THE IDEA
Most of what you see is nothing
Data arrives constantly. Sales are up this week. Website traffic dropped yesterday. A customer complained. An employee left. Each event feels meaningful. Each invites a story, a cause, an explanation. And most of the time, the event is noise - random variation that means nothing, signifies nothing, and will reverse itself next week without any intervention.
The challenge of signal vs noise is one of the oldest in statistics and one of the most practical in everyday decision-making. Signal is meaningful variation - a real pattern, a genuine change, a trend with a cause. Noise is random variation - the background hum of a complex system doing what complex systems do. The problem is that signal and noise look identical in the moment. You can’t tell them apart by staring at a single data point. You can only tell them apart by looking at enough data, over enough time, with enough discipline to resist the urge to explain every fluctuation.
The human brain is spectacularly bad at this. We’re pattern-recognition machines. We see patterns in randomness, stories in noise, and causation in coincidence. A stock goes up three days in a row and we call it a trend. Sales dip for a week and someone launches an investigation. A team has two bad sprints and the manager restructures. Most of these reactions are responses to noise - expensive, disruptive responses to things that would have corrected themselves.
IN PRACTICE
Reacting to randomness
A marketing team notices that email open rates dropped from 22% to 19% this month. Alarm spreads. The subject lines are rewritten, the send time is changed, the segmentation is reworked. Next month, the open rate bounces back to 23% - which it would have done anyway, because the original drop was within normal variation. The team attributes the recovery to their changes, reinforcing the false belief that the noise was signal and the intervention worked.
A football manager loses three matches in a row. The press calls for a new strategy. The board gets nervous. The manager changes the formation, drops two players, and brings in a sports psychologist. They win the next two games and everyone credits the changes. But three consecutive losses in a competitive league isn’t unusual - it’s within the expected range of variation. The reaction wasn’t to a problem. It was to randomness wearing the disguise of a crisis.
A parent notices their child had a bad week at school - low test score, an argument with a friend, a grumpy evening. The parent worries something is wrong. They schedule a meeting with the teacher, have a serious talk, consider whether the child needs more support. The following week is perfectly normal. Nothing was wrong. The bad week was noise in a complex little life. But the parent’s intervention - the worried conversations, the teacher meeting - sent a signal of its own: that normal variation is cause for alarm.
WORKING WITH THIS
Learning to wait
The most important skill is patience. When a data point changes, the first question should be: is this within the normal range of variation? If it is, do nothing. Watch. Wait for more data. The pattern will either persist (signal) or revert (noise). Reacting to noise is one of the most expensive habits in any organisation.
Use baselines and control charts. Know what normal variation looks like for your key metrics so you can distinguish genuine change from random movement. A metric that fluctuates between 18% and 24% every month hasn’t changed when it hits 19%. It’s changed when it consistently moves outside that range.
When you do spot what looks like signal, test it before acting. Look for corroborating evidence from other sources. Ask whether the pattern persists over multiple time periods. Check whether someone else sees the same thing. Signal tends to be consistent across multiple indicators. Noise tends to be isolated. If one metric moves and nothing else does, it’s probably noise.
THE INSIGHT
The line to remember
Most of what looks like a pattern is randomness, and most responses to randomness make things worse. The discipline of doing nothing in the face of noise is one of the hardest and most valuable skills in any system.
RECOGNITION
When this is in play
You’re reacting to noise when the same metric triggers action every month in a different direction. When last week’s crisis is this week’s non-issue without anything having changed. When every fluctuation has an explanation and every explanation leads to an intervention. When the team is exhausted from responding to variation that the system was always going to produce. When someone says “this month is different” and last month was also different, and the month before that.