The phrase “near miss” sounds like jargon, but the idea behind it is something everyone already understands. You step off a kerb, a bicycle brushes past, and your heart jumps. Nobody was hurt, yet you will look both ways more carefully for a week. In software delivery and operations the same thing happens constantly, and teams that learn to notice it improve faster than teams that only react to damage. This article explains the concept in everyday words and shows how a DevOps team can put it to work.
The idea in one sentence
A near miss is something that might have done damage and, by luck or design, did not, which hands you the lesson minus the injury. That is all. The reason the idea deserves its own name is that human attention is lopsided: we study accidents carefully and forget close calls within hours, even though the two share causes.
Why the same cause appears in both
Imagine two engineers who each run a command in the wrong environment. One of them is at a terminal where the environment happens to have no live data, so nothing happens. The other is at a terminal where it does, and an outage follows. The action was identical; only the circumstance differed. If you would investigate the second event, you have equal reason to investigate the first, because the weakness it reveals, an environment that is hard to tell apart, is exactly the same.
This is the central insight of the whole subject. Harm depends partly on chance, but the conditions that allow harm do not. By studying events where chance was kind, you can repair the conditions before chance turns unkind.
What near misses look like in DevOps
Here are examples that engineers often recognise at once.
- A deployment that would have gone to production but was stopped because someone noticed a version number was off.
- A backup that was found to be empty during an unrelated check.
- A permission granted too widely, discovered and narrowed before anyone used it.
- A change made late on a Friday that was rolled back within minutes because a dashboard showed an odd pattern.
- A warning in a log that was ignored for weeks until someone finally asked what it meant.
None of these produced a customer-visible problem. Each tells you something specific about a missing check, an unclear name, or an alert nobody reads.
A simple three-question method
You do not need a complicated framework to learn from a near miss. Three questions are enough, and they can be asked in a five-minute conversation.
- What nearly happened? Describe the event in plain terms without blame.
- What prevented it? Was it a designed safeguard, a habit, or sheer luck?
- What would make prevention reliable? Choose the smallest change that would stop the problem without depending on luck.
The second question is the most revealing. If the answer is “a designed safeguard”, you have confirmation that a control works. If it is “a habit”, consider making the habit part of the tooling. If it is “luck”, you have found a place to act.
What “no blame” really means
People often misunderstand the blameless approach. It does not mean that nobody is responsible or that carelessness is excused. It means that the aim of the conversation is to improve the system, so questions focus on conditions and not on character. “Why was the environment name so easy to confuse?” leads somewhere useful. “Who was careless?” ends the conversation and discourages the next report.
The wider cultural side of this is covered in a practical overview of managing culture change in DevOps, which pairs naturally with the near-miss habit since the habit only works when people feel safe to speak.
How near misses relate to risk management
Risk management usually starts from imagination: what could go wrong? Near-miss learning starts from observation: what almost went wrong? The two complement each other. Imagination is broad but sometimes off target; observation is narrow but grounded. A team that combines them keeps its risk list honest by adding items it has actually brushed against. For a broader treatment of the subject, see the article on managing risk while shipping software dependably and securely.
Where to look for a related reading list
Engineers who keep online services running have written for years about studying close calls. For general readers, SRE School shows a blog-style home page of short articles on reliability themes. Reading a few pieces there shows how a neighbouring community words the same instinct, but the three questions above work without any of that background.
Three kinds of luck to tell apart
When the second question, “what prevented it?”, is answered with luck, it helps to be more precise, because not all luck is equal.
- Timing luck. The problem happened to occur when someone was watching, or when traffic was low.
- Knowledge luck. The right person happened to be nearby and recognised the pattern.
- Scale luck. The mistake touched a small test group and not the whole user base.
Naming the type points toward the remedy. Timing luck calls for an alert that works whether or not anyone is looking. Knowledge luck calls for writing down what the knowledgeable person knew. Scale luck calls for limiting the blast radius of every change by default.
Practical steps for a team this month
Start small and keep the cost low.
- Open a shared note called “Close calls” and tell the team it exists.
- Invite everyone to add one line whenever something nearly goes wrong.
- Once every two weeks, spend twenty minutes reading the lines and picking one fix.
- Announce the fix, and thank the person whose report led to it.
After two months, review the list for repeated themes. Teams often find that a small cluster of causes explains the bulk of the lines, for instance muddled naming or a missing check on one input field.
What to do with old events you remember
Teams often have a store of half-remembered scares from before the habit began. Ask a few veterans for the two or three they still think about, and add them to the list as historical entries. They tend to be vivid precisely because they were frightening, and they usually point at structural weaknesses that are still there. Treat them gently, since people may feel embarrassed, and focus on the conditions that allowed the event instead of on who was involved.
Common questions, briefly answered
Is this the same as an incident review? Not quite. An incident review responds to harm that has occurred. A near-miss review responds to harm that was avoided, which makes it lighter and more frequent.
Is special software required? Not at first. A shared document will do, and tools can be added later if the volume demands it.
What if people do not report? Make reporting easy, thank reporters publicly and never use entries to judge individual performance. Reports rise when people see them lead to visible improvement.
How detailed should an entry be? Three lines are enough: what nearly happened, what stopped it, and what would make the stop dependable.
A brief example in everyday words
Suppose a small team deploys an update every Thursday. One week, an engineer prepares the update and, while double-checking, notices that the package name in the release file is misspelled. The deployment tool would have accepted the typo and failed halfway, leaving the service in a half-updated state. The engineer corrects the name and the release proceeds normally.
In the near-miss review, the team asks the three questions. What nearly happened? A release with a bad package name. What prevented it? A habit of rereading the file, which not everyone shares. What would make prevention dependable? A validation step that checks the package name against the repository before the release starts. The step takes an hour to add. From then on, the typo cannot reach the half-update stage whoever is on duty.
Why the plain-language approach matters
Terms like “resilience engineering” and “hazard analysis” are useful in their place, but they can make newcomers feel the subject is not for them. The near-miss idea works because anyone can take part from day one. A new starter who notices that two folders have almost identical names has made a valuable observation, and a team that welcomes it grows stronger than one that waits for formal expertise.
The core advice fits in a line: speak up whenever something nearly goes wrong. Write it down, ask the three questions, and fix the smallest thing that makes the next close call less likely.









