Why the rewrite fails
A rewrite has to reach feature parity with a system that has absorbed a decade of edge cases, regulatory patches and undocumented business rules before it can deliver a single unit of value. For that entire period the business is paying two teams — one to keep the old system alive, one to build the new — and getting nothing new in return. Meanwhile the old system keeps changing, so the target moves as you chase it.
By the time the rewrite is ready, the requirements that justified it have shifted, the champions who approved it have moved on, and the appetite for a risky midnight cutover has evaporated. This is not a hypothetical failure mode; it is the default one.
Strangle, don't replace
The strangler-fig pattern is the antidote. Instead of replacing the system, you grow the new one around it, routing individual capabilities to new services one at a time while the old system keeps running everything you have not migrated yet.
- Put a routing layer in front of the legacy system so traffic can be redirected capability by capability.
- Carve off the highest-value, lowest-risk slice first and move only that.
- Run old and new in parallel, comparing outputs, until you trust the new path.
- Retire each piece of the legacy system only once its replacement is proven in production.
Value ships continuously
The decisive advantage is that value arrives throughout the journey, not at the end of it. Each migrated capability is live and delivering the moment it is ready. Risk is contained to one slice at a time, every step is reversible, and the business can pause, reprioritise or stop entirely at any point with real gains already banked. Momentum and trust compound instead of draining away.
Data is the hard part
The code is rarely the difficult bit — the data is. Legacy schemas encode assumptions and history that no one fully remembers, and moving to a modern data model without losing fidelity is where most modernization effort actually goes. We treat data migration as a first-class workstream with its own reconciliation, validation and rollback, running dual writes and continuous comparison until the new store is provably correct before anything depends on it alone.
Key Takeaways
- 1The big-bang rewrite pays nothing until the end and usually never gets there.
- 2Strangle the legacy system capability by capability behind a routing layer.
- 3Every step is live, reversible, and banks value — momentum compounds.
- 4Data migration is the real work; treat it as a first-class, reconciled workstream.