By C. Todd Lopez
WASHINGTON (Oct. 23, 2020) -- The Army has multiple enterprise financial management IT systems, some decades old, that account for the majority of the service's financial, acquisition, and logistic transactions. These systems communicate transactions in a variety of ways through a complex architecture, but that communication is far from perfect. When a financial system communicates inconsistent or no data to another system, it creates a problem that requires human intervention.
"One of our consistent issues within Army financial management is caused by our large portfolio of legacy systems executing hundreds of thousands of transactions per week with one another,'' said Jonathan Moak, who serves as Army's principal deputy assistant secretary for financial management and comptroller within the Office of the Assistant Secretary of the Army, Financial Management and Comptroller (ASA (FM&C)). ''Incorrect information is often generated or reflected in a system during these constant transactions, which can create the issue called an unmatched transaction.''
The Army experienced somewhere between two and three million of these unmatched transactions (UMTs) in the 2019 fiscal year, driven by its multiple enterprise system, each built with different requirements. These UMTs, all of which need correcting, have included inaccurate data of obligations for financial payments and issues with the mismatching fields between a financial and logistics supply system.
"It's a big problem...with a total value of several billion dollars,'' said Moak. ''These mismatches lead to a general lack of accountability, funds control, and have a negative impact on buying power -- all of which are critical to auditability."
Before the Army can be deemed auditable, a majority of these UMTs must be resolved. Resolving just one UMT is a labor intensive process that can take multiple hours. Since there are currently millions of UMTs that need resolving, the problem cannot be solved by manual labor alone.
The Army is building the requirements to field a core enterprise system where UMTs are not created in the first place. In the interim, Moak said that a team led by Chase Levinson in ASA (FM&C) has implemented robotic process automation, or RPA, to help resolve UMTs as they occur.
RPA is an automation tool where RPA ''bots'' follow the strict business rules given to them by developers. "RPA relies on conditional statements that say if you see this, then do this, which requires a well-defined process with very clear rules to fix a UMT,'' said Levinson.
However, Levinson said that some UMTs do not always have a clear path to resolution, which means that when the RPA bot can't resolve the UMT, humans still have to fix them manually.
According to Moak, continuing this process with UMTs is not an optimal use of manpower, which can be redirected to higher-level financial management operations and analysis by implementing a more effective solution.
''Our priority is to give the right tools to our workforce to accomplish the mission at the greatest level of efficiency,'' said Moak.
To further add to its capabilities to solve the UMT problem within its financial management enterprise, the Army turned to the Joint Artificial Intelligence Center and the Defense Innovation Unit.
While the Army was already using robotic process automation as a partial solution, a new solution, driven by artificial intelligence, could be even better. A well-trained AI-driven solution would better be able to deal with the variety of nuances that crop up due to the large number of financial systems the Army uses and the wide variety of transactions that take place.
"The Army came to us with this issue, saying that they had tried to do some initial modeling to automatically resolve these errors without human intervention, but hadn't been successful" said Rachael Martin, Mission Chief for Business Process Transformation at the Joint Artificial Intelligence Center.
Martin said the Army wanted to know if there was anything JAIC could do to help them not only build out a capability to help resolve problems with UMTs, but also have the capacity to do a better job in the future building predictive models and unburden human analysts from unnecessary work.
Martin said JAIC worked with the Defense Innovation Unit to identify existing best practices in the industry that might help the Army fix the problem with UMTs. She also said that at the same time, the Comptroller within the Office of the Secretary of Defense became interested in the effort, as OSD had problems similar to those of the Army.
"We were able to leverage the resources we allocated to support the Army to bring in a similar use case from Comptroller, and do some really innovative industry competition," Martin said. "We awarded contracts to two different companies with different approaches to building a UMT model so that we could test both and find the most promising solution.''
It was DIU that put out a request for industry solutions. The DIU, with headquarters in Silicon Valley, maintains relationships with best-in-breed vendors focused on solving similar problems for commercial customers from across the country and can quickly bring the best of what's happening in the commercial world to the Defense Department.
"When DIU put out requests for solution briefs, we received over 50 from commercial AI companies. And from there, they were down-selected through a couple of different rounds, and used demonstrations to help us down-select and interview them," said Eric Dorsey, the government contractor who is the project manager on the DIU team. "It finally came down to two companies to award what's called prototype contracts."
Now, each of those companies has been assigned to work with either the Army or the OSD comptroller to find an AI-driven solution to solve problems like the Army has with UMTs, Dorsey said.
The two companies will have about six months to accomplish their work, Dorsey added.
"Using machine logic, which is more sophisticated than RPA -- the goal is to solve up to 70% of these UMTs automatically and correct them," Dorsey said. "The result is we could save millions of labor hours for the Army and the DOD each year."
After six months of work, around mid-March 2021, both companies will have completed their work with the Army and the OSD comptroller and will have working prototypes. The Army and comptroller will then go into field trials and decide, ultimately, if they want to go into production with the product, Dorsey said.
"We're hoping to take lessons learned with these efforts and scale to the other branches of the military -- Navy, Air Force, Marines -- to also help them solve some of their unmatched transaction problems in their accounting system," he said.
While the work of the JAIC and DIU may eventually help multiple services and the OSD, Moak said bringing in AI assets for financial work is going to be a necessity for the Army going forward.
''Our collaboration with DIU and JAIC on this effort is a tremendous opportunity for Army financial management, and will hopefully inspire others,'' said Moak. ''Applying these innovative tools now allows us to shift our focus to optimize our systems environment, and building a core enterprise system -- a sustainable solution for improving financial operations.''