Your CI bill is a teachable moment.

Most engineering organisations pay 2 to 5 times more for CI than they need to, because build optimisation is nobody's full-time job. GitHub Actions minute multipliers are quietly compounding on every repository. Nobody audits concurrency. Caching was set up once in 2019 and never revisited. Test suites run in full on every PR instead of sharded. External runners like Depot, Namespace, BuildJet, Blacksmith, and Ubicloud exist and are 40 to 90% cheaper on average, and most teams do not know they exist.

Adrian Cockcroft, formerly of Netflix and AWS, argued that cost should be a first-class architectural concern: the same way you instrument latency or error rate, you should instrument spend per unit of work. Corey Quinn of Last Week in AWS puts it this way: the cloud bill is a fossil record of every architectural decision. The same logic applies to CI. Your GitHub Actions bill is a fossil record of every workflow you have ever added, every matrix strategy nobody questioned, every macOS runner someone used because it was the default in a StackOverflow answer.

The FinOps Foundation framework defines unit-cost economics as the practice of understanding cost per meaningful unit of work, whether that is per deployment, per engineer, per feature shipped, or per test suite run. CI is the canonical unit-cost problem in engineering: it is invisible, it scales with team size, and it is optimisable. This site exists because the math is not complicated, but it is invisible, and most teams are paying for the darkness.

The seven-step optimisation stack below covers the full range from “30-minute fix” to “one-week migration.” Work through them in order. The first four are free. The last three involve a decision. In most cases, teams recover 40 to 90% of wasted spend within two weeks of starting. The consulting CTA at the bottom exists because sometimes you need someone to tell you which two anti-patterns are responsible for 80% of your bill, and do that work in two days rather than two weeks.