Legacy system modernization: when to modernize and how to plan it

Share in
19-11-2026

Legacy system modernization should start with the systems that block your business today, not with the oldest code you own. That ranking often stalls, because most legacy systems still run and nothing forces a budget.

Waiting is the default even at the largest scale. According to the U.S. Government Accountability Office (GAO), federal agencies typically report spending about 80 percent of their IT investment on operations and maintenance. By February 2025, agencies had finished only three of the 10 critical legacy modernizations GAO flagged in 2019.

The way out is to rank before you rebuild. You score each system on business value and technical health, modernize the blockers first, and give every system you keep a review date. The sections below build that method, from warning signs to approaches, scoring, and a phased roadmap.

Back to top

What is legacy system modernization?

Legacy system modernization means updating or replacing outdated software, infrastructure, and data platforms to meet current business, security, and integration needs. It preserves the business logic and data built up over years.

A legacy system is any business-critical system that has become hard to change, support, or secure. Typical examples in mid-size and large enterprises include:

  • Custom ERP on an aging technology stack: an in-house enterprise resource planning (ERP) system running finance or inventory on old frameworks.
  • Oracle Forms applications: data-entry screens built with Oracle's older form-based tool.
  • AS/400 and IBM i systems: IBM's midrange server platform, now called IBM i, often running RPG, an IBM business language.
  • VB6 and early .NET apps: Visual Basic 6 (VB6) and older Microsoft .NET Framework apps, near or past end of support, when security patches stop.
  • COBOL on a mainframe: core processing in COBOL (Common Business-Oriented Language) on IBM's large central computers, covered in our mainframe modernization guide.

Support status is a better test than age for any of these. Microsoft stopped supporting the VB6 development environment on April 8, 2008. It now states there is "no supported method to create or maintain Visual Basic 6 applications," although those applications still run.

Support can also end on a fairly recent release. IBM shipped IBM i 7.6 in April 2025, yet IBM i 7.4, released in 2019, left standard support on September 30, 2026. So even a young system can turn legacy once standard support ends or nobody on your team understands it.

Timeline showing that support status, not age, makes a system legacy: VB6 development environment support ended on April 8, 2008 while its apps still run, and IBM i 7.4, released in 2019, left standard support on September 30, 2026 after IBM i 7.6 shipped in April 2025, with IBM Service Extension running to September 30, 2029Back to top

How common are legacy systems?

Legacy systems are widespread in large organizations, and the most detailed public data comes from U.S. federal agencies, where even critical systems run for decades. The 11 federal systems GAO ranked in 2025 as most in need of modernization range from about 23 to 60 years old. Together they cost about $754 million a year to operate and maintain.

In the private sector, the same burden shows up as technical debt, the accumulated cost of shortcuts and outdated technology that someone must fix later. A 2020 McKinsey article cites a survey that puts a number on it. CIOs (chief information officers) reported that 10 to 20 percent of their new-product technology budget is diverted to resolving technical debt issues.

Analysts now frame that debt as a business problem. Gartner tells infrastructure and operations leaders to reposition technical debt "as a business risk and not a technology risk." It calls that debt "a direct barrier to strategic AI platform investment and deployment."

For your organization, the useful figure is your own split between run cost (keeping current systems running) and change budget (building anything new). The signals in the next section tell you when that split has tipped too far.

Server room photo with a split bar showing that U.S. federal agencies typically report spending about 80 percent of their IT investment on operations and maintenance, according to GAO, and that the 11 federal legacy systems most in need of modernization are about 23 to 60 years old and cost about $754 million a year to operate and maintainBack to top

Signs it's time to modernize a legacy system

Modernize a legacy system when it shows two or more of the signals below. They fall into three groups, plus a formula that prices the wait:

Technical signs

Three technical signals matter most:

  • Approaching end of support: after that date, regular security updates stop, and any extension is usually paid and time-limited. Windows Server 2016 and .NET Framework 4.6.2 both lose support on January 12, 2027.
  • Key-person risk: only one or two people understand the system. GAO notes two Treasury systems use COBOL, which has "a dwindling number of people available with the skills needed to support them."
  • Integration blocks: you cannot connect the system to SaaS (software as a service) tools without file transfers or re-keying (typing data in twice).

Business signs

Business signs show up in budgets and in daily work:

  • Small changes take weeks: a rule change needs weeks of analysis because nobody knows its side effects.
  • Run cost dominates: McKinsey says spending over half the IT project budget on integrations and legacy fixes likely means "paying interest only."
  • Workarounds spread: spreadsheets and re-keying fill the gaps, hiding labor cost and error risk.

Why legacy security won't scale in 2026

Legacy security won't scale in 2026 because software flaws are now the attackers' top way in, and unsupported systems stop getting fixes. Verizon's 2026 Data Breach Investigations Report, based on 2025 data, found 31% of breaches began with vulnerability exploitation, ahead of stolen credentials.

Older systems also struggle with zero trust, which NIST defines around protecting "users, assets, and resources" rather than network perimeters. Agencies told GAO that eight of the 11 most critical federal systems cannot implement zero trust without modernization. One agency cited missing MFA (multi-factor authentication).

Every API, a connection point other software calls, widens an older system's attack surface, the points an attacker can probe. Each one exposes whatever flaws sit behind it. GAO found known vulnerabilities in 7 of those 11 systems, so those flaws may already be on record.

Illustration of a legacy system past end of support, where a SaaS tool, a mobile app, and a partner portal each connect through a new API that leads straight to unpatched flaws, with Verizon's 2026 finding that 31% of breaches began with vulnerability exploitation, ahead of stolen credentials

Turn the signs into a cost of waiting

This formula gives your CFO (chief financial officer) a number:

Annual cost of waiting = Extra run cost + Workaround labor + Delayed-change cost + Expected incident cost (probability × impact)

You can estimate each variable with your finance team:

  • Extra run cost: licenses, extended support fees, and specialist contractors you pay only because the system is old.
  • Workaround labor: staff hours spent on spreadsheets and re-keying.
  • Delayed-change cost: revenue or savings lost while changes wait.
  • Expected incident cost: breach or outage likelihood multiplied by impact.
Back to top

Legacy system modernization approaches at a glance

Modernization approaches differ mainly in how much of the system changes. That sets their speed and risk.

The eight approaches covered here overlap with the "7 Rs," a common shorthand for modernization options. Our legacy application modernization guide covers each R in depth, so this section stays at the portfolio level.

The table below compares the eight approaches by scope, speed, risk, and fit, rating them against one another rather than against measured benchmarks:

Approach

What changes

Speed (relative)

Risk (relative)

Choose it when you

Encapsulate (API layer)

Nothing inside; you wrap the system in APIs

Fast

Low

Need to expose data to new channels quickly

Rehost

Infrastructure only; the system moves as-is

Fast

Low

Are leaving aging hardware or a data center

Replatform

Runtime (the layer that executes code) or database, with minimal code change

Medium

Medium

Want to cut license and operations cost

Refactor / re-architect

Code restructured, often with a new architecture

Slow

Medium to high

Own a system that changes often

Rebuild

Everything, written from scratch

Slowest

High

Depend on the system as a core differentiator

Replace (SaaS/package)

The system is swapped for a product

Medium

Medium

Run a commodity function such as HR or customer relationship management (CRM)

Retire

The system is removed

Fast

Low

Find little remaining business value

Retain

Nothing, for now

Not applicable

Rising over time

Have a stable, low-risk system

Encapsulation, rehosting, and replatforming leave business logic intact. That makes them fast and low-risk, but hard-to-change code stays in place.

Rebuild sits at the far end of the table, and McKinsey treats a new "greenfield stack" (systems built from scratch) as "a last resort." Keeping old systems has a limit too. By McKinsey's measure, their risk and cost start to outweigh their benefits once technical debt passes half of a division's technology asset value.

Retain is a valid choice when you can state why and set a date to revisit it. A system you retain with no review date risks becoming the next emergency.

Avoiding that outcome means deciding which systems get which approach, and in what order.

Back to top

How to decide which systems to modernize first

Modernize first the systems that score high on business value and low on technical health. Formal scoring has a public precedent. GAO ranked 69 federal legacy systems on 16 attributes, including age, vendor support, legacy programming languages, cybersecurity risk, and operating costs.

Ranking your own portfolio takes three steps:

Score each system on business value and technical health

In the worksheet below, you score seven criteria from 1 to 5, grouped into two axes:

Axis

Criterion

Score 1 to 5

Business value

Revenue or operations that depend on it

 

Business value

Number of users and processes it touches

 

Business value

Strategic importance over the next 3 years

 

Technical health

Vendor support status (5 = fully supported)

 

Technical health

Ease of change (5 = easy)

 

Technical health

Integration readiness (5 = APIs available)

 

Technical health

Security and compliance fit

 

Treat business value of 10 or more out of 15 as high, and technical health of 10 or less out of 20 as low. Score with both the business owner and an engineer in the room, because each side may overrate its own axis.

These thresholds are a starting point. Adjust them once you see how your scores cluster, but keep them fixed during a ranking round so every system faces the same bar.

Map systems with the Gartner TIME model

The Gartner TIME model (Tolerate, Invest, Migrate, Eliminate) is a portfolio framework that places each system on two axes. SAP LeanIX describes it as a matrix of technical fit, how sound the technology is, against functional fit, how well the system serves the business. Those two axes match your technical health and business value scores, so the worksheet feeds the matrix directly:

 

Low technical health

High technical health

High business value

Migrate (modernize first)

Invest (extend)

Low business value

Eliminate (retire)

Tolerate (retain and monitor)

In TIME, "Migrate" means moving the business capability to a better technical base, so rehosting, replatforming, refactoring, rebuilding, or replacing can all apply. Because these systems are both valuable and fragile, your modernization budget goes to the Migrate quadrant first. Eliminate systems free up budget, and Tolerate systems get a review date.

Rerun the matrix every budget cycle. A support deadline, a new regulation, or an acquisition can move a system from Invest to Migrate within a single year.

Factor in dependencies and timing

Before you lock the sequence, adjust it for two constraints:

  • Dependencies: when many systems call one legacy system, you put an API layer in front of it first. You can then modernize the system behind that layer without breaking the systems that call it.
  • Timing: you align each project with end-of-support dates and your budget cycle. IBM's Service Extension, its extended-support offering for IBM i 7.4, runs until September 30, 2029, which buys planning time rather than a long-term home.
Back to top

A phased legacy modernization roadmap

A phased legacy system modernization roadmap avoids betting everything on one cutover, the switch from the old system to the new one. It splits the work into four smaller decisions, each gated by its own exit test:

Phase 1: discover and document

Build an inventory of components, a dependency map, and notes from SME (subject-matter expert) interviews before you choose an approach. Record every integration and batch job (a scheduled, unattended task such as a nightly export), since a hidden one can surface as a cutover outage.

Martin Fowler, who writes about software architecture, explains why this step is harder than it looks. "Replacements seem easy to specify, but often it's hard to figure out the details of existing behavior," he writes.

Phase 2: pilot a low-risk, high-visibility slice

Pick a module with clear business value and few dependencies. Measure a baseline first, such as lead time for a change (days from request to release), defect rate, and monthly run cost.

Without that baseline, you cannot prove the pilot worked. A measured, visible win makes the case for the next phase.

Phase 3: modernize incrementally with the strangler fig pattern

The strangler fig pattern replaces a legacy system piece by piece while the old system keeps running. Fowler took the name from a vine that gradually replaces its host tree. In practice, you route one function at a time to the new component, then expand once it proves stable.

The pattern needs temporary code that lets the old and new systems coexist. Fowler argues that the lower risk and earlier value outweigh its cost.

Strangler fig pattern in three stages: a routing layer first sends every function to the legacy system, then routes functions one at a time to a new component while temporary code lets old and new coexist, and finally the legacy system is switched off after reconciliation passes and the rollback window closes

Phase 4: cut over and decommission

Run data reconciliation, an automated comparison of records between the old and new systems, before the cutover. Clean data alone does not prove readiness, as TSB, a UK bank, learned. Its data migrated successfully in April 2018, yet its new platform failed immediately.

Some TSB customers were affected until December 2018, and in 2022 regulators fined the bank £48.65 million, according to the Bank of England.

Keep a rollback window, the period when you can still revert. Switch off the old system only after reconciliation passes and that window closes. Write the rollback steps before cutover day, not during it.

IT team monitoring screens at night, with a gate showing the old system is switched off only after data reconciliation passes and the rollback window closes, and the TSB case: data migrated successfully in April 2018, the new platform failed immediately, some customers were affected until December 2018, and the bank was fined £48.65 million in 2022

Phase 4 is where a gap between "migration done" and "safe to switch off" hurts most. When you assess a migration partner, ask where testing sits in their process and who signs it off. The System Migration process at GMO-Z.com RUNSYSTEM, for example, makes data-integrity and compatibility testing its own phase, after migration and before deployment.

Back to top

Common legacy modernization challenges and how to handle them

Most modernization risks are predictable, starting with weak planning. Of the 11 federal systems GAO ranked most in need of modernization, only three had fully documented modernization plans.

Five risks recur across programs, and each has a control you can put in place before your team writes new code:

  • Missing documentation: your team runs code analysis and SME sessions before design. It then records each business rule as a testable statement.
  • Data migration quality: you profile your data early, meaning you scan it for gaps, duplicates, and format errors. Then you cleanse it and automate reconciliation, so the tooling checks every data load (each batch of records you move) in full.
  • Running two systems at once: you budget people and money for both tracks across the full overlap. The old system still needs patches and fixes until the day you switch it off.
  • Testing gaps: your team builds a regression suite, a set of automated tests confirming existing behavior, on the old system before changing it. That suite becomes your definition of correct.
  • Change resistance: you bring the heaviest users into the pilot so they help shape the new workflow and explain it to their peers. Their early feedback also catches gaps that specifications miss.

These controls cost time up front, but they spread the effort across the program instead of saving it for one big push. McKinsey warns that infrequent IT megaprojects pose "high execution risk." It recommends paying down technical debt "consistently, predictably, and over a strategic time horizon" instead.

The controls also reinforce one another. A regression suite gives reconciliation a reference for correct output, and SME sessions give the regression suite its test cases.

The five controls map to four points on the roadmap:

  • Phase 1: you run documentation work and data profiling.
  • Phase 2 pilot: you involve the heaviest users.
  • Before phase 3: your team builds the regression suite.
  • Business case: you fund the overlap budget, so the old system keeps getting patches mid-program.
Back to top

Four trends are changing how enterprises modernize legacy systems in 2026:

  • AI-assisted code comprehension and conversion: agentic AI, which plans and runs multi-step tasks, now reads old code and drafts documentation and tests. AWS describes AWS Transform for mainframe as covering work "from assessment and business rule extraction through requirement generation."
  • API-first, composable architecture: API-first means you expose each legacy function through an API, one at a time, and build new services on those APIs. Composable means built from independent parts you can replace separately, so you can retire the core later without breaking the systems that call it.
  • Data modernization first: AI projects can expose weak data foundations, which moves data cleanup earlier in the plan. An October 2026 Gartner playbook says AI implementation "can act as a technical debt stress test" that exposes weaknesses "across infrastructure platforms, integrations and data."
  • Security-driven modernization: with vulnerability exploitation now the top breach entry point in Verizon's data, support dates and zero trust readiness increasingly set the schedule.

None of these trends changes the ranking logic. AI speeds up discovery and test generation, but your team still has to validate its output against real system behavior.

Use these tools to move faster through phases 1 and 3, not to skip scoring. Ask any vendor pitching AI conversion how it proves the converted code behaves like the original.

Back to top

Legacy system modernization starts with what blocks the business

Modernization works best as a ranking exercise followed by phased delivery. Score each system on business value and technical health, map it on the TIME matrix, and modernize the Migrate quadrant first. Then deliver in gated phases, so no single cutover carries the whole program's risk.

Ranking alone does not get systems modernized. Federal agencies had finished only three of 10 critical modernizations nearly six years after GAO flagged them.

Not every system needs to change this year. Each system you keep needs a review date, because support dates and attack patterns keep moving even when your code does not.

Your first pass through the portfolio fits in one quarter:

  • List the systems: you name your 10 to 20 most important ones.
  • Score them: you rate each one with the worksheet.
  • Price the wait: you estimate the cost of waiting for each system in Migrate.

The result gives your CFO a ranked case instead of a wish list.

If you want a second opinion before you commit budget, talk to the GMO-Z.com RUNSYSTEM System Migration team. Its process starts with an analysis phase that covers system and data complexity, potential risks, and the migration options that best fit your business.

Back to top

Legacy system modernization FAQs

Short answers to four common questions follow:

What is an example of a legacy system?

Common legacy system examples include custom ERPs on aging technology stacks, Oracle Forms applications, AS/400 (IBM i) systems, VB6 applications, and COBOL mainframes. Each is still business-critical but has become hard to change, support, or secure. Microsoft, for instance, stopped supporting the VB6 development environment in April 2008.

How long does legacy system modernization take?

A full legacy system modernization program for core systems often takes several years, which is why phased delivery matters. GAO flagged 10 critical federal modernizations in 2019. Of the seven still unfinished in February 2025, agencies planned to complete two in five or more years.

What's the difference between legacy system and legacy application modernization?

Legacy system modernization covers the whole system, including infrastructure, data, integrations, and applications. Legacy application modernization focuses on the code and architecture of individual applications. System-level decisions come first, because they set which applications you rehost, refactor, replace, or retire.

Is it cheaper to maintain or modernize a legacy system?

Maintaining a legacy system is usually cheaper in the short term, so compare both options over three to five years. Weigh the annual cost of waiting against the modernization cost. Incident cost alone can tip that comparison, since IBM puts the 2026 global average breach cost at $4.99 million.

Back to top
20年の経験を持つGMO-Z.com RUNSYSTEMの専門家からテックコンサルティングを受けましょう
GMO-Z.com RUNSYSTEM