How often will a deploy break?

Ship often enough and a non-zero change-failure rateA DORA metric: the share of deployments to production that cause a failure needing remediation (a rollback, hotfix or patch). Elite teams sit at 0โ€“15%, low performers at 40โ€“60%. makes a bad day a statistical certainty, not bad luck. Every deploy is a coin flip; the count that fail in a window follows the binomial distributionThe distribution of how many of N independent trials succeed (here: fail), each with the same probability p. Its mean is Nยทp.. Feed in your cadence and read off the odds.

The cadence
We shipdeploys per, and% of changes fail.
Over a window ofweeks, that's40 deploysโ€” about 6 expected to fail.
Each failed deploy takes abouth to recover.
6expected to fail
0.15%clean windowThe chance the whole window passes with no failed deploy at all: (1 โˆ’ rate) raised to the number of deploys. It shrinks fast as you ship more.
12 hrecovery timeExpected failures ร— your recovery time โ€” roughly how long you'll spend firefighting failed deploys over the window.
The odds
The chance of at leastfailed deploy in the window
99.8%P(โ‰ฅ 1)

Over the window, the chance of at least 1 failed deploy is 99.8%.

17.4%0 failed deploys: 0.15%1 failed deploy: 1.1%2 failed deploys: 3.6%3 failed deploys: 8.2%4 failed deploys: 13.3%5 failed deploys: 16.9%6 failed deploys: 17.4%7 failed deploys: 14.9%8 failed deploys: 10.9%9 failed deploys: 6.8%10 failed deploys: 3.7%11 failed deploys: 1.8%12 failed deploys: 0.77%13 failed deploys: 0.29%14 failed deploys: 0.1%15 failed deploys: 0.03%16 failed deploys: 0.0084%17 failed deploys: 0.0021%
024810121416ฮป = 6

number of failed deploys in the window โ†’

exactly 1: 1.1% ยท at most 1: 1.2% ยท at least 1: 99.8%

At this cadence and rate, that's โ‰ˆ 78 failed deploys a year. Shipping more often raises the count but shrinks each change โ€” the lever that actually moves the firefighting total is the failure rate and how fast you recover, not slowing down.

DORA change-failure rate, run through the exact binomial distribution