Centrelink's canned $191m engine took minutes to do what existing system did in seconds

A revamped entitlements calculation engine (ECE) for Centrelink was scrapped because it took “minutes” to produce results the existing engine could do in two seconds.

Executives for Services Australia, which oversees Centrelink and Medicare, told an audit committee on Friday that the ECE - produced by Infosys, using Pega technology - had proven slow to perform, and did not play nicely with other systems in production.

The program was scrapped in July and the $191 million spent was written off; it was used to process only 784 claims, and only ever run in parallel with the existing system, ISIS.

Services Australia - and more importantly, the government at the time - had wanted ECE because it promised to allow welfare policy changes to be implemented into Centrelink’s systems more quickly.

Chief information and digital officer Charles McHardie - who inherited the already-active project when he joined in mid-2020 - said that any change to entitlement rules currently needs to be “hardcoded” into the ISIS system, using the Model 204 programming language.

Services Australia’s CEO Rebecca Skinner - who also joined when the project was already underway - said that made change difficult.

“At the moment if we’re asked to make a change [to the entitlements calculation] based on a policy change, we probably regularly frustrate by saying, ‘That will take us several months to do because we’ll need to build it in this code, and we’ll need to test it because you’re in this entitlement calculation piece, you don’t want that going wrong for broad payments’,” she said.

“I think sometimes others would like us to do that quickly, and we can’t do that.”

ECE was meant to fix that, making the engine simpler to change, such that the work could also be performed by people with lower levels of technical specialisation.

By having to scrap it, “the piece that’s lost as part of this is the ability to have faster and more flexible changes made,” Skinner said.

“Changes to entitlements will still be a costly and somewhat slow exercise.”

The executives said the ECE was ultimately scrapped for two reasons: $191 million was spent on a system that couldn’t get close to the existing performance the organisation was used to, and that did not perform well in a production setting.

McHardie said that the ECE was capable of calculating a payment at the right amount and for the right date: producing an identical result to the existing system, ISIS.

But it took far too long to produce the result to be of any use.

“The ISIS system at the moment does its calculations within two seconds,” he said.

“For a claims processing officer that has done all of their work around eligibility assessment, they push what’s called the assessment button, it gets handed off to ISIS, ISIS does all the calculations it needs to and within two seconds it has the answer back to the staff member. 

“So that’s obviously a big productivity issue for us. We can’t have a system where a claims officer is sitting there waiting for minutes for an assessment to come back, and that was pretty much where we were. 

“We had a system that could calculate but was just not performing and we did have problems with scalability.”

Skinner noted that Centrelink handles about half a billion dollars in payments each day; at that volume, the two-second benchmark was a necessity.

McHardie said ECE also had issues when it was put into production.

For most of its ‘life’ at the agency, ECE ran in either development or test environments. The latter sought to replicate the production environment as much as possible, but ultimately could not simulate all eventualities.

“The problem we had when we put it into the production environment where it had to sit there and interoperate with ISIS, and with SAP [was] we [had] a lot of performance issues, we had a lot of defects and problems,” McHardie said.

“We got it into production and running but it was slow and had defects.”

The executives indicated that errors were essentially made at the design phase for the ECE, in the approach taken to building the system.

A “bottom-up” approach was taken initially, where the ECE tried to convert existing code in ISIS into pseudocode and then [into] Pega rules”.

“Infosys thought they had a credible solution, I think, which is if they could speed up the process by extracting the rules, understanding the rules, transpiling the rules into Pega rules, it would’ve been fine, but it didn’t perform.

“So what happened during the life of the program is Infosys had a pivot and took a ‘top down’ approach, pulled out the lines of legislation around aged pension and used a coder army - pretty much - to write the rules,” he said.

When it became clear that the project had issues, Services Australia even took a drastic step of “de-scoping it to de-risk it”.

However, Skinner said that the findings from a gateway review process led to the recommendation to discontinue the ECE and write it off.

“We’d been engaging through our gateway review process, which finally also came back to us and said: ‘You’re really going to be in a position where you need to decide whether the conversation you have with government is about large amounts of more funding to try to get this thing to work for every payment across the board; or you stop and have a rethink about what’s the best way to go ahead’,” she said.

“At that point it would’ve been a very expensive proposition and many many years of work to try to get to the point where you could use that technology base that had been developed as part of the new ECE, which just did some work on aged pension. 

“Based on our professional knowledge and experience, we felt that the best thing to do there was to call a stop and impair the asset.”

Skinner said some of the IT hardware used for the ECE had since been repurposed.

The project had also informed other uses of Pega inside the agency, which were occurring in spaces “where the policy environment is significantly simpler” than welfare.