If you follow technology news on LinkedIn, you have seen the posts. “We migrated our legacy system in a week with AI.” “80% of migration automated with Copilot.” “From legacy to modern in days, not months.”
These claims are misleading. Not because AI does nothing — it does a lot — but because they confuse what AI speeds up with what a migration project actually involves. The result is unrealistic expectations that lead to blown budgets, missed deadlines, and frustrated stakeholders.
This article is the antidote. It is based on evidence from real migration projects, published benchmarks, developer community consensus, and our own experience modernising enterprise systems. No hype. Just honest numbers.
What the LinkedIn posts actually mean
When someone says “we migrated in a week,” they almost always mean one of three things:
They upgraded, not migrated.Moving from .NET 6 to .NET 8 is a version upgrade within the same framework. It was always a one-to-two-week job. That is not the same as moving from .NET Framework 4.x WebForms to .NET 9 with a React frontend — which is a fundamentally different system.
They measured “code compiles” not “system works in production.” Getting legacy code to compile on a new framework is roughly 5% of a migration project. The other 95% is testing, validation, user acceptance, data migration, and deployment. Those posts were written before the project was actually finished.
It was a small system. Migrating a 5,000-line internal tool is fast with AI. Migrating a 500,000-line system with 200 screens, dozens of integrations, and 50 users who each have different expectations is a completely different problem.
Where AI genuinely saves money
The savings are real. They just come from a different place than the marketing suggests. The primary saving is team size reduction, not timeline compression.
| Dimension | Traditional | AI-augmented | Saving |
|---|---|---|---|
| Team size | 10–12 people | 3–4 people | 70–75% |
| Duration | 12 months | 9–10 months | 15–25% |
| Total cost | £800K–£1.2M | £200K–£350K | 60–75% |
| QA effort | Baseline | ~40% automated | Partial |
| User acceptance | Baseline | Same | 0% |
One developer with AI tools produces code at roughly 2.5 to 3 times the rate of one developer without AI, for mechanical migration tasks. Two developers with AI produce the coding output of five or six traditional developers. That is genuine and measurable.
But coding is only 20 to 25% of a migration project. The rest — understanding the system, testing it, getting users to validate it, making architecture decisions, deploying it safely — all of that remains human-paced.
Why the timeline barely changes
This is the uncomfortable truth that no AI vendor will put on a slide deck. Here is where time actually goes in a migration project:
AI dramatically speeds up that 20% coding slice. But there is a hidden cost: reviewing AI outputis a new activity that did not exist in the traditional model. AI makes subtle mistakes that are harder to catch than human mistakes — like silently changing how null values are handled, or adding error handling that changes the system’s behaviour.
The honest maths:AI saves roughly 7 weeks of coding time on a 12-month project. But reviewing AI output adds 4 weeks back. Net saving on the critical path: 3 weeks. That is real — but it is 3 weeks out of 52, not 48 weeks out of 52.
The human bottlenecks that AI cannot compress
User acceptance testing
Your users have day jobs. Getting a two-hour testing session scheduled takes one to two weeks. They need at minimum three rounds of testing — initial review, fixes, and final sign-off. Each round has a fixed feedback cycle. If your users give feedback monthly, you get four cycles in four months. AI does not make calendars move faster.
Stakeholder decisions
A migration discovers 50 to 200 decision points. “Should we preserve this bug or fix it?” “Is this hidden feature in scope?” “Is this UX change acceptable?” Each requires human judgment, often from people who are busy with other priorities. Average decision cycle: three to five business days. AI does not attend meetings.
Architecture
AI can suggest architectures. It cannot take responsibility for them. Choosing between a strangler fig pattern and a big-bang rewrite, defining service boundaries, deciding the authentication migration path — these decisions have downstream consequences that compound over months. Getting them wrong early costs far more than the time AI saves on coding.
Where timeline CAN be shorter
AI enables meaningful timeline compression — but only when specific conditions are met:
| Condition | Timeline impact |
|---|---|
| Users available weekly for testing (not monthly) | Saves 6–8 weeks |
| Feature freeze on the legacy system during migration | Saves 4–6 weeks |
| Stakeholders available for quick decisions | Saves 4–6 weeks |
| Existing test suite with good coverage | Saves 2–3 weeks |
| Good documentation exists | Saves 2–3 weeks |
Best case with all conditions met: 12 months becomes 6 to 7 months. Typical case with some conditions: 9 to 10 months. Worst case with none: 12 to 14 months — AI review overhead can actually extend the timeline while still reducing cost.
The real timeline win: parallelism, not speed
The biggest acceleration from AI is not faster coding. It is the ability to run workstreams in parallel that a traditional team would run sequentially. With AI, a small team can have one developer migrating the backend while another rebuilds the frontend simultaneously, with automated QA agents running continuous validation.
Traditional sequential approach: 36 weeks. AI-augmented parallel approach: 24 weeks. That is a genuine 33% reduction — but it comes from parallelism, not from magic speed.
What this means for your business
If you are considering modernising a legacy system, here is what you should expect:
Cost saving: 60–75%.This is real, significant, and comes primarily from needing fewer people. A project that would have cost £960,000 with a traditional team of twelve can be delivered for £250,000 to £330,000 with a small AI-augmented team.
Timeline saving: 15–25%. Modest but real. Do not plan on more unless you can guarantee user availability, feature freeze, and fast stakeholder decisions.
Quality: equivalent or better.AI-assisted QA agents can compare every screen, every API response, every database state change between old and new systems without getting tired at screen 150. Human QA still decides what matters — but the mechanical comparison is more thorough.
The bottom line: If someone tells you AI can migrate your 500,000-line legacy system in a week, ask three questions. How many lines of code was the system? Did real users validate the result in production? How long did that validation take? The answers will tell you everything.
How to scope a legacy migration properly
The difference between a successful migration and a failed one is almost always in the preparation, not the execution. Before writing a single line of code, you need alignment on four fundamental questions:
Is this a platform upgrade, a modernisation, or a full rewrite?Each has a different cost profile, timeline, and team structure. Most legacy projects are modernisations — new frontend, preserved business logic — but many are pitched as simple upgrades, leading to scope explosions.
What exactly is changing and what is being preserved? Business logic, data, integrations, UI, and UX are different layers that can each be handled independently. The biggest mistake is treating all layers the same way.
Who will test it and when? If you cannot name the person who will sign off on the migration, and when they are available, your project has a timeline risk that no amount of AI can solve.
What happens if it goes wrong?Rollback strategy, feature flags, parallel running — these must be planned before go-live, not improvised during an incident.
These are not technical questions. They are business questions. And they require a technology leader who can translate between the engineering reality and the business expectation — which is exactly what a fractional CTO does.
Three questions to ask any migration vendor
Whether you are evaluating an agency, a consultancy, or an internal proposal, these three questions separate honest estimates from fantasy:
1. What percentage of your estimate is testing and user acceptance? If the answer is less than 30%, the estimate is unrealistic. Failed migrations consistently underspend on testing. Successful ones allocate 30 to 40% of total effort.
2. Have you budgeted for reviewing AI-generated code? If they are using AI tools but have not accounted for the review overhead, they will either deliver buggy code or blow the timeline. AI output review is a new cost that adds 10 to 15% to the project.
3. What is your rollback plan? If they do not have one, they have never done this before.