Publicité
ERP IMPLEMENTATION
🇫🇷 Lire en français →

ERP Parallel Run: Making Old and New Systems Coexist Before Cutover

ERP parallel run explained: when it pays off, which processes to double (payroll, close), how long, how to reconcile and exit. Decision guide and checklist.

ERP Parallel Run: Making Old and New Systems Coexist Before Cutover

Running two ERPs side by side is not insurance, it is a cost: you need to know what you are trying to prove. Too many steering committees order a “parallel run” out of reflex caution, without defining what will be compared, by whom, within what tolerance, and under what condition they stop. The result: teams keying everything twice for weeks, variances nobody can explain, and a cutover decided by attrition rather than evidence.

This guide is for CIOs, project managers and CFOs of mid-market companies. It does not repeat the cutover checklist from D-30 to D-3 or the legacy system decommissioning plan. It addresses the question that comes before both: should you run in parallel, what, for how long, and how do you get out of it?

Parallel Run: Definition and Difference from Big Bang, Phased and Pilot

A parallel run (or dual run) means operating the same business process in both the old and the new system for a defined period, then comparing the results. The old system remains the official system of record; the new one is “under observation”. At the end, you either switch the system of record or fix and repeat.

Do not confuse it with the other three cutover strategies:

  • Big bang: the whole scope moves to the new ERP on a given date. No coexistence, apart from read-only access to the old system.
  • Phased: you switch module by module, entity by entity or country by country. Each wave is a big bang, but the project is sliced.
  • Pilot: one real entity or site goes live on the new system while the others stay on the old one. This is real production, not comparison.
  • Parallel: the same transactions are processed twice, to prove the new system produces the same results.

The distinction matters: pilot and phased approaches expose production to a limited risk; a parallel run does not expose it, but costs human time.

What “Parallel” Means by Domain

The word covers very different realities:

  • Payroll: you calculate the same payslips in both systems from the same inputs, then compare employee by employee and pay element by pay element. The old system stays official until the comparison is validated. Vendors and integrators describe the same practice: on SAP payroll projects, Rizing notes that contracts typically provide for two parallel payroll cycles.
  • Accounting: you replay a monthly close (entries, reconciliations, trial balance) in the new system and compare the general ledger balance, the subledger balances and the VAT returns.
  • Invoicing: you generate invoices in the new tool as dry runs (not issued) and compare them with those actually sent.
  • Inventory and logistics: a parallel run is generally impossible as is (see below). Replace it with cutover stock counts and consistency checks.

When a Parallel Run Is Justified (and When It Is a Trap)

Decision Criteria

Four criteria are enough to decide, process by process:

  1. Criticality: does an error have an immediate, irreversible effect on a third party (an employee paid wrongly, an incorrect tax filing, a customer billed incorrectly)?
  2. Reversibility: can you correct afterwards without damage? A payroll corrected the following month is still visible to employees; a journal entry corrected before close is not.
  3. Volume and rule complexity: the more rules there are (collective agreements, multiple VAT rates, consolidation), the more value the comparison adds, because unit tests do not cover every combination.
  4. Legal obligations: payroll, social security filings, VAT, period close. The statutory calendar does not move because the project slips.

Simple rule: a parallel run is justified when the calculation is complex, verifiable by comparison, and an error is expensive to correct. It becomes a trap when the process is a real-time flow, when nobody can say what a “correct” result looks like, or when the team lacks the capacity to process the same volume twice.

Table: Big Bang, Phased, Pilot or Parallel

CriterionBig bangPhasedPilotParallel (on a scope)
Production riskHigh, concentratedMedium, spread outLimited to one entityLow (the old system is authoritative)
Team workloadPeak at cutoverSpread out, longMedium, with dual supportHeavy and continuous during the period
Coexistence durationNoneLongLongShort and bounded (1 to 3 cycles)
Direct costLowTemporary interfacesTemporary interfacesExtra staff and double-keying hours
What you proveNothing before cutoverEach wave independentlyThe system under real conditionsAccuracy of the calculations
Typical caseSMB, simple scopeMulti-entity mid-marketMulti-country groupPayroll, close, VAT

These strategies are not mutually exclusive. The most common setup in mid-market companies: a phased or big bang approach for operational flows, combined with a targeted parallel run on payroll and close.

Scope: What to Double, What Never to Double

Double first

  • Payroll, over one or two full cycles, including a month with special events (bonuses, leavers, long absences).
  • The monthly accounting close, with bank reconciliations, fixed assets, accruals and the VAT return. Our monthly close checklist (in French) provides the outline of the controls to compare.
  • Statutory reporting calculations (tax returns, regulatory statements) if the business rules have been redeveloped.

Never double

  • Customer and supplier flows (EDI, e-invoices, orders): sending the same document twice to a third party creates duplicates, disputes and confusion. Test these flows in a test environment with the partner, not in dual production.
  • Real-time inventory: two systems decrementing the same physical stock diverge at the first exception (breakage, return, transfer). Freeze, count, switch.
  • Payments and direct debits: only one system issues payment orders. The second produces comparison files only, never transmitted to the bank.
  • Processes with external side effects: anything that triggers an action at a third party (document dispatch, API call, statutory notification).

The principle: you double the calculation, never the effect. The new system calculates, compares and stays silent.

Reconciliation Method

Without a method, a parallel run produces dozens of variances with no diagnosis. A proven method has four elements.

1. Frozen test sets. Before kickoff, fix the test populations: all employees or a representative sample of each special case (part-time, executives on fixed-day contracts, expatriates); for accounting, all companies or the most complex ones. Rizing recommends starting small and early, validating the extraction of cumulative balances and the comparison tools during unit testing, not at the first parallel cycle.

2. Variance tolerances defined in advance. Decide in writing what an acceptable variance is: a rounding difference of a few cents with no impact, or zero variance required on net pay and VAT. Without an explicit threshold, every variance becomes a debate.

3. Daily or per-cycle reconciliation reports. For each domain, one standard report: number of objects compared, number of variances, cumulative variance value, cause. The same format every day, produced by the same person, makes the trend visible.

4. Root-cause classification. Each variance gets a cause from a small number of categories: wrongly migrated data, configuration, poorly specified business rule, interface, expected variance (the old system had a defect). Only classification tells you whether you are converging.

Variance Indicators to Track

  • Number of open variances, by domain and by day.
  • Value of those variances (in currency, and as a share of the total processed).
  • Dominant cause, to steer corrections.
  • Average resolution time and number of variances that reappear after correction.

A converging parallel run looks like this: the number of variances falls from one cycle to the next, and the remaining causes are rare and understood.

There is no universal duration; these are orders of magnitude from practice, to be adapted to complexity:

DomainTypical durationNote
Payroll1 to 2 full monthly cyclesPlan a catch-up cycle if the first reveals fundamental errors
Accounting close1 to 2 monthly closesIdeally one with VAT and one with accruals
Invoicing (dry run)A few days to one billing cycleMainly for discount and tax rules
Consolidation1 quarterly close if the scopes require itCostly, reserve for groups
Inventory, EDINo parallel runCutover count and partner testing

The Hidden Cost: Double Keying, Fatigue, Extra Staff

The visible part of the budget is the infrastructure and licences of the old system, which you keep anyway. The real cost is human:

  • Double keying: payroll officers or accountants process everything twice, on top of their regular workload. If the team has no slack, the parallel run degrades the quality of the official system, the opposite of the goal.
  • Fatigue: the parallel run happens at the busiest times (payroll, close). By the second cycle attention drops, and variances are closed too fast or labelled “not important”.
  • Scarce skills: the person who can explain a payroll variance in the old system is often the same one who validates the new one.

So budget explicit extra capacity (contract accountants or payroll specialists, an integrator consultant, backfill to free the experts) rather than assuming the team will “absorb it”. This is what the vendor fails to cost when proposing “two parallel cycles” in its offer. And specify in the contract who, client or integrator, produces the comparison reports: ambiguity over responsibility is one of the pitfalls identified by Rizing.

Responsibility Matrix

ActivityCFO / HR DirectorProject managerKey usersIntegratorCIO
Define scope and tolerancesARCCI
Produce test dataIARCC
Run both processesIIRCI
Produce the comparison reportCARRC
Analyse and classify variancesCARRI
Fix configuration and dataIACRR
Go/no-go decisionARCCC

(R: responsible, A: approves, C: consulted, I: informed.)

Exit Criteria (Go/No-Go) and Rollback Plan

The parallel run must end with a decision, not with weariness. Set the criteria before starting:

  • Zero variance (or below the agreed threshold) on amounts payable, VAT amounts and cash balances.
  • All remaining variances are explained and classified, none “unknown”.
  • Trend: at least two consecutive readings with no new variance cause.
  • Key users have signed off on their scope.
  • Resources for the cutover period are available (the parallel run must not have exhausted the team before go-live).

If the criteria are not met, there are three options: extend by one cycle (once, no more), reduce the cutover scope, or postpone. An accepted postponement costs less than an erroneous payroll.

Rollback plan. As long as the old system is authoritative, going back is trivial: you abandon the new one. The critical point is the moment the system of record switches. From then on you need documented reversibility: a frozen backup of the old system, maintained access, and a clear rule on the deadline beyond which rollback is no longer possible (for example after the first real payroll is paid or the first VAT return is filed). This window must appear in the cutover plan.

How It Fits with UAT and Hypercare

A parallel run is not acceptance testing. UAT verifies that the system does what was specified, on constructed scenarios. The parallel run verifies that it gives the same results as the old one on real data. One does not replace the other: a parallel run launched before UAT is finished spends its time discovering configuration defects.

After cutover, the hypercare phase takes over: the first real payroll or close cycle stays under heightened monitoring, with the parallel run’s variance reports as the baseline.

Typical Case: Payroll and Monthly Close Over One to Two Cycles

A manufacturing mid-market company with several hundred employees migrates its accounting and payroll to a new ERP. It chooses a big bang for purchasing, sales and inventory, and a parallel run only for payroll and close.

  1. M-3 to M-2: frozen test sets, comparison tool validated on a sample of employees, tolerances written down.
  2. M-1 (first cycle): payroll is calculated in both systems. The old one actually pays. Variances are classified; most come from migrated data (cumulative balances, leave) and a badly configured bonus rule.
  3. M (second cycle): after correction, a new calculation. Residual variances explained. The accounting close is replayed in parallel, with trial balance reconciliation.
  4. Go/no-go committee: criteria met, sign-off by the HR Director and the CFO, system of record switched at the next cycle.
  5. M+1: first real payroll on the new ERP, heightened hypercare.

What makes this case work: a narrow scope, written criteria and extra capacity budgeted for two cycles.

Five Common Mistakes and Countermeasures

  1. Starting the parallel run too early, before acceptance testing is finished. Countermeasure: an explicit entry condition (UAT validated, migrated data of good quality).
  2. Doubling everything “to be safe”. Countermeasure: apply the four decision criteria process by process and refuse a parallel run on flows with external effects.
  3. No variance threshold and no exit criteria. Countermeasure: tolerances and go/no-go criteria signed before day one.
  4. Exhausting the team. Countermeasure: budgeted extra staff, parallel periods outside peaks, no more than two cycles without a decision.
  5. Comparing without explaining. Countermeasure: every variance has a classified cause; an “unknown” variance blocks exit.

Checklist to Copy

  • Candidate processes assessed on the four criteria (criticality, reversibility, volume, legal)
  • Doubled scope written down; excluded processes listed (EDI, inventory, payments)
  • Entry condition: UAT validated, migrated data checked
  • Test sets and comparison tool validated before the first cycle
  • Variance tolerances written and signed
  • Extra capacity budgeted and owner of the comparison report named
  • Indicators tracked: number, value, cause, resolution time
  • Go/no-go criteria and maximum duration fixed in advance
  • Rollback plan with a reversibility deadline
  • Variance reports handed over to the hypercare team

Going Further

Download our ERP checklist and evaluation scorecards to frame your cutover, and read alongside it the cutover checklist from D-30 to D-3 (in French) and the UAT guide.