Weeks to Seconds — Services Australia's AI Engine for Sorting the Document Deluge

Centrelink, Medicare and Child Support share a single document lodgement service in Australia's Services Australia. Caseworkers previously had to manually open and classify every citizen-uploaded file (e.g. medical certificates, bank statements) before an assessment could begin, and this manual process was overwhelmed by the 2020 COVID-19 claims surge.

From mid-2019, Services Australia worked with delivery partner Capgemini to automate document intake, aiming to classify and extract data from citizen-submitted files as part of the Document Management Modernisation through Intelligent Automation project.

The agency and Capgemini built an AI-based document management modernisation engine combining optical character recognition (OCR) and natural language processing (NLP) to automatically read, classify and extract data from uploaded documents.

By October 2020 the engine was processing more than 25,000 document lodgements a day with over 95% classification accuracy. Capgemini reported that the time from lodgement to caseworker-ready assessment fell from weeks or days to seconds. The agency said it was still working to fully quantify downstream outcome metrics, and no independently audited figures on error-correction rates, caseworker time saved, or claim-outcome accuracy have been separately published.

The strongest public evidence for this practice remains contemporaneous trade-press interviews with Capgemini and Services Australia rather than an independent evaluation; the rubric assessment (overall score 60/100) notes strong scalability (already handling full national volume) but weaker transparency and limited demonstrated transferability.