Veterans’ Affairs tests using AI to tackle 82,645 unprocessed claims

The Department of Veterans’ Affairs (DVA) is preparing to pilot artificial intelligence to help manage a growing caseload of 82,645 ex-service personnel seeking financial support.

The agency has developed a proof-of-concept tool called MyClaims, which uses AI to extract specific medical details from the extensive documentation that accompanies veterans’ claims for compensation or benefits.

DVA was among the first agencies to trial a secure Azure-hosted environment established by the Department of Finance aimed at supporting AI development and testing under the federal GovAI strategy.

Speaking at the AI Government Showcase in Canberra, the DVA’s chief data officer Alicja Mosbauer revealed that, prior to the last year, it had taken an average of 315 days to process veterans’ claims.

While recent increases in staffing have helped reduce that figure in the past year, Mosbauer said the process remains slow as staff may spend days or even weeks manually reading and assessing medical documentation.

“We have a massive number of people trying to do this claims process,” she said.

“They’re not trained medically. They have to go through and manually tag a series of different PDFs to understand the nature of the claim and which body parts it relates to.”

DVA began using the GovAI environment in May as part of the program’s beta phase.

“We don’t have access to this [same] technology in our current environments through our current service arrangements,” Mosbauer said.

“It was a really great opportunity for us to test some ideas that we’ve had about where we could speed up some of those things.”

For the proof of concept, DVA assembled a synthetic dataset and used AI to extract and summarise the required information to assess a claim.

“It goes through and identifies all of the potential body systems and body parts that are part of that claim,” Mosbauer explained.

“It’s important for us to be able to identify the date, the date information for body systems, what body part, but then also generate a summary.

“We’ve been able to very quickly and very easily prove out that we can do this a whole lot faster.”

To mitigate privacy risks associated with veterans’ personal and medical information, DVA partnered with Circle T to develop a redaction feature that removes sensitive details, such as names, when documents are stored in SharePoint.

“Then, in the [GovAI] environment, AI can still be utilised to pull out and extract the medical information that we need to use to determine a particular claim,” Mosbauer said.

With the redaction capability in place – and with users’ consent – Mosbauer said the agency is now ready to move into the pilot stage and begin testing on real users’ data.

She noted that staff within Veterans’ Affairs—many of whom are ex-service men and women—have volunteered their own past medical records as use cases for pilot testing.

“Having a real use case will help us build out that claims [tool] a little bit faster,” Mosbauer said.

“The GovAI environment has demonstrated how, [since May] we can go very quickly and very easily into a really well-developed concept and have an understanding of what our roadmap is and how we can start taking those steps forward.”