The challenge
An eleven-week delivery cycle, manual weekend releases, and a post-deployment rework rate above thirty percent.
The outcome
A move to weekly delivery, automated deployments during working hours, and a rework rate brought below eight percent in nine months.
The context
Releases happened on Saturdays, manually, following a twenty-two-step procedure held in a shared document. Each deployment tied up four people. Rollback was never tested.
Our approach
We did not start with tooling but with measurement: lead time, deployment frequency, change failure rate, time to restore. Those four indicators framed the whole programme.
- Progressive automation — each manual step was scripted and then wired into the chain, in order of how much human time it cost.
- Reproducible environments — infrastructure described as code, with test environments identical to production.
- Progressive deployment strategy — rollout in small batches, with automatic rollback above an error threshold.
- Skills transfer — two engineers from the in-house team co-built the chain with us and now own its evolution.
The outcome
Nine months in, the team ships weekly, during working hours, with a rework rate below eight percent. The twenty-two-step procedure no longer exists.