Claims document reader pilot

DVA developed the Claims Document Reader to help authorised staff find, organise and review information in large and complex claim documents. The Reader uses artificial intelligence (AI) to reduce time spent searching, so staff can focus on assessing the evidence. The Reader does not make claim decisions. DVA staff remain responsible for checking the evidence, applying the law and making every decision. 

This webpage explains the Reader’s controlled pilot, initial evaluation and safeguards; it does not describe final rollout arrangements.

About the pilot

DVA, with support from RSL Queensland, established a controlled pilot of the Claims Document Reader using real initial liability claim documents from veterans who volunteered to participate in the pilot. The pilot tests the Reader’s suitability to help staff find and organise claim information.

DVA conducted the pilot within GovAI, a secure Australian Government platform. 

Choosing to take part

DVA ran the pilot with support from RSL Queensland, which helped identify veterans who may be interested in participating. Around 270 veterans voluntarily consented to their claim documents being used and received information to support their decision.

Participants received information and consent material explaining the pilot’s purpose, what claim information could be used and how AI-enabled technologies could interact with that information. DVA staff remained responsible for checking the evidence and making every claim decision.

Claims continued to be reviewed and decided by DVA staff. The consent and privacy safeguards outlined in this webpage apply to the pilot.

About the Reader

Why the Reader was built

Claims can contain hundreds of pages, with relevant evidence spread across many documents. The Reader was built to help staff find and organise this information more efficiently, so they can focus on the evidence and each veteran’s circumstances.

The Reader supports staff; it does not replace their judgement or responsibility for claim decisions. It does not change the claims process or evidence requirements, or guarantee a faster decision.

How the Reader was developed

DVA staff built the Reader pilot within the secure GovAI platform. DVA did not contract a vendor to build the Reader.

The Reader and the information it processes are stored in DVA’s secure computer systems, supported by Services Australia. The Reader uses GovAI, an Australian Government service managed by the Department of Finance, to access AI tools.

How the Reader works

What the Reader does

The Reader helps DVA staff find, organise and review information in large and complex claim files. Its results link to the original document and page so staff can check the evidence. The Reader:

• Reviews supporting medical, service and other claim documents made available to DVA for assessment.
• Identifies and organises information relevant to claimed conditions.
• Links findings to the original document and page.
• Helps staff identify information that needs further checking or explanation.

How the Reader processes documents

The Reader uses AI to read typed text, handwriting and information in images across documents of different formats and quality. The pilot uses a single AI model within the secure whole-of-government GovAI platform. DVA data is not used to train the AI model.

What the Reader does not do

The Reader does not:

• Decide whether a person can receive benefits, use legal powers to decide claims or replace DVA staff.
• Give a medical opinion or assess a person’s health.
• Create new evidence or change original documents.
• Change the law, policy or evidence needed for a claim.
• Make treatment decisions, decide whether treatment is medically needed or approve funding for allied health services or therapy.

Safeguards and oversight

Human review and accountability

Authorised DVA staff remain responsible for every claim decision. They must check relevant Reader outputs against original documents, consider missing or conflicting information, seek further information when needed, and apply the relevant legislation, policy and procedures.

Privacy and information protection

The Reader pilot uses safeguards to protect personal information and control access and use.

GovAI is the Australian Government’s secure whole-of-government platform for controlled access to AI models. Its safeguards protect government information and reduce the risk of unauthorised access or use.

• Access was limited to participating claims and approved users.
• Claim information and Reader outputs were not used to train the AI model.
• Only authorised DVA staff could use the Reader during the pilot.
• Staff checked that only documents from participating veterans entered the pilot.
• Copies with identifying details removed were stored in each veteran’s official record.

DVA handles, stores and disposes of information in accordance with government privacy, security and recordkeeping requirements.

During the initial evaluation, DVA staff removed personally identifying details before the Reader processed documents in GovAI. Any future processing without redaction would remain subject to accreditation of an Official Sensitive GovAI platform and approval of the relevant use cases.

Access controls and security

• Access is restricted to approved staff.
• User and administrator roles are separated to support oversight.

Governance and assurance

Before and during the pilot, DVA reviewed privacy, ethics, security, legal requirements and safe AI use. Ongoing checks cover access, monitoring, information handling and staff checking original documents. These safeguards work together to protect information and support staff review. DVA does not rely on any one safeguard alone.

Staff training and support

All DVA staff must complete AI training. Reader pilot users also receive training on their role, claims processing and handling sensitive personal information. This covers:

• Staff responsibilities when using AI and the limits of information it produces.
• Privacy, records and information-handling responsibilities.
• Safe use and checking information against original documents.
• Responding to missing, incomplete or conflicting information.
• How the Reader fits into claims processing.
• How to report errors, get help and refer serious issues.

Training supports safe use. Claims support staff and staff authorised to make decisions remain responsible for their work.

Pilot evaluation

What the evaluation tested

DVA conducted an initial evaluation of the pilot to check how well the Reader worked with real veteran documents and how it helped claims staff. The evaluation covered:

• How well the Reader summarised large documents and identified information relevant to claims processing.
• Document review time, accuracy and consistency.
• Staff confidence, trust, satisfaction and user experience.
• How the Reader fits safely into existing claims processes.

Initial findings

DVA completed the initial evaluation six months after the pilot began. The evaluation compared how staff reviewed claims with and without the Reader.

Initial findings suggest the Reader helped staff review claim information faster.

• Staff completed the activities tested around 2 times faster on average.
• For larger claims of more than 100 pages, staff completed the activities tested around 3 times faster.

These findings relate to the activities tested during the initial pilot and do not measure the total time a veteran waits for a claim decision.

The evaluation grouped several types of issues together, including Reader errors, missing documents, incomplete veteran records and errors in the manual comparison process.

The initial findings were based on a small number of claims. DVA is doing more testing to check time savings, accuracy and consistency, and how the Reader fits into day-to-day claims processing.

Limitations identified

The evaluation helps DVA identify where the Reader can help staff, where information may be wrong or incomplete, and what safeguards claims staff need.

• The Reader may miss relevant information.
• Handwriting or poor-quality scans may be misread.
• Information may be copied incorrectly or without enough surrounding detail to understand it.
• A summary may be incomplete or need further explanation.

How limitations are managed

• Reader summaries link back to the original documents and pages.
• Staff check relevant information against original documents.
• Missing, unclear or conflicting information is reviewed and resolved by staff.
• Feedback, error reporting, investigation and monitoring support ongoing improvement.

Initial evaluation findings do not remove the need for staff to check original documents, oversee use of the Reader and carry out more testing.

Supporting claims staff

Claims can contain hundreds of pages, with evidence spread across many documents. The Reader is intended to help staff find relevant information more consistently so they can focus on the evidence and each veteran’s circumstances.

The Reader does not guarantee a faster claim decision or change the normal claims process, evidence requirements or decision-making responsibilities of authorised staff.

The Reader helps staff work with information. It does not replace staff judgement or their legal responsibility to make decisions. Staff must still consider the evidence and circumstances of each claim.

Next steps

DVA plans to conduct a final evaluation using a larger number of initial liability claims to test the Reader’s performance and the initial findings before embedding it within DVA claims processing systems. DVA and RSL Queensland are also planning a separate pilot for permanent impairment claims.

More information

To ask about DVA services or the Claims Document Reader pilot, call 1800 VETERAN (1800 838 372). You can also read DVA’s AI Transparency Statement on the DVA website.