Parallel running buys real confidence, sometimes.

What parallel running actually proves.

Parallel running is the test where the same business event is processed in two systems, and the outputs are compared. The customer order is entered in both systems. The supplier invoice is recorded in both ledgers. The employee payroll is calculated in both engines. The outputs are compared on volume, value, and detail. Differences are investigated. The buyer learns what the new system does differently and whether the differences are correct or defects.

What parallel running proves is fitness for purpose under real operating volume. Mock cutovers and user acceptance tests use scripted scenarios. Parallel running uses live transactions, with the full diversity of edge cases that the business actually sees. The result is much stronger evidence that the Newco system will hold under production load. The discipline is to define what counts as proven and what counts as still at risk before the parallel period begins.

What parallel running does not prove is the integration to external systems. Parallel running typically tests the core processing system in isolation. The bank integration, the EDI integration, and the tax filings continue from the production system. The other integrations are tested separately. Buyers that conflate parallel running with full integration testing are taking on more risk than they realize. The integration platform pattern that handles this is in carve-out integration platform rebuild.

Three models of parallel.

Model one is full parallel. Both systems run for a defined window. Both systems receive the same transactions. The legal system of record is one of them, typically the seller's during the parallel period. The other system is run for comparison. After the window, the buyer cuts over and the comparison system becomes the system of record. Full parallel is expensive because both systems require staff and licenses. It buys the highest confidence.

Model two is shadow parallel. The Newco system processes a copy of the transactions but does not act on them. No invoices are sent. No payments are made. No filings are submitted. The shadow system is purely for comparison. Shadow parallel is cheaper than full parallel because it does not require operations staff for the shadow side. It buys moderate confidence because the shadow system is not actually operated.

Model three is sampled parallel. A subset of transactions is processed in both systems for comparison. The subset is large enough to be representative but small enough to be operable. Sampled parallel is the right model when full parallel is impractical and shadow parallel does not exercise the new system enough. Sampled parallel requires careful sampling design so that the subset covers the diversity of transaction types and edge cases.

When parallel running is worth the cost.

Parallel running is worth the cost in three patterns. Payroll cutovers, where a payroll error has immediate employee impact and is hard to unwind. Tax engine cutovers, where the calculations have regulatory consequences and the differences may not surface until quarter end. And revenue recognition cutovers, where the calculations feed the income statement and the audit requires comparison. In these three patterns, the cost of parallel running is small compared to the cost of a defect that surfaces in production.

Parallel running is rarely worth the cost in patterns where the comparison is hard to make or where the system has already been validated through other means. Order management cutovers, where the comparison requires every order to be entered twice. CRM cutovers, where the data quality in the seller's system is itself questionable. And reporting cutovers, where the outputs are easy to verify after the fact through normal data validation.

The TSA cost implications of parallel running are real. The seller may continue to charge for the legacy system during the parallel period at full TSA rates. Extension fees may apply. The Newco system has its own operating cost during the same window. The discipline is to model the full parallel cost before committing and to weigh it against the risk reduction the parallel period delivers. The benchmarks for this trade off are in TSA exit cost benchmarks.

The comparison discipline that makes it real.

A parallel run without a structured comparison is a waste of money. The comparison discipline starts with a defined set of comparison points. For payroll, the gross pay, the net pay, the tax withholding, and the benefit deductions for each employee. For tax, the tax base, the tax rate applied, and the tax amount for each transaction class. For revenue, the revenue recognized, the deferred revenue, and the contract liability for each contract.

Differences are categorized. Acceptable differences are those that result from intentional configuration changes between the two systems. For example, a rounding rule that the buyer chose to change. Defects are differences that result from configuration errors, data errors, or logic errors in the Newco system. Differences in the seller's system that the Newco system corrects are not defects but are documented as expected differences.

The exit criterion for the parallel period is a defined ratio of clean transactions over a defined window. For example, three consecutive payroll cycles with zero employee defects above a defined dollar threshold. The exit criterion is set before the parallel period begins, signed by the governance committee, and not adjusted during the period. The discipline of a pre committed exit criterion prevents the parallel period from extending indefinitely.

Exit the parallel and own the result.

The parallel period ends with a formal acceptance decision. The governance committee reviews the comparison results, the defect log, and the remediation status. The committee decides yes, no, or hold. Yes means cutover to Newco as the legal system of record on the agreed date. No means rerun the parallel after a defined remediation period. Hold means continue parallel for a defined extension while specific items are addressed.

After acceptance, the seller's system is decommissioned according to the data retention policy. The data is extracted, the access is revoked, and the seller's TSA obligations for that service end. The disciplined buyer plans the decommission carefully because it is the point of no return. After decommission, the comparison is no longer available, and any defects that surface in Newco must be resolved in Newco alone.

Hypercare continues for a defined period after parallel exit. The team that ran the parallel period stays available, the issue triage continues, and the response times are tighter than steady state. The hypercare exit criterion is similar to the parallel exit criterion but applied to single system operations. The discipline that drives a clean transition through parallel and into steady state is part of our TSA exit acceleration service.

More on TSA exit strategy.