An AI-based document-processing solution has proved invaluable to helping Services Australia keep up with the administrative work stemming from surging demand for government benefit programs during the COVID-19 pandemic.

The agency’s success came from ongoing work with consultancy Capgemini, which has been working to apply machine-vision techniques to speed the reviewing and validation of information in more than 25,000 documents lodged daily through its online Document Lodgement service.

The documents – which support citizens’ applications for benefits under programs such as Centrelink, Medicare and Child Support as well as COVID-era programs like JobKeeper and JobSeeker – were being manually reviewed until the Document Management Modernisation through Intelligent Automation (DMMIA) project applied a range of machine learning techniques to the problem.

Using an API-based architecture, the six-month pilot program developed a platform that allows citizens to submit documents online, then have their data automatically extracted and verified against internal and external sources using optical character recognition (OCR) and machine-vision code that applies AI-based learning to validate the information it extracts.

Teaching machines to read

The Python-based project “utilises really cutting-edge approaches” to do much more than simple OCR, Capgemini ANZ CEO Olaf Pietschner told Information Age, noting its use of neural-network techniques like periodic spectral ergodicity (PSE) and ResNet for regression-based deep learning.

This level of complexity is important because many users submit photographs of documents that may be rotated, low-contrast, of poor quality or riddled with smudges, stains, creases and other obfuscating details.

Once the information is extracted from the submitted documents, it is fed through a natural-language processing engine that validates the content, confirms citizens have uploaded the correct document, and automatically classifies the documents.

Demonstrated accuracy of more than 95 per cent “has resulted in a high degree of confidence that the approach really works and scales,” Pietschner said, “and the outcome has been a significant reduction in the time it takes – from weeks and days, down to seconds.”

“This really allows the Services Australia team to focus on the edge cases, where documents haven’t been automatically processed.”

Automation rules

The project, which kicked off in mid 2019, was already underway when the COVID-19 pandemic hit.

However, the Capgemini and Services Australia team – operating in lockstep thanks to an extensive effort around agile development – increased the pace of the pilot program as it became clear that efficient, accurate document processing would be critical to keeping up.

“COVID has meant there is more demand, and more citizens lodging requests, through the department than ever before,” Capgemini vice president for federal government Lysandra Schmutter said, noting that the success of the pilot program would see it scaling upwards and across Services Australia, where “there is a broader demand outside of those 25,000 documents per day”.

It could also be applied to volume document-processing environments in other departments – and, potentially, for tasks like verifying proof-of-identity document within financial services firms.

“We’ve got a big focus on helping government agencies to make it easier for Australians to access their services during the pandemic,” she explained.

“A lot of those interactions moved to digital channels during lockdowns, and we believe this can be applied to more use cases.”

Streamlining the mundane

Despite the controversy over the government’s ‘robodebt’ debacle, robotic process automation (RPA) has become a major focus for enterprises seeking to trim inefficient processes from their operations, with ever-better AI threatening to upend conventional roles as repetitive work activities are automated.

Gartner, for one, recently predicted that 90 per cent of organisations will have adopted RPA by 2022 – with overall spend increasing around 10 per cent annually and large organisations tripling their RPA capabilities by 2024 as they seek to automate low-value added activities.

“The decreased dependency on a human workforce for routine, digital processes will be more attractive to end users not only for cost reduction benefits,” distinguished research vice president Cathy Tornbohm said, “but also for insuring their business against future impacts like this pandemic.”

The novelty and value of the solution were this month recognised as the combined team accepted a regional innovation ‘imagination’ award for the project.

“Through their digital transformation, the agency continues to re-design the way they do business using new and emerging technology to improve their systems so they are easy to use and responsive to customer needs,” Capgemini noted in accepting the award.

“The work Capgemini and Services Australia is undertaking to provide an updated service that assists customers to submit, classify and verify documentation online has never been more important.”