← All perspectives

Perspective / Software & CRM

Legacy modernisation as a sequence of reversible decisions

Legacy replacement risk is shaped by dependencies, data ownership and the order of migration. A phased architecture needs explicit coexistence rules, reconciliation evidence and an honest account of when rollback stops being straightforward.

10 min readReviewed

Restored brick building connected to a contemporary glass-and-steel extension.
Conceptual editorial image.

Modernising a business-critical system is a sequence of changes to responsibility, data and operating risk. Replacing the interface is often the easiest part. The difficult decisions concern which application can accept a transaction, where authoritative information lives, and how the organisation returns to a known state if a transition fails. A useful modernisation plan makes those decisions visible before delivery begins. This article uses an illustrative order-management migration to show how sequencing, evidence and reversibility can guide investment and cutover.

The executive view

  • Select migration slices that reduce a meaningful burden and test important assumptions.
  • Make data authority, reconciliation and coexistence responsibilities explicit at every stage.
  • Treat rollback limits and legacy retirement as decisions requiring evidence and ownership.
01

Assess the burden before selecting the replacement

Age alone is a poor investment criterion. A long-running system may remain economical and reliable; a recent one may already prevent the business from changing. Assess the burden through concrete constraints: unsupported dependencies, scarce operating knowledge, unreliable recovery, expensive changes and inability to satisfy current business needs.

The UK government’s revised legacy IT definition, updated in August 2026, centres on unacceptable burden or risk. Although its formal scope is public-sector assessment, that framing is useful for commercial investment discussions.

Translate each issue into a decision. An unsupported component may justify urgent containment; a slow workflow may need targeted redesign; an expensive reporting process may be addressed without replacing the transaction engine. The modernisation case should explain the burden being removed and the new operating obligations created by the proposed solution.

Create a burden register with observable evidence. Instead of “the platform is old”, record a missed recovery objective, an unsupported library, a contract-renewal constraint or the time required to make a specific pricing change. Distinguish a vulnerability exposure from a realised incident, and a proposed control from one already verified. This gives the board a way to compare containment, targeted remediation and replacement.

Estimate the cost of inaction over the same period used to assess modernisation. Include the support estate, specialist dependency and workarounds, but avoid assigning speculative catastrophe costs as though they were certain savings. The case becomes more credible when it explains which risks will remain after replacement. New software can remove a technical dependency while creating a new supplier, data or operational dependency.

02

Map dependencies that a demonstration will not reveal

Start with the system’s real responsibilities and callers. Include scheduled jobs, exports, spreadsheet workarounds, supplier interfaces, reporting queries and manual approvals. A dependency that runs once at month end can be more commercially important than a screen used every day.

Separate read dependencies from write dependencies. Determine which systems need historical information, which can create commitments, and which assume a transaction has completed when it has only been accepted for processing. Record ownership, frequency and failure consequences for each connection.

The resulting graph should support migration sequencing. A capability with few dependencies may be a useful first slice, but ease alone is not the objective. Select a slice that reduces a meaningful risk or proves a difficult architectural assumption. Document what it teaches the organisation about later transitions, including the limits of what a successful pilot actually establishes.

Observe actual behaviour across a complete business cycle. Combine code and configuration review with transaction traces, scheduled-task inventories and interviews with the people who repair failures. A month-end export used for finance reconciliation may be absent from the application's main interface and still determine whether retirement is possible.

For every dependency, record the business identifier, timing expectation and completion semantics. A batch file that is technically delivered may still require a manual approval before finance treats its contents as accepted. Replacing that connection with an API can change the control accidentally. Treat hidden operational rules as requirements to be assessed, not automatically as waste to eliminate. The first migration slice should test at least one important boundary without concentrating every unknown into the same cutover.

03

Worked example: extracting order status before order acceptance

Consider an illustrative distributor with a legacy application handling order acceptance, stock reservations and customer status enquiries. Replacing all three together would combine several independent risks. A staged plan could begin by extracting the read-only status experience while leaving transaction authority intact.

StageNew capabilityAuthority retained in legacy system
ObserveNew reporting and dependency inventoryAll order writes
ExposeCustomer-facing order-status serviceAcceptance and stock reservation
ValidateRehearsed replacement acceptance flowProduction order commitment
TransferNew acceptance service for an agreed cohortUnmigrated cohorts
RetireVerified removal of remaining dependenciesNone

The early status service tests identifiers, event freshness and exception handling. It does not prove that stock reservation or payment-related commitments can safely transfer. Treating the first slice as evidence for those separate decisions would create false confidence.

Define what the early read-only service is allowed to claim. If it consumes events that can be delayed, “latest known status” is a more accurate representation than guaranteed real-time fulfilment. Include an observation timestamp, a freshness limit and a customer-support path for contradictory information. This is an architectural requirement that changes the interface and the operating procedure.

A useful acceptance rehearsal deliberately drops a status event, delivers a duplicate and reverses the arrival order. The replacement should converge to the approved state without inventing another shipment or losing the exception. This rehearsal demonstrates the read model's recovery properties. It does not establish that the future order-write path can prevent overselling; that later authority transfer needs its own stock-reservation and commitment evidence.

04

Design coexistence as temporary architecture

A phased migration introduces components that may be needed only during transition: routing rules, translation layers, synchronisation processes and additional reconciliation. They have a cost, an owner and a removal condition.

The Strangler Fig pattern is a reference for progressively routing responsibility from an existing system to a replacement. Its usefulness depends on being able to intercept relevant interactions and operate both environments during the transition. It is not automatically appropriate for a small, easily replaced application or a system that must be retired immediately.

An anti-corruption layer can translate legacy concepts at a defined boundary rather than spreading them throughout the replacement. Keep its responsibilities narrow. If the layer quietly becomes the location of new business policy, the programme may create another poorly understood core instead of reducing dependency on the old one.

Budget for the transition layer as a product with a defined lifespan. It needs deployment, monitoring, incident ownership and tests while it routes business traffic. Give each temporary rule a removal condition, such as the verified migration of its final consumer. A target date alone is weak evidence because dependencies can remain after the calendar deadline passes.

Avoid dual writes without a recovery model. If an application updates the legacy database and then the replacement independently, a failure between the two operations creates disagreement. Decide how pending changes are recorded, retried and reconciled, and identify which system remains authoritative while repair occurs. Where a transactional event mechanism is feasible, evaluate it against the actual platform constraints rather than assuming a message broker alone makes the two changes atomic.

05

Specify data authority at every migration stage

For each domain, identify which system can write, how changes reach the other environment, and how conflicts are resolved. “Both systems stay synchronised” conceals the most important questions: which value wins, what happens during an outage, and how a missed update is recovered.

A rehearsal should test business invariants. For orders, these might include stable customer references, preserved financial totals and agreement between accepted orders and reservation outcomes. Counts alone are insufficient: two datasets can have equal row counts while containing different commercial commitments.

Record the evidence needed to transfer authority. Use representative historical records, current transactions and deliberately difficult exceptions. If the business cannot define what a correct result looks like, it is not ready to approve the cutover. The acceptance process must also consider the practical ability of staff to recognise and handle an inconsistent state.

Create a stage-by-stage authority table for every material entity. Orders, customer relationships, product references and historical invoices may transfer at different times. Record the permitted writer, routing rule, read source and reconciliation procedure in each stage. This avoids an ambiguous halfway state where two teams each believe their system owns the same commercial fact.

Check the difficult records before declaring a mapping complete: split orders, partial cancellations, changed tax treatment, discontinued product codes and customers whose legal identity changed. Reconciliation should preserve meaning as well as amounts. A transformed total can match while discounts or liabilities are assigned incorrectly. Keep source-to-target lineage and the approved transformation version so an operator can explain a disputed transaction after the migration team has moved on.

06

Define rollback by state, not by reassurance

Rollback is easiest before the replacement creates information the old system cannot represent. Once new transactions or rules diverge, returning traffic to the old application may require transformation, replay and manual reconciliation. A backup does not automatically reverse external commitments already made.

Create a cutover state model. For each stage, document the last safe reversal point, retained data, replay mechanism and decision authority. Include the conditions that pause migration, such as unresolved reconciliation differences or inability to process a defined exception.

Test the reversal procedure, including the time needed to understand the situation and obtain approval. Where reversal is no longer practical, define forward-recovery options instead. Executives should see this distinction plainly: a transition may be controlled without remaining cheaply reversible at every moment. Removing old tables, interfaces or specialist knowledge should therefore be a deliberate, evidenced decision.

Run a timed rehearsal with the people who will actually make the cutover decision. Include detection, diagnosis, escalation, traffic changes and reconciliation, rather than measuring only the script's execution. The business impact depends on the full interruption, including time spent deciding whether rollback is safe.

Distinguish technical reversal from commercial compensation. If the new system sends a supplier instruction or takes a payment, restoring an earlier database cannot undo that external action. The runbook must identify affected commitments and the forward correction required. Define a decision deadline before recovery options narrow further, and retain a transaction journal across the transition. A clean application dashboard is insufficient if the business cannot account for work that crossed the boundary during the incident.

07

Fund retirement and operational ownership explicitly

The commercial case should include the full coexistence period, additional support, reconciliation, staff preparation and eventual decommissioning. Otherwise, a phased approach can appear inexpensive while accumulating two permanent operating estates.

Track retirement evidence alongside new features. Examples include callers removed, reports redirected, historical access validated and recovery procedures accepted by the receiving team. Make those outcomes visible in programme governance so that a successful new interface does not disguise an unfinished transition.

A decision-ready modernisation pack contains a burden assessment, dependency graph, staged authority model, reconciliation criteria, cutover runbook and retirement plan. It should show both technical progress and the commercial exposure remaining at each stage. The objective is a system the organisation can operate and change with confidence, supported by evidence about what moved, what remains and who owns the next decision.

Test the financial effect of delay. In an illustrative estate, a replacement costs £14,000 per month to operate while the legacy estate still costs £18,000. Six additional months of coexistence add £192,000 of combined operating spend, before migration-team cost; only part of that total is avoidable by retiring sooner. Show the counterfactual clearly rather than presenting the entire figure as an available saving.

Make retirement a funded workstream with explicit acceptance evidence: no remaining callers, retained records accessible under the agreed policy, credentials revoked and billing obligations resolved. Archive the explanation of historical data and transformations. Removing production infrastructure while retaining undocumented exports can merely relocate the risk. The programme is complete when operational responsibility and historical accountability are sustainable, as well as when the new interface is live.

08

A staged modernisation decision for executives

At each stage, approve the next transfer of responsibility rather than the abstract ambition to become modern. The decision should state the burden reduced, the authority moved, the evidence obtained and the dependencies still retained. Separate progress measures such as features delivered from risk-reduction measures such as a verified recovery exercise or a retired unsupported dependency.

Use stage gates to resolve uncertainty. A discovery gate might require a credible dependency inventory; a coexistence gate might require reconciliation and failure handling; a cutover gate might require an exercised recovery plan. Thresholds depend on the business and should be agreed before evidence is collected.

This approach supports commercially useful early releases while preventing a successful demonstration from authorising unrelated high-consequence changes. It also makes stopping or reshaping the programme a deliberate option when evidence shows the proposed replacement would add more operating burden than it removes.

Further questions

Questions this raises

Does phased modernisation always reduce risk?

It can separate decisions and contain exposure, but it adds temporary routing, synchronisation and support costs. Its suitability depends on system boundaries, available integration points and whether both environments can operate during transition.

Is a database backup enough for rollback?

No. External transactions, changed business rules and information created by the replacement may require replay or reconciliation. A rollback plan must explain how the business returns to a coherent operating state, not simply restores files.

What evidence should authorise a cutover?

Agreed business invariants, representative migration rehearsals, resolved exception cases and operational readiness. Approval should identify the accepting owner, remaining risks, reversal limits and the process for recovering if the transition fails.

References

Sources & further reading

  1. Strangler Fig pattern 2 June 2026 (updated)
  2. Updated Definition of Legacy IT 17 August 2026 (updated)

Connected capabilities

Software & CRMStrategy & digital transformation

From thinking to doing

What could this mean
for your business?

Your next chapter

What could your
business do better?

Tell us where you want to go, what’s getting in the way, or what you need to deliver. We’ll start there.

Start a conversation

Start a conversation

Tell us what
you have in mind.

Interactive preview. Your information stays in this page and is not sent.