Members of the Public Service Alliance of Canada protest on the three year anniversary of the launch of the Phoenix pay system, in Ottawa, on Feb. 28, 2019.Justin Tang/The Canadian Press
Dorothy Eng is chief executive officer of Code for Canada.
Launched in 2016, problems with the Phoenix pay system have cost the government at least $3.5-billion, while thousands of public servants have spent years waiting to resolve issues with their pay. At the end of last year, the system had a backlog of 383,000 transactions – 52 per cent of them more than a year old.
The government now wants to clear those old transactions before adopting a new, desperately needed payroll platform later this year, so the clock is ticking to deliver results. Enter: AI.
Since the fall, Public Services and Procurement Canada has been testing a virtual assistant tool that uses artificial intelligence to speed up the backlog-clearing process. Now, they’re looking to expand its use.
The problem? No number of new tools will solve the issues at the heart of the federal government’s payroll issues.
If used correctly, AI can benefit the public service by reducing administrative burdens and unlocking potential efficiencies. However, the federal government has a track record of making costly mistakes when implementing new technology, and there is reason to fear that it may repeat them with Phoenix’s replacement.
Why did Phoenix fail? There are many reasons, but to name a few: an overwhelming number of rules and processes, including 72 job classifications and 80,000 pay rules, requiring more than 300 customizations built into the payroll system; a lack of proper testing with users before a major rollout; and dated procurement processes that favour large vendors and waterfall methodologies.
Concerns are already being raised that the potential new payroll platform, Dayforce, may have similar issues. The Public Service Alliance of Canada has said that its members aren’t being meaningfully consulted in the selection process, pointing out that even planned feedback sessions with public servants are focused on testing simple navigation tasks, such as updating a beneficiary or requesting leave, instead of ensuring reliability and accuracy of pay transactions.
The federal government’s own report on Dayforce as a replacement says it is “viable” but that there are gaps in its ability to handle HR and pay functions, requiring either the simplification of rules and processes or additional cloud extensions.
This all adds up to a government that continues to avoid best practices when it comes to the procurement and deployment of technology. By focusing on large off-the-shelf products from big-name vendors and avoiding the hard work of meaningfully testing and developing solutions with public servants, the government continues to set itself up for public failure.
This brings us back to the use of AI to address Phoenix’s backlog. By all accounts, the backlog-clearing work is a solid use case for an AI tool, mirroring other successful case studies such as the U.S. Department of Veterans Affairs’ use of AI to process a backlog of disability benefit applications. PSPC is also applying several ethical AI principles to the work, building in human oversight, rolling out the tool in stages and monitoring its performance closely before expanding its use.
But we shouldn’t be leveraging the potential of AI just to clean up messes created by bad technology selection and implementation – especially when there’s no guarantee we won’t repeat the same mistakes.
And the stakes are incredibly high. Public trust in government continues to fall, and the misuse of AI has the potential to erode it further. The federal government is taking a serious risk by tying its use to a system known across the country as a public failure.
Unless the government makes serious changes to how it selects, tests and deploys technology within the public service, it continues to be at risk of headline-grabbing failures such as Phoenix and ArriveCAN. AI is a new technology, but it is being used within the existing systems that have repeatedly failed us. If we don’t start doing things differently, Ottawa may have a new mess to clean up soon.