The Joint Artificial Intelligence Center is working to develop machine learning models that can go beyond the capabilities of robotic process automation and make decisions about financial transactions.
The Defense Department is building intelligent tools to spot and fix accounting errors without the assistance of humans – or the basic bots that have increasingly be deployed across government.
The Joint Artificial Intelligence Center (JAIC) in partnership with the Defense Innovation Unit (DIU) is working to develop artificial intelligence/machine learning models that can go beyond the capabilities of robotic process automation and make decisions about financial transactions.
By coupling machine learning with RPA, JAIC hopes to solve some of the more complex challenges in financial data.
Automated financial management services are already used throughout DOD to resolve accounting errors and process data for a variety of audit requirements. The Army, for example, used automated solutions to eliminate nearly 30,000 labor hours in FY19 and expects to free up more than 100,000 labor hours in FY20, according to a JAIC blog post.
RPA, however, is unable handle unmatched transactions, which are those where the invoice number, amount, method of payment or any number of details cannot be reconciled. It can take humans anywhere from a day to months to track down and resolve these unmatched transactions, adding up to billions of dollars in unresolved financial activity for DOD. The current backlog of about a year compounds the problem.
According to JAIC, unmatched transactions are too complex for RPA and other types of automation to handle, leaving humans as the resolvers of last resort.
The new automated ML models would “be able take irregular, complex financial information and data, make decisions and apply judgments and solutions to a high level of accuracy or confidence without having to have a human in the loop,” said Rachael Martin, Mission Director for JAIC’s Business Process Transformation Mission Initiative.
To find companies to build these models, JAIC called on DIU to identify commercial solutions. More than 50 vendors provided solution briefs, and within four months, contracts were awarded to two vendors Vertosoft and Summit2Sea as part of the Humanless Unmatched Transactions (HUnT) program.
Both firms will integrate ML and AI software into existing RPA infrastructure, each working with different business systems to see which company delivers the better solution.
According to JAIC, the HUnT models will first learn to categorize unmatched transactions as either simple or complex. Then they will learn to triage the transactions, sorting them by ease of resolution, level of importance and the likelihood RPA could fix the issue with additional instructions. Some will still be so complex that a human must resolve them.
If the HUnT models are successful, they will not only save time and money, but they’ll help DOD increase financial compliance and get a better idea how its dollars are being spent, JAIC said. Additionally, the software that’s developed will be built out into Advana advanced analytics platform to facilitate agile development and experimentation with new AI capabilities.
“This is definitely one of our most important proofs of concept,” Martin said. “It’s a capability that right now people are trying to figure out how to incorporate and leverage, so this is a good opportunity to be able to help provide that.”
NEXT STORY: Ready for a VR-enabled work environment?