Inside the battle to stop Phoenix from infecting troubled PS pensions
‘Contaminated’ pay data spells new trouble for retirees, beneficiaries
No one thought it would come to this.
Miramichi and Shediac should have been working in tandem. Instead, two years after the launch of the unfortunate Phoenix pay system, these New Brunswick towns find themselves in an awkward embrace of technology and management gone wrong.
Miramichi of course is the epicentre of the government’s frantic effort to fix Phoenix — the deeply flawed decade-long project to modernize and consolidate a pay system that affects 300,000 government employees, about two-thirds of whom have their pay processed through the Miramichi office.
Nearly 700 pay administration workers there are struggling to whittle down a backlog of some 600,000 pay queries and transactions, most dealing with incorrect amounts on employees’ pay stubs.
Now these errors are threatening the government employees’ pension plan, run by a staff of 730 plus in nearby Shediac.
The Shediac employees are well placed to watch the drama unfold. They operate a call centre and handle the intricate details of a system that includes more than 850,000 government employees, pensioners and beneficiaries.
The centre’s basic mission is to calculate retirees’ pensions using multiple pieces of software including a calculation engine known as Penfax. Before Phoenix launched on Feb. 24, 2016, the job was relatively straightforward.
Pension administrators collected pay data from human resources colleagues throughout government, and combined it with information about employees’ pensionable service. Then they applied a formula to determine the pension to be paid out.
But for the past two years, pension administrators have watched with mounting concern as pay errors piled up in Miramichi and began infecting the pension system as well.
“This would be our worst nightmare,” says the administrator of a pension system for provincial government employees, “to be importing polluted salary data into our system.”
The extent of the pollution is unclear. But the resources devoted to cleaning it up are both significant and growing. Prior to February’s federal budget the government had hired nearly 400 extra staff to deal with public servants’ salary errors. About 100 pay specialists were hired at Miramichi, the rest at temporary satellite offices in Ottawa and elsewhere. The government this year is adding another 500 pay employees, which would bring the total to 1,500 compared to the originally targeted 550.
The goal is to fix salary errors before they are forwarded to Shediac. However, pension administrators there can’t yet count on a clean stream of salary data — not when pay administrators have yet to make an appreciable dent in the backlog at Miramichi.
In response Shediac has added staff as well. Two months before Phoenix went live, the pension centre employed 650. It has since added 85 people. Significantly, 55 of the new hires have been assigned the job of “validating data received from Phoenix.”
This is intensive work. The LinkedIn job description of one Shediac employee notes he has been tasked to “review and correct all incongruencies in Penfax generated by Phoenix pay system.” This exercise involves recording all pension account activity “for future reference and to keep track of errors and patterns causing the inconsistencies.”
Correcting mistakes is labour intensive. To verify a pension, administrators often must track a career through multiple federal departments and collect pay and pension documents going back more than 30 years. The pension calculations must take account of career interruptions triggered by training courses or maternity leave.
The good news, if it can be called that, is that the number of government workers potentially affected so far seems manageable.
While a majority of the 300,000 federal government employees who rely on Phoenix have received incorrect salaries since 2016, fewer than 30,000 have so far entered the pension system. This includes 20,800 public service employees, 1,650 RCMP members and 5,460 members of the Canadian Armed Forces.
Equally fortunate, military and non-civilian RCMP personnel had not yet switched their pay systems to Phoenix when problems with the latter became apparent. The result: Their salary data is not plagued with the same kinds of errors as that of other federal government employers.
Public Services and Procurement Canada, the federal department responsible for both the pay and pension centres, said a recent audit of pension payments revealed an error rate of just two per cent — about the same as during the two years prior to Phoenix’s launch.
However, this statistic doesn’t tell us much about the extra effort now being expended — such as assigning extra staff to double-check Phoenix data — before sending out pension cheques to new retirees.
Through spokesman Pierre-Alain Bujold, Public Services confirmed that in the case of workers employed by the federal government before Phoenix went live, the number of records requiring a pay correction has increased 25 per cent. (For government workers hired in the past two years, there’s been a 60-percent jump in the amount of effort required to correct pay records but since these employees won’t be retiring anytime soon, their experience has little bearing on the pension system.)
“The situation was fine until Phoenix came along,” said an industry official familiar with the inner workings of the new pension system.
There is also a side issue whose importance will only become clear in the months ahead. It concerns the quality of some of the salary and employee service data entered into Penfax after the 2003 launch of the project to modernize the pension system — data that is suddenly relevant for people retiring now.
Penfax, which was designed by James Evans and Associates of Victoria and is employed by other large pension plans across the country, does not appear to be the issue. Several pension experts we consulted say they have manually double-checked pension and other calculations produced by Penfax.
“I’ve never had a problem with it,” said an actuary based in the Ottawa area.
Rather, the difficulty appears to lie with process errors such as inputting data using incorrect software codes. This appears to crop up during calculations that involve anomalies from full-time, regular service — leaves without pay, parttime service and maternity leave, for example.
If this seems an echo of the Phoenix project, it’s not surprising. The government’s pension-system upgrade, like that of the pay system, did not initially include software patches to include all the variables involved in an employee’s career, which forced pension and pay workers to do manual workarounds. The projects were also driven by the same two themes.
In both cases, Public Services and Procurement Canada invested in efficient new software systems, then centralized administration in New Brunswick. The thinking was that predicted efficiencies would allow the government to run its pay and pension systems with substantially fewer employees. As we know, the two systems played out very differently.
In the pension project, the government hired Hewlett Packard Enterprises (now DXC) and James Evans and Associates (among other tech firms) to implement Penfax and related technologies.
“It was a huge effort to clean up the data,” said a private sector engineer involved with the project in its early days. “We were dealing with a 35-year-old pension system.” Nevertheless, the engineer insisted the data “was for the most part fixed during conversion to the new pension system.” The major issue today, he added, is the pay data being imported from Phoenix.
It may be more complicated than that. Consider the case of Laura Prevost, a former Canada Revenue Agency employee who retired in 2014 and, as she did so, arranged to buy roughly 18 months’ worth of extra pension credits (as she was entitled to do). Her initial pension did not include the impact of the extra credits, known as a buyback.
Prevost anticipated that her pension payments would increase once the paperwork involving the buyback went through. Instead, in 2016, she was shocked to discover her pension deposit actually dropped by roughly $100 per month.
This was the start of a two-year odyssey dedicated to uncovering the source of multiple errors in her pension account. Along the way she has spent thousands of dollars on legal and actuarial fees. Prevost’s quest has been all the more painful because 2016 was also the year after she separated from her husband.
Prevost and her lawyer Eric Letts learned that, during her first couple of years of retirement, she had been receiving too much pension because the system had over-estimated the number of years during which she contributed toward her pension.
The details are complicated, involving stretches of maternity and other types of leave. In some cases Prevost opted to continue contributing towards her pensions, in other cases, not, as the rules permitted. Emails obtained through access to information offer a revealing glimpse into how pension administrators cope with the pension system’s various quirks.
For instance, a note from Lisa Hébert, a Public Services official, informed Prevost her service credit (number of years worked) had been overstated by more than a year because some stretches of her leave-without-pay had not been “coded correctly in the system.”
More revealing is a note penned in 2016 by Jean-Philippe St- Onge, another pension official, who explained why Prevost had received an incorrect estimate for the cost of buying additional pension credits. “The information in the system was erroneous, which led to erroneous counselling,” St-Onge wrote. “This often happens with old service (records) from Canada Revenue Agency as the information recorded in the system back
then (pre-Penfax) was often erroneous.” He added this was why it was so important to verify salary and service data before issuing pensions.
This email surprised Prevost. “My legacy data and CRA records were correct,” she said. “What I don’t know is how the data that wound up in Penfax became incorrect.”
Errors do happen in pension systems, especially one as large as the one run by the federal government. The key is how administrators deal with them. Part of the reason the pension system run by Shediac didn’t collapse into Phoenix-like crisis is that administrators were given sufficient time to sort out most of the issues — the system wasn’t rushed into service.
The pension system’s call- centre technology and software programs were developed from 2005 to 2013 — roughly double the length of time government project managers spent crafting the Phoenix pay system.
Not only that, the public service, RCMP and military were added to the new pension system in stages from 2013 to 2017, giving administrators plenty of time to learn the quirks of Penfax, Siebel and Oracle software and to develop workarounds where necessary. In contrast, Phoenix was launched in two phases early in 2016, separated by mere weeks.
Nevertheless, while this helps to explain why the project to modernize pensions worked when implementing Phoenix did not, Prevost’s situation also points to some serious potential problems as pension administrators try to cope with the rush of Phoenix errors. Phoenix is nowhere near being repaired and the flow of retirees is both significant and unrelenting.
Not only that, dealing with pension quirks is extremely labour intensive — no fewer than six pension administrators were involved in counselling Prevost about her service buyback options, for instance. Nor is her lawsuit against the government resolved. While pension officials, according to email correspondence, have acknowledged errors in processing her account, there is still a dispute about the timing of her pension service buyback — a matter Prevost estimates involves some $40,000 worth of pension assets.
While cases such as Prevost’s could be unusual, they do illustrate the vulnerabilities in the pension system.
The administrators in Dieppe have enough on their hands without having to worry about the extra complications of Phoenix pay errors. Unfortunately, this is where they are — playing preventive defence in a game that may have years to play out.
This would be our worst nightmare, to be importing polluted salary data into our system.