In 2009 Canada began transforming pay for its ~290,000 federal employees: centralize the work of 101 departments into one pay centre in Miramichi, New Brunswick, and replace a 40-year-old pay system with a customized PeopleSoft platform called Phoenix. IBM was hired to design, build and deploy it — the only company to bid. The customisation was enormous: more than 80,000 pay rules from 105 collective agreements, interfaces to 34 separate HR systems, and over 200 custom-built programs bolted onto the package.

The staffing cut was the fatal companion. About 1,200 pay advisors were eliminated across the 46 participating departments, replaced by 460 advisors plus 90 support staff in Miramichi — and those departments lost direct access to enter or correct their own pay data. Phoenix went live in two big waves, 34 departments in February 2016 and 67 more in April, with the old system retired behind it. Errors surfaced immediately — thousands underpaid, overpaid or not paid at all — and by fall 2017 the Auditor General found outstanding pay requests had quadrupled since launch.

The bill compounded. The Auditor General warned a sustainable fix would take years and far exceed the $540m the government budgeted. CBC's access-to-information release showed IBM's stage-one contract of $5.7m had grown to $185m across 39 amendments, with more than 1,000 software defects still open. By July 2024 a Public Services and Procurement Canada official put the total cost of fixing Phoenix at about $5.1 billion — against roughly $309m originally spent to create it.

The replacement, Dayforce, was procured in June 2024 on a ten-year, $350.6m contract with an explicit commitment: no big-bang cutover this time. Departments will adopt it in small parallel-run phases — a design dictated entirely by the memory of Phoenix.

$310m to build; $5.1bn officially spent fixing pay errors by 2024; IBM's stage-one contract of $5.7m grew to $185m across 39 amendments

Staffing was cut in lockstep with the cutover: 1,200 advisors to 460, while the departments that understood their own pay rules lost the ability to fix errors directly.

An off-the-shelf product was stretched far past its design: 200+ custom programs over PeopleSoft to encode 80,000 rules meant every interface was a new failure point.

The go-live was big-bang across 101 departments in two waves before the first wave was stable, with no parallel-run fallback once the old system was retired.

Procurement concentrated risk instead of pricing it: one sole bidder, a 1,700-page contract that itself assigned 'critical decisions' to the government, and 39 amendments with no visible cost cap.

The Auditor General found no comprehensive governance structure to resolve pay problems sustainably — triage replaced engineering.

Don't cut the experts and their system access in the same stroke you switch systems. Phase the move, run old and new in parallel, and keep enough people who know how the pay actually gets made.

Phoenix remained in service through the 2020s, partly because a successor had to be built around its failure: manual workarounds, satellite pay centres and an ongoing error backlog cost hundreds of millions a year. Pay-related class actions and damages added to the tally. The 2024 Dayforce contract institutionalizes the counter-lesson — phased adoption, parallel testing, no single cutover weekend — while taxpayers continue running two payroll platforms at once until the migration completes.

FOLLOW THE EVIDENCE

The sources

  1. Report 1—Phoenix Pay Problems (2017 Fall Reports of the Auditor General of Canada) web.archive.org
  2. IBM contract cost for failure-plagued Phoenix payroll system jumped to total $185M cbc.ca
  3. Fixing problems with Phoenix payroll system cost taxpayers $5.1 billion: official thecanadianpressnews.ca