Platform engineering has been a fashionable phrase for three years and a useful idea for ten. The pitch is the same in every deck: golden paths, internal developer platforms, paved roads. The push-back is the same in every CFO meeting: prove it.
Here’s our data from twenty-two platform-engineering engagements we’ve run since 2022, with permission to publish the medians.
What our clients actually gained.
- Median time from new-repo to first production deploy: 11 days → 1.5 days.
- Median time to onboard a new engineer to productive: 6 weeks → 9 days.
- Median P1 incidents per service per quarter: 1.8 → 0.4.
- Median cloud spend per service: −18% in year one, after the migration drag.
What you give up.
Velocity, for a quarter. Every team we’ve worked with sees a measurable dip in feature throughput in the first three months as engineers learn the new paved road. The numbers above are post-dip. If you don’t budget for that dip publicly, your platform team will be on the chopping block by month four.
What we recommend if you’re starting.
- Pick a target — ‘new service in a day’ is a good one — and ruthlessly optimise for it.
- Build the paved road for the third team, not the first. The first team will tolerate sharp edges; the third team is the early-majority.
- Publish a service health score that includes platform-adoption signals. Make the trade-off visible.
- Hire the platform engineering lead from outside if you can’t free your best staff engineer for it.