate finance. But, unless a bot has some AI capability for, say, speech recognition, RPA in finance is not strictly a type of AI for finance. Such chatbots are considered rules-based, meaning they can only take the if-then steps programmed into them. Bots with AI are algorithmic: They use algorithms to analyze unstructured data and learn over time, according to Woods.
View the distinctions in terms of attended vs. unattended and rules-based vs. algorithmic, Woods said.
"If you think about attended rules-based, that's just traditional desktop RPA, [someone] using a bot to orchestrate certain tasks, but the person is involved through the entire process, ensuring that the bot executes based on rules that that person has defined," he said.
Next is unattended rules-based, which is a middle ground between RPA and IPA, Woods said. A person established the rules but lets the bot run unattended until an exception calls for intervention.
"As you move to the algorithmic, the difference between rules and algorithmic is: You're not creating a distinct [if-then] decision tree," Woods said. "You're creating a set of parameters."
The bot learns from the parameters and modifies its own rules within them.
This unattended algorithmic type of bot is using true AI to make decisions on its own.
Such applications are still "very early days" as companies sort out security and privacy concerns, Woods said.
There are other areas where using AI for finance shows promise.
"More and more companies and CFOs are starting to take a look at machine learning and AI to help with forecasting," Tay said. "They're also using that to help their employees in the initial problems of baseline budgets, which are typically a very intensive task."
Some companies use NLP-equipped bots to answer questions, such as a person's expenses for the year and whether they were paid.
"Reporting is also another big area in terms of natural language generation, automating some of the initial insights by using the machine versus the analysts writing it up themselves," Tay said.
NLP also enables searching of finance data as machine learning takes over some of the data management of analysts. It can also read and analyze contracts that can number in the thousands.
Many experts think this ability of AI to spot discrepancies in huge databases is tailor-made for the auditing services of the Big Four accounting firms: Deloitte, EY, KPMG and PwC.
"It can actually scan through and identify terms or conditions or things that we have trained it to look for and highlight those for an auditor," said Will Bible, partner in the audit and assurance practice at Deloitte US. "The auditor can follow up and review and understand the context around those items. It definitely provides the first pass that accelerates and expands how much our auditors can cover," he said.