Big-bang vs. strangler-fig: choosing without betting the business

Every legacy migration eventually has to answer the same question: does the new system replace the old one all at once, or piece by piece? The answer shapes everything downstream about risk, timeline, and what happens if something goes wrong.

Big-bang: fast on paper, risky in practice

A big-bang cutover moves everything at once, on a scheduled date. It’s simpler to plan and easier to explain to stakeholders — there’s one date, one go/no-go decision, one thing to coordinate. The problem is that it also means the entire system’s risk is concentrated into one event. If something’s wrong, you find out with everything already moved, and the fastest fix is often rolling everything back, not just the broken part.

Strangler-fig: slower to finish, contained when it breaks

A strangler-fig migration routes traffic to the new system incrementally — one feature, one customer segment, or one data domain at a time — while the old system keeps handling everything not yet migrated. Failures are contained to whatever slice just moved, and the team building confidence in the new system does so gradually, with production feedback at each step instead of all of it at once.

When big-bang is actually the right call

Strangler-fig isn’t always better. For small, self-contained systems with limited blast radius, or ones where the old and new systems genuinely can’t run side by side, the coordination overhead of an incremental approach can cost more than it saves. The deciding factor is blast radius: if a bad cutover would be a minor incident, big-bang is fine. If it would be a business-threatening one, incremental is worth the extra time.

The actual decision framework

Ask what happens to the business if the migration fails on day one. If the honest answer is “we’d notice and fix it,” big-bang is a reasonable choice. If the honest answer involves the word “outage” attached to a system customers depend on, strangler-fig is the safer default — and the extra weeks it takes are cheap compared to what a failed big-bang cutover costs. This is exactly the kind of judgment call a migration risk audit is built to make explicit before you commit to either path — see the full service ladder for how that fits into an engagement.