The slide said 99.6 percent payroll accuracy. Green tile, second bullet, quarterly operations review. Twelve countries, about 5,000 employees.
The CFO had spent the previous fortnight inside a purchase price allocation and was in no mood. Ninety-nine point six percent of what?
The payroll director said accuracy of payroll processing. She asked again. Per employee, per payslip, per line item? Nobody knew. The figure came off the provider’s monthly service report and the report did not say.
Two weeks later someone worked it out. The provider was counting pay elements. The arithmetic was correct. The metric was worthless.
Ninety-nine point six percent of what
Do the maths in the open. Five thousand employees, paid monthly. Count the elements on one payslip: base, overtime, premiums, allowances, both sides of each social contribution, income tax, the year-to-date carriers. Forty is a conservative average. Deloitte’s 2025 payroll benchmarking survey, 15 multinationals with a median headcount of 87,104, found 47 percent maintain more than 1,000 active pay codes.
Forty elements times 5,000 payslips is 200,000 elements a month. An element-level error rate of 0.4 percent is 800 wrong elements.
Errors do not scatter one per payslip; they cluster. One wrong tax code moves the tax line, the net, the employer contribution and four year-to-date accumulators. Eight bad elements per affected payslip is normal when you count. Eight hundred divided by eight is 100 payslips.
One hundred payslips out of 5,000 is 2 percent. One employee in fifty got the wrong net pay. The same month, the same data, reads as 99.6 percent or 98.0 percent depending on the denominator. The board sees the first number. Employees live in the second.
Most teams are not gaming anything. They never defined it. Deloitte’s global payroll benchmarking survey, 750-plus organisations across 55 countries, found 89 percent of payroll KPIs were not monitored or tracked. PayrollOrg’s 2025 Getting the World Paid survey of 585 respondents found 38 percent track no payroll metrics at all.
Two accuracy numbers that survive the follow-up question
Payroll accuracy is payslips requiring a correction divided by payslips issued, per country, per period. Not pay elements, which is what you count when you want the number to look good. OpsDog defines payroll accuracy against total payments processed; CloudPay reports issues per 1,000 payslips.
The unit has to be the payslip: that is what an employee receives and what a labour inspector asks to see. A correction is any change to an already-approved payslip: paid off-cycle, netted next period, or fixed after approval but before the bank file went out.
Corrections caught internally still count. That rule is the one that makes the metric real. Count only what employees complained about and you are measuring their attentiveness, not your payroll. So report the total, then the split between caught before payment and caught after, which is the only way to tell a team with good controls from a team with a clean month.
The trap is definition drift. Somebody will soon argue a €2 rounding fix is not a correction. Resist a materiality floor in year one, because the floor always rises to wherever the number needs it.
What is first-time-right at the run level?
The share of pay periods closed with no off-cycle correction run and no adjustment carried into the next period. The denominator is runs, not payslips. Germany monthly, twelve runs a year, ten closed clean, 83 percent.
CloudPay’s Payroll Efficiency Index, drawn from more than a million payslips across 130-plus countries, put first-time approval at 74 percent globally and 69 percent in the US. Not the same measure, but a quarter of runs do not clear on the first pass.
First-time-right is binary per run: a country with one wrong payslip scores the same as one with 400. Never publish it without the payslip number beside it.
Error source attribution is the number that gets the fix funded
Every correction gets one primary source code, entered by whoever fixes it: HR master data, time and attendance, inbound input file, configuration, calculation, or statutory change not applied. Six codes, mandatory before the correction can close.
Deloitte found 38 percent of off-cycle payments were caused by upstream processes, with terminations at 30 percent, missed or inaccurate HR updates at 20 percent and time and attendance at 17 percent. CloudPay’s index puts nearly two thirds of payroll issues, 63 percent, down to data input mistakes. PayrollOrg’s 2025 survey ranked the top three causes of accuracy problems as poor quality data inputs, late time-tracking data and inputs arriving after cutoff. Not one is a calculation problem.
This metric moves behaviour because it changes the ask. Show that 61 percent of last quarter’s corrections traced to one HRIS field three recruiters type by hand and you are not asking for headcount. You are asking for a validation rule on one field, and that gets approved.
Attribution needs the same code set on the same correction record in every country, which is where a per-country patchwork stops you. In HR Blizz a correction runs as its own controlled off-cycle run, so it stays countable as a correction instead of disappearing into the next regular cycle.
Cycle time has four segments and an average that hides the constraint
Measure cutoff to calculation complete, calculation to approval, approval to bank file transmitted, and elapsed calendar days from cutoff to pay date. Per country, per period, the first three in hours.
An aggregate of 5.5 days to run payroll gives you nothing to fix. If cutoff to calculation is four hours and calculation to approval is three days, your constraint is a controller in another time zone, not your engine. Deloitte found more than 30 percent of organisations take over four days to complete processing, with over a third of EMEA and APAC respondents exceeding six days against under 8 percent in North America.
The average hides the problem because your close is gated by the slowest country. Twelve countries averaging three days with one at eleven is an eleven-day close. Report the maximum with the country named, and anchor the clock to the published cutoff timestamp, not to when payroll felt ready.
Off-cycle ratio and retro volume belong beside cycle time because they move first. Off-cycles divided by total payments, and pay elements with an effective date inside a closed period per 100 employees. Deloitte’s global average was 0.23 manual off-cycle payments per employee per year, 0.30 in North America and 0.03 in EMEA. Accuracy only moves after something went out wrong. Off-cycles and retros climb the week upstream data starts arriving late, a quarter before accuracy notices.
Cost per payslip and penalties, without flattering yourself
Fully loaded means internal payroll labour, HR and finance time on payroll tasks, vendor fees, per-country statutory and filing fees, technology allocation and the cost of corrections, divided by payslips issued. Annually, by country.
Then the benchmark trap. Deloitte’s global payroll benchmarking survey put cost per payslip at $3.70 in North America, $6.31 in APAC and $21.39 in EMEA. By headcount band it ran from $33.60 below 500 employees to $2.14 above 25,000, and the US-only bands move differently again, $14.60 below 500 employees against $2.16 in the 10,000 to 25,000 band. Any global average across 40 countries where half your entities employ fewer than 50 people is an artefact of your entity structure. Segment by country and by headcount per entity or do not publish it.
Put corrections in the cost. EY’s December 2022 survey of 508 US payroll practitioners put the average error at $291, so a hundred corrections a month is $349,000 a year, which is what funds the validation rule.
For the board, two compliance numbers: filings submitted by the deadline divided by filings due, per country, and the money value of penalties, interest and late fees with the reason attached. The IRS failure-to-deposit ladder runs 2 percent at one to five days late, 5 percent at six to fifteen and 10 percent beyond fifteen; HMRC’s RTI penalties run £100 to £400 per late month by employer size. Alight’s 2024 study of around 300 payroll professionals found 53 percent had incurred a payroll penalty in five years, rising to 67 percent among companies operating in two to five countries. That peak in the two-to-five country band reads as a coordination problem rather than a calculation problem, which is why it shows up in the penalty line before it shows up in accuracy.
Both run off a filing inventory that is almost certainly incomplete, and the returns that bite are the municipal ones nobody listed.
The reason code nobody enters, and the metric that never existed
Query volume per 100 employees per period, with reason codes and first-contact resolution. Deloitte’s 2025 survey found half of respondents have no resolution SLA at all and a quarter target 72 hours or more. The reason codes are the value: if 30 percent of queries concern one allowance, your payslip label is wrong, not your calculation.
Manual touch count per run predicts every other number. Count every human action that changes data after cutoff: uploads outside the standard interface, direct edits to a calculated result, overrides. Deloitte found 30 percent of organisations that outsource named manual entry and adjustment their most time-consuming activity.
None of this exists unless two things are true. Correction runs have to be tagged as corrections, distinguishable from a genuine off-cycle bonus, or your numerator is fiction. And the reason code has to be mandatory at entry, not a reconciliation project at quarter end. An optional code is blank or “Other” on most records within two periods, and then the metric never existed. Keep the list closed, six to ten values, owned by one person.
Five numbers on one page
Put these in front of the CFO.
1. Payslips corrected as a percentage of payslips issued, by country, split before and after payment.
2. First-time-right runs by country.
3. Top three error sources by share of corrections.
4. Worst-country cycle time from cutoff to bank file, country named.
5. On-time filing rate and penalty value year to date.
The fantasy target is 100 percent accuracy, every country green, a slide that never changes. The real one is a first year where the number gets worse before it improves, because you started counting the corrections you used to fix quietly. Nobody runs 30 countries on three HR systems and manual timesheets at 99.9 percent of payslips uncorrected. My own rule of thumb, from years of counting these, is that somewhere around 98.5 percent of payslips uncorrected, with error sources concentrated in two named upstream systems and falling, is a good operation. That is a working benchmark, not a published one, so set your own from your first four honest quarters.
On Monday, pull every off-cycle run from the last quarter onto one sheet and write next to each one where the error came from. If you cannot answer for more than a third, you have found your first project, and it is not an accuracy project. If you want to see what tagged correction runs and a 30-column audit trail look like inside a platform, ask HR Blizz to show you the correction records.
The penalty rates cited here are general information rather than tax advice, and they are set by each authority.
FAQ
Q: How should payroll accuracy be calculated?
Use payslips requiring a correction divided by payslips issued, per country per period. The payslip is the right unit because it is what the employee receives and what an inspector examines, whereas measuring accuracy per pay element inflates the result: a 0.4 percent element error rate on 40-element payslips can still mean one employee in fifty received the wrong net pay.
Q: What share of payroll errors originate outside the payroll team?
Published data consistently puts the majority upstream. CloudPay’s Payroll Efficiency Index, based on more than a million payslips across 130-plus countries, attributes 63 percent of payroll issues to data input mistakes, and Deloitte’s global payroll benchmarking survey found 38 percent of off-cycle payments were caused by upstream processes, led by terminations, missed HR updates and inaccurate time and attendance data.
Q: What is a realistic cost per payslip benchmark?
There is no useful global figure. Deloitte’s global payroll benchmarking survey of more than 750 organisations across 55 countries reported cost per payslip at $3.70 in North America, $6.31 in APAC and $21.39 in EMEA, with a range by headcount from $33.60 per payslip below 500 employees to $2.14 above 25,000. Cost per payslip only means something when segmented by country and by headcount per legal entity.
Q: Why do payroll KPI programmes fail?
Usually because the reason code is optional. If a correction can be closed without recording its source, the field is blank or “Other” on most records within two pay periods and error source attribution becomes impossible. Correction runs must also be tagged as corrections in the system so they can be separated from legitimate off-cycle payments.