ERP

How to Run an ERP Parallel Test Before Go-Live (2026 Guide)

A step-by-step ERP parallel test guide for governments: payroll and billing runs, cycle count, staffing, reconciliation checklist, variance log, exit criteria.

Chris Rodney 19 min read
On this page
  1. At a glance
  2. What is an ERP parallel test?
  3. When do you need a parallel test, and when is a mock conversion enough?
  4. How many parallel payroll cycles should you run?
  5. How to pick the pay periods (and billing cycles) to test
  6. Who staffs the parallel test, and how do you handle backfill?
  7. Entry criteria: what must be true before the first parallel cycle
  8. How to reconcile old vs. new: the parallel test reconciliation checklist
  9. How to categorize variances: expected, explainable or defect
  10. Defect triage and retest
  11. Exit criteria and sign-off for the go/no-go decision
  12. After the parallel: what carries into cutover
  13. Frequently asked questions
  14. Before your first parallel cycle

An ERP parallel test runs the same live pay period (or billing cycle, or AP check run) through your old system and your new ERP, then reconciles the two results line by line. You run it after conversion and user acceptance testing, before cutover. Every difference gets explained, fixed and retested until the remaining variances are ones you understand and accept.

For a government, the parallel test is the last point where a configuration mistake costs you a spreadsheet row instead of a short paycheck, a grievance or a late retirement report. This guide covers how to plan it, staff it, reconcile it and decide when it has passed.

At a glance

A parallel test is a side-by-side run of real transactions in both systems, reconciled to the penny or to an agreed tolerance.

Payroll comes first. Utility billing and AP check runs need one too whenever an error would reach residents, vendors or employees.

As a rule of thumb, KCT plans for three payroll parallel cycles, with the last two clean.

You compare gross-to-net by employee, deductions, employer contributions, retirement reporting, taxes, leave balances and GL postings by fund.

Every variance is tagged expected, explainable or defect, and each defect is retested in the next cycle.

The output is a signed variance log plus a short exit memo for the go/no-go meeting.

What is an ERP parallel test?

An ERP parallel test is a controlled rehearsal in which you process actual production inputs (time, pay changes, meter reads, invoices) in both the legacy system and the new ERP for the same period, then compare outputs. The legacy system stays the system of record. Nobody gets paid or billed from the new one yet.

The point is to prove that configuration, converted data and integrations produce the same answer your current system produces, or a different answer you can explain. KCT’s ERP implementation service puts it plainly in its problem statement: integration and data issues often stay hidden until “the first parallel run exposes it.”

Parallel payroll vs. parallel financial, AP and billing runs

Payroll is the most common parallel test because the error tolerance is close to zero and every mistake has a name attached. The other parallel runs are narrower and usually shorter.

Swipe to see all columns →

Parallel typeWhat you run in both systemsWhat you compareTypical trigger to require it
PayrollOne full pay period per frequency (biweekly, semi-monthly, monthly)Gross-to-net per employee, deductions, employer costs, taxes, leave, retirement file, GL distributionAlways, for any payroll replacement
TimekeepingThe same period’s punches and timesheetsHours by pay code, overtime, shift differential, comp timeNew time system or new pay rules engine
AP / check runOne check or ACH run from approved invoicesPayment amounts, vendor remit-to, 1099 flags, fund and account coding, positive pay fileNew vendor master or bank file format
Financial periodOne month’s journal activityTrial balance by fund, cash, encumbrances, budget-to-actualHeavy custom posting logic or new chart of accounts
Utility billingOne billing cycle or routeBill amount by account, rate class totals, late fees, payment plans, AR agingAny customer information system replacement

Utility billing parallels deserve their own plan. In Richmond, Virginia, WTVR reported in September 2026 that customers on a payment program were wrongly charged late fees the day after launch, which the vendor described as “conversion related.” The article does not say whether a parallel bill run was performed. Late fee logic and payment plan balances are exactly what a parallel bill run is built to catch.

When do you need a parallel test, and when is a mock conversion enough?

You need a parallel test whenever the new system calculates something that reaches a person: pay, a bill, a vendor payment, a pension contribution. A mock conversion and reconciliation is enough when the new system only stores and reports data you already calculated elsewhere.

A mock conversion proves the data moved correctly. A parallel test proves the system does the right thing with that data. The two answer different questions, so a clean mock load does not excuse you from a payroll parallel.

Swipe to see all columns →

SituationMock conversion and reconciliationEnd-to-end testParallel test
Payroll moving to a new systemRequiredRequiredRequired
Utility billing moving to a new CISRequiredRequiredRequired (at least one full cycle)
AP moving with a new bank file formatRequiredRequiredRecommended (one check run)
GL and budget only, no new calculationsRequiredRequiredUsually optional; reconcile trial balance instead
Upgrade on the same platform, same rulesTargetedRegression testingOnly for changed pay or rate rules
Fixed assets, purchasing, permittingRequiredRequiredRarely needed

When the honest answer is “we’re not sure our rules came across correctly,” run the parallel. You pay for it in staff hours; skipping it tends to get paid for at the payroll window and the utility counter.

How many parallel payroll cycles should you run?

As a rule of thumb, KCT plans for three payroll parallel cycles: the first to stabilize data and tooling, the second to prove fixes, and the third to confirm a clean result. Add a cycle for each pay frequency or employee group that behaves differently, such as a 24-hour fire schedule.

Some public bodies set three as a floor. The London Fire Brigade’s November 2025 board report on its HR and payroll replacement states that three parallel runs “are always considered a minimum requirement” for an ERP implementation. After three parallel payrolls, the brigade still had what the report calls “intolerable variances,” traced them to a legacy system that had been in use for over 30 years, and pushed go-live from October 2025 to November 2025 and then to February 2026.

The parallel did its job there: the brigade, with 5,877 staff and a complex pay structure, found the problem on test payslips.

Plan your timeline so a failed third cycle does not break the project. A slip of one or two pay periods is far cheaper than a bad first live payroll. For how parallel testing fits the larger schedule, see the government ERP implementation timeline.

How to pick the pay periods (and billing cycles) to test

Pick periods that exercise your hardest rules, not your quietest weeks. A parallel run on a calm, no-overtime period proves very little.

Look for periods that include as many of these as you can:

Overtime, call-back pay, shift differentials and holiday pay

A retroactive adjustment or a step increase

A deduction change (open enrollment, a new union dues rate, a garnishment)

New hires, terminations with leave payouts, and a rehire

An employee group not covered by Social Security, if you have one

A month-end period, so the GL distribution and benefit billing line up with a close

For utility billing: a cycle with a rate change, estimated reads, final bills, and accounts on payment plans

Also avoid stacking risk in the period after go-live. The January 2025 metering and billing audit for Billings, Montana, prepared by SLserco for the city, found that the new customer information system went live during the year’s peak water use, shortly after a rate change.

The resulting exception overload delayed bills; Yellowstone Public Radio reported that bills went out on 60- or 90-day cycles instead of 30. The audit concluded that “the exception overload from a rushed go-live process was the cause of the problem.”

Who staffs the parallel test, and how do you handle backfill?

Your own payroll, HR, finance and billing staff run the parallel, because they are the only people who know what a correct result looks like. The integrator supports configuration fixes and comparison tooling, but should not be the party that declares the variances acceptable.

During a payroll parallel, your payroll team is processing the live payroll in the old system, re-keying the same inputs into the new system, and investigating variances. That is roughly two jobs at once for several weeks.

The Billings audit is direct about what happens when that load isn’t planned. Errors during the transition were “primarily due to inaccuracies in the initial setup of the new system and human error, driven by a lean staff, lack of training, system knowledge and documented procedures,” and the billing area “was staffed too lean for this type of project.”

Plan the backfill before the first cycle:

1

Name a parallel test lead who owns the variance log and has authority to assign work across payroll, HR and IT.

2

Move routine tasks (filing, data entry, phone coverage) to temporary staff or other departments for the test window.

3

Budget overtime for payroll and billing staff, and get it approved before the window opens.

4

Pause non-essential projects and reporting requests for the same weeks.

5

Have the integrator or implementation partner automate input loading where they can, so staff are not hand-keying every timesheet twice.

6

Embed project team members in departments as go-to experts during the test, as Accomack County, Virginia described doing in its September 2025 audit committee minutes while it ran two weeks of parallel payroll testing ahead of a September 28 go-live.

Department heads should also hear early that their timekeepers will be asked to correct entries quickly. Staff buy-in matters here; the guide to managing staff resistance to a new ERP covers how to bring them along.

If your team cannot absorb that load, treat it as a planning input and resource it early. KCT’s ERP implementation team leads end-to-end and parallel runs on the government’s side during the test, migrate and validate phase, reconciling until the numbers tie out, which takes the coordination and variance triage off your payroll manager’s desk while your staff keep ownership of the sign-off.

Entry criteria: what must be true before the first parallel cycle

Do not start a parallel cycle until the inputs are stable enough that variances point to real problems. Starting too early fills the variance log with noise and burns a cycle.

Use a short written checklist and get the project sponsor to confirm it:

The final mock conversion has been reconciled to legacy totals (headcount, year-to-date gross, deductions, leave balances, open AR) and signed off.

User acceptance testing is complete, with no open critical or high defects in payroll, benefits or billing.

Pay rules for every bargaining unit and employee group have been reviewed against the current contracts.

Interfaces are built and tested: time to payroll, payroll to GL, retirement system file, bank and positive pay files, benefits carriers.

Comparison reports exist in both systems at the same level of detail (employee, earning code, deduction code, fund).

Security roles are set so testers see production-like data without the ability to issue real payments.

The variance log template, categories and owners are agreed.

Backfill and overtime are approved.

The data validation step is the one most often rushed. A February 2026 performance audit presentation to the North Carolina State Board of Education on Winston-Salem/Forsyth County Schools found that financial data had been imported “without proper validation by Finance staff,” noted the payroll/HR go-live had moved from January 2026 to July 2026 due to readiness concerns, and recommended validating phase 2 data before that go-live.

How to reconcile old vs. new: the parallel test reconciliation checklist

Reconcile from the top down: control totals first, then employee-level gross-to-net, then the outputs that leave the building (bank file, tax deposits, retirement file, GL entries). Matching totals can hide offsetting errors, so you always go down to the employee.

Swipe to see all columns →

AreaWhat to compare (legacy vs. new)Level of detailPrimary ownerTolerance (KCT rule of thumb)
Control totalsHeadcount paid, total gross, total net, total employer costPayroll runPayroll leadExact match after explained items
Gross payEach earning code: regular, overtime, differentials, holiday, leave paid, retroEmployee by earning codePayrollZero unexplained difference
Pre-tax deductionsHealth, dental, deferred comp, flexible spending, retirement employee shareEmployee by deduction codePayroll and benefitsZero unexplained difference
TaxesFederal withholding, state and local withholding, Social Security, Medicare, taxable wage basesEmployee by taxPayrollRounding only, documented
Post-tax deductionsUnion dues, garnishments, child support, charitableEmployee by deduction codePayrollZero unexplained difference
Net payNet amount and split between direct deposit accountsEmployeePayrollExact match to the penny
Employer contributionsEmployer retirement share, health premiums, Medicare, workers’ comp basisEmployee and planFinance and benefitsZero unexplained difference
Retirement system reportingContribution file fields: earnable compensation, service credit, contribution amounts, member codesEmployee and file totalHR and payrollFile passes the retirement system’s format check; amounts match
Leave balancesAccruals, usage, payouts, carryover caps for vacation, sick, comp timeEmployee by leave typeHR and payrollExact match in hours
GL postings by fundSalary and benefit expense, liabilities, cash, by fund, department, project and grantFund and account stringFinanceExact match by fund
Position and budgetEncumbered salaries or position budget consumed, if you use position controlPosition and fundBudget officeExplained differences only
Payment outputsBank ACH file totals, positive pay file, check registerFile and runTreasuryExact match
Third-party filesBenefit carrier files, deferred comp vendor, garnishment remittancesFilePayroll and benefitsExact match

A few notes from public-sector projects:

State retirement systems often have their own format and timing rules, and rules vary by state. Confirm with your system whether you can submit a test file before go-live.

On GL distribution, your auditors will look at payroll allocations across funds and grants when they prepare the ACFR, so a payroll that is correct by employee but wrong by fund is still wrong.

Leave balances are often converted once and never compared again. Compare them every cycle, because accrual rules are where bargaining unit differences live.

How to categorize variances: expected, explainable or defect

Every variance gets one category, one owner and one status. Three categories cover nearly everything.

Expected variances are differences you designed on purpose, such as a new rounding method for overtime rates that was approved in writing during design.

Explainable variances come from the test itself or from the old system: a timesheet keyed differently in each system, a pay change entered in legacy after the parallel extract, or a legacy calculation that was wrong all along.

Defects mean the new system produced the wrong result because of configuration, converted data or an interface. Every defect is fixed and retested.

Watch the “legacy was wrong” items closely. When the parallel reveals the old system has been miscalculating something, someone with authority decides whether to correct it going forward, correct it retroactively, or bargain it. Log that decision; it may need HR, legal or labor relations, and possibly a council briefing.

Setting a variance tolerance

As a rule of thumb, KCT sets net pay tolerance at zero unexplained cents per employee, and treats a rounding difference as acceptable only if it is documented, consistent and approved. Aggregate tolerances (such as “within $50 for the run”) are risky because one employee overpaid by $25 and another underpaid by $25 nets to zero.

For utility billing, agree on a tolerance per bill (often exact, aside from documented rounding) and review rate class totals separately. For AP, payment amounts and vendor bank details must match exactly.

Sample variance log layout

Keep the log in one shared workbook, one row per variance per cycle, with the cycle number in the ID. The rows below are illustrative only.

Swipe to see all columns →

IDEmployee / accountElementLegacy vs. newCategoryRoot cause and fixOwnerStatus
P1-014Emp 10482, FireOvertime, 24-hour shift412.18 vs. 398.40DefectWork period set to 14 days instead of 28; correct the rule and retest in cycle 2IntegratorClosed, passed cycle 2
P1-022Emp 20991, ClericalUnion dues28.50 vs. 31.00ExplainableNew dues rate keyed in the new system only; align inputsPayrollClosed
P1-031All, PoliceNet pay0.01 roundingExpectedApproved rounding change, documented in design decision 47Payroll leadAccepted
P2-006Fund 501 WaterGL salary expense84,210.55 vs. 83,977.10DefectTwo positions mapped to the wrong fund; fix mapping and retest in cycle 3FinanceOpen
P2-011Emp 33017Sick leave balance96.00 vs. 92.00 hrsExplainable, legacy errorLegacy accrued past the cap for 3 years; correction decision neededHREscalated

Add columns for date found, date closed and the person who verified the retest, so an auditor can follow each item from discovery to closure.

Defect triage and retest

Triage defects daily during each cycle, and fix them before the next cycle’s data is loaded. A defect without a retest is not closed.

A simple rhythm that works for most cities, counties and districts:

1

Hold a short variance standup each working day of the cycle with payroll, HR, finance, IT and the integrator.

2

Rank defects by impact: anything affecting net pay, taxes, retirement or a bank file is critical.

3

Assign each defect a named owner and a fix date that lands before the next cycle’s cutoff.

4

Re-run the affected employees in the test environment as soon as a fix is in, then confirm the fix again in the next full cycle.

5

Watch for regressions: a fix to one bargaining unit’s overtime can break another’s.

6

Report open counts by category and severity to the project sponsor at the end of every cycle.

If the same defect comes back twice, stop patching records and look for the root cause in converted data or design. The London Fire Brigade report traced its persistent variances to the legacy database rather than the new configuration, and reset its approach around that finding.

Exit criteria and sign-off for the go/no-go decision

The parallel test passes when the final cycle meets written exit criteria agreed before the first cycle started. Writing them in advance keeps a deadline from redefining “good enough.”

Typical exit criteria, in our experience:

The final cycle (or final two cycles) completed with zero open critical or high defects.

Net pay matches for every employee, or every difference is categorized as expected or explainable and approved.

GL postings tie by fund, and the payroll clearing account nets to zero.

The bank file, positive pay file and retirement contribution file have been validated.

Leave balances match for every employee.

All “legacy was wrong” decisions are documented with an owner.

Staff who will run the first live payroll have run at least one parallel cycle themselves, start to finish, without integrator hands on the keyboard.

Sign-off should come from the payroll manager, HR director, finance director or CFO, IT lead and the project sponsor, in writing. The signed variance log and a one-page exit memo then go into the go/no-go meeting as evidence.

The stakes of skipping this gate are well documented. When Oregon’s state government moved payroll to a new system, about 4,500 employees, roughly 10% of the workforce, were underpaid or overpaid in January 2023, and more than 2,000 were improperly paid in each of the next two pay periods, according to Oregon Capital Chronicle reporting republished by Salem Reporter in June 2024. The Secretary of State’s statewide audit concluded that testing of the configuration was either “not sufficiently scoped” or not properly conducted.

For a wider view of how weak testing gates show up in failed projects, see why government ERP projects fail.

After the parallel: what carries into cutover

Once the parallel passes, the risk moves to cutover and the first live cycles.

Carry these items forward:

Freeze pay rule and configuration changes between the final parallel and the first live payroll, unless a change goes through the same retest.

Keep the legacy system available read-only so you can compare the first live payroll if needed.

Repeat the reconciliation checklist on the first two live payrolls, with the same owners.

Hold the integrator and your own team on site or on call through the first live pay date and the first month-end close.

KCT’s approach follows that pattern. In the City of Oxnard, KCT split go-lives one module at a time, Finance, then HCM, then EPL, with each cutover rehearsed (Oxnard case study).

Frequently asked questions

What is parallel testing in payroll?

Parallel testing in payroll is running the same pay period through the legacy payroll system and the new system using identical inputs, then comparing gross pay, deductions, taxes, employer contributions and net pay employee by employee. Employees are still paid from the legacy system, and every difference is explained or fixed before go-live.

How long does a parallel payroll test take?

A parallel payroll test usually spans several pay periods, because each cycle needs time to process, reconcile, fix defects and retest. Accomack County, Virginia, for example, described two weeks of parallel testing ahead of its September 2025 go-live. As a KCT rule of thumb, plan for three cycles plus contingency.

What is the difference between parallel testing and user acceptance testing?

User acceptance testing checks that each process works as designed using scripted scenarios. Parallel testing checks that the whole system produces the same results as your current system on real production data for a full period. UAT comes first; the parallel test is the final proof before cutover and the go/no-go decision.

What variance is acceptable in a payroll parallel run?

An acceptable variance in a payroll parallel run is one you can explain and have approved, such as a documented rounding change. As a KCT rule of thumb, net pay should match to the penny for every employee after explained items, and GL postings should tie by fund. Aggregate run totals alone are not enough.

Do you need a parallel test for utility billing?

A parallel test for utility billing is strongly advisable when you replace a customer information system, because rate calculations, late fees, payment plans and estimated reads all reach residents directly. Run at least one full billing cycle in both systems and compare bill amounts by account and totals by rate class before the first live bill run.

Who should sign off on a parallel test?

The parallel test should be signed off by the people accountable for the results: the payroll manager, HR director, finance director or CFO, IT lead and the project sponsor. Their written sign-off on the variance log and exit memo becomes part of the evidence for the go/no-go decision and the project’s audit trail.

Before your first parallel cycle

A parallel test eats staff hours, and nobody enjoys re-keying a pay period. Still, it is the last stretch of the project where a wrong overtime rule or a mis-mapped fund costs you a correction instead of a paycheck.

Write the exit criteria and get the backfill approved while the schedule still has slack, and leave room in the plan for a fourth cycle in case the third one fails.

If you want a partner to lead the parallel runs, variance triage and cutover on your side of the table, KCT’s ERP implementation team does that work and stays through the first fiscal close and payroll cycles. Book a conversation with KCT to talk through your test plan.

Ready when you are

Let's talk before it slips.