Case studies

How to evaluate architecture and delivery decisions using evidence.

Talk to Keith

How to assess a case study

Start with the situation and the baseline. A percentage improvement means little without a definition, a time window, and a clear comparison.

An illustrative example

For a frequently failing release, compare the rate of failed deployments and recovery time before and after one narrow change. Track rollbacks and customer impact as well. This is a hypothetical evaluation plan.

Worked example: replacing one write path

Suppose a team wants to move customer preferences from a shared database to a separate service. Start with an internal cohort. Record every write in both stores, compare normalized values, and investigate discrepancies before changing the read path. This is a hypothetical example.

  • Keep an operator-controlled kill switch and document which store is authoritative at each stage.
  • If writes can continue during rollback, agree on how to reconcile changes made during the test.
  • Keep deletion or data-retention changes outside the rollout until their implications have been reviewed.

A CTO who stays through delivery

Talk to Keith