Asked how common something is, almost nobody counts. They try to think of instances, notice how easily instances arrive, and answer from that.

The shortcut is not unreasonable. Common things generally are easier to recall than rare ones, so ease of recall carries real information about frequency. Tversky and Kahneman named the move in the early 1970s, and the point of naming it was that the correlation breaks in predictable ways.

What makes something available besides being common

Recency. What happened last month comes to mind faster than what happened three years ago, regardless of which is more typical.

Vividness. A dramatic case is retrieved more easily than a mundane one. One client who blew up is more available than thirty who were fine.

Emotional charge. Cases that carried fear, embarrassment or anger are retrieved preferentially. An error that humiliated you in front of a client is permanently available in a way that a larger error nobody saw is not.

Personal involvement. Your own cases outrank everybody else's, which is why individual experience produces such confident and such badly calibrated estimates.

Repetition. Something heard several times feels more common, even from a single source repeated. This is most of why a well-covered risk feels larger than a widespread one.

Where it operates in a business

Risk assessment. The dramatic failure gets guarded against and the quiet one does not. Most businesses are far better protected against the thing that once went spectacularly wrong than against slow margin erosion, which has never been vivid on any single day.

Client selection. One bad experience with a type of client can close off a category for years. The sample was one, it was memorable, and no amount of ordinary counter-evidence is as retrievable.

Estimating demand. If three people asked for something last week it feels like a trend. Three is a number, and the ease with which it comes to mind is doing the work rather than the number.

Hiring after a bad hire. The next process over-screens for whatever the last failure exhibited, which is fighting an available case rather than the actual distribution of risk.

What helps

Count something. Any actual count beats recall. How many clients like this have there been, and how did they go? The number is usually retrievable from records in ten minutes, and it routinely contradicts the impression.

Ask what you are not recalling. Available cases are loud and absent cases are silent, which is the same structure as survivorship bias. If a category comes to mind through one example, that is a fact about the example.

Be suspicious of recency in any judgment about a pattern. A quarter is not a trend, and it is the most available evidence you have.

Write things down as they happen. A log is unweighted by memory. This is the single most effective correction, because it replaces retrieval with a record, and it is the same fix that works for cognitive dissonance and confirmation bias — all three are failures of a memory that edits, and all three are defeated by not relying on it.