The Pareto Principle

The problem it solves

Effort is usually spread across a list as though the items on it were roughly equivalent, because a list gives no indication that they aren't. Forty features, two hundred customers, a quarter's worth of tickets. Each one occupies the same amount of vertical space and none of them announces how much of the outcome it is carrying.

The outcomes themselves are almost never spread that way. Counted rather than listed, revenue tends to pile up behind a few accounts, crashes behind a few defects, and support load behind a handful of flows that nobody thinks about until somebody counts. The gap between the flat list and the lumpy reality is where a lot of effort goes to die.

The expensive version of the mistake isn't spreading effort evenly, though. It's concentrating it confidently on the wrong few, which is what happens when a team nominates its own heavy items from memory instead of measuring them.

The idea

The principle says that effects tend to be concentrated, so a small share of causes produces a large share of the result. Joseph Juran named it after Pareto in 1951 and spent the following decades saying he'd got the attribution wrong. His own phrase for it is the better one.

The vital few and the trivial many.

That phrasing describes what the tool does rather than who first drew the curve, and it keeps the two halves in view, which the numbers on their own don't.

Enlarge: The shape holds up far better than the ratio does.

Ten bars in a row, ranked by size from tallest to shortest. The first two are far taller than the rest and are marked by a bracket labelled most of the effect. A dashed line separates them from the remaining eight. The first group is labelled the few and the second the many, over a line reading that the split is rarely as tidy as eighty and twenty.

The shape holds up far better than the ratio does.Close

The shape holds up far better than the ratio does.

The eighty and the twenty are a memorable pair of numbers and they are not a finding. Real distributions run at seventy and thirty, ninety and five, or barely uneven at all, and the two numbers don't have to add up to a hundred, since they're measuring different things. What holds up is the shape, and the shape is worth checking each time rather than assumed.

The idea came out of an income distribution and transfers well to anything with a long tail of small contributors, which covers most things a team measures. It transfers badly wherever the items interact. Ranking twenty ingredients by how much they contribute to a recipe is a category error, and so is ranking the steps of a process that produces nothing until they have all run.

How to use it

Count something you currently rank by opinion. Tickets by root cause, revenue by account, incidents by service, hours by project. An afternoon of counting is the whole method and the disagreement between the count and the room is the finding.

Rank it, then draw the running total, because the ranking says which items are large and the running total says where the meaningful break is. Sometimes there isn't one, and a flat distribution is a real result worth recording rather than a failed exercise.

Where a break exists, work out what a change to the top group would be worth before doing anything about it, and against what else the same effort could buy. Concentration says where the mass sits and says nothing about whether the mass can be moved.

Then decide what happens to the tail deliberately. It carries real obligations and real customers, and the useful question is what the cheapest adequate handling of it looks like rather than whether it can be dropped.

Where it breaks down

Every concept here has one. It is the section most summaries leave out.

It ranks on one variable, and decisions need at least two. Features ranked by usage produce a different order from features ranked by revenue at risk, and the ones near the top of both lists are the only items the analysis has actually settled. A single ranking presented as a conclusion is the commonest misuse and the hardest to argue with, because the chart looks like evidence.

Lopsidedness by eye is very weak evidence. Most published claims that a distribution follows a power law don't survive being tested against their own data, and a bar chart that slopes steeply is a much lower standard than that. The practical version of this warning is that with enough slicing, a lopsided chart can be produced from almost any dataset.

The tail is where the obligations live. Rare and low-volume covers the regulatory work, the safety cases, the enterprise customer with an unusual integration and the failure mode that happens twice a year. Cutting the trivial many is the move the name invites and it's the one that generates the incident.

It describes and doesn't prescribe. A finding that four accounts carry most of the revenue supports serving those four harder, and it supports treating the concentration as the largest risk the business has. The data is equally comfortable with both, and whichever the team was already inclined towards is the one it will hear.

Applied to people it becomes a bad management theory. The step from most of the output comes from a few of the causes to most of the output comes from a few of the people is short, popular and unsupported, and it ignores that contribution is measured on whatever the organisation happens to count. The principle has nothing to say here and gets quoted here constantly.

In one line

Go and count which few carry the weight, because the ones you'd have guessed are the recent and annoying ones rather than the heavy ones.