Every team says they want reliability. Then a revenue feature ships on Friday and the SRE's spreadsheet is overruled in a hallway. An error budget that loses arguments is decoration — and the fix isn't a better spreadsheet, it's making the budget cost something real.
Make the budget bite
- Write the number into the deploy pipeline. If the last 28 days of availability fall below target, deploys to that service need an explicit override with a name attached. Not blocked — named. Humans behave differently when their name is on the risk.
- Spend it like money. A marketing launch that takes the API down 2% is a purchase. Was the launch worth the spend? Nobody asks when the budget is invisible; everybody asks when the invoice is.
- Report burn rate, not uptime. "99.7%" means nothing at 9am. "We burned 40% of this month's budget in the last 3 days" starts conversations that end in fixes.
- Never let the budget fund heroics. If the only tool you have is someone working the weekend, your budget is being paid in burnout, and it will be collected.
The uncomfortable part
The first month after we enforced this, deploy frequency dropped 30%. The engineering director was furious. The second month, incidents dropped 60%, deploys recovered past baseline, and the pages at 3am stopped. Error budgets don't slow good teams — they expose the releases that were never safe.
All drills ship in the Ops Starter Kit ($29). Every kit in one download: Ops Mega Bundle ($49). Want it run for you, 24/7? Always-on ops desk.
Top comments (0)