The situation

Monday review. The scrap report lists nine defect codes. Everyone has a favourite. The quality engineer blames solder paste. The supervisor blames the placement machine. The plant manager says “all of them, fix them all.” Three weeks later nothing is better, because the team spread its effort across nine problems.

A Pareto chart ends this meeting in about ten minutes. It shows which problems are big and which are small, using your own data.

What a Pareto chart is for

A Pareto chart is a bar chart with a line on top. The bars show how often each problem happened, sorted from tallest to shortest. The line shows the running total, in percent.

The idea behind it is simple. In many processes, a small number of causes create most of the loss. Joseph Juran called these the “vital few”. The rest he called the “trivial many”. You do not need an exact 80/20 split for the chart to be useful. The point is to see where the weight sits.

Terms used here:

  • Category: one type of problem, such as one defect code or one downtime reason.
  • Cumulative percent: the share of the total you reach after adding each category, from the biggest down.
  • Vital few: the first categories that together make up most of the loss.

A short history

Vilfredo Pareto was an Italian economist. In 1896 he published a study of how land and wealth were spread in Italy. He found that a small part of the population owned most of it. In 1941 Joseph M. Juran noticed the same uneven pattern in quality problems. He named it after Pareto and described the chart in his Quality Control Handbook (first edition, 1951). Some accounts of this history differ on dates and details, so treat the exact years as “as commonly reported”.

Where it applies, and where it does not

It works well for:

  • scrap and defect codes, in a daily or weekly review;
  • downtime reasons, to choose which machine stop to attack first;
  • the problem description in an 8D report or the current-state box of an A3;
  • showing before and after, to prove that a fix worked.

It works badly when:

  • Severity matters more than count. A rare safety defect must not wait behind a common cosmetic one.
  • You have very little data. With fewer than about 30 events, one extra defect changes every percentage.
  • All bars are the same height. A flat chart says the problem is the whole process, not a few categories.

How to do it

We follow one example. At Pomerania Electronics, a fictional plant, Mateusz (quality engineer) and Agnieszka (SMT line supervisor) review Line 1. In May the line built 40,000 inverter control boards and scrapped 1,640. That is 1,640 ÷ 40,000 = 4.1%.

  1. Collect the data. Count scrapped boards by defect code over a period long enough to be typical. Do not hand-pick the good weeks.

  2. Total each category. Mateusz gets: solder bridging 820, tombstoning 410, solder balls 164, misalignment 82, insufficient solder 66, other 98.

  3. Sort from largest to smallest. Put “Other” last, even if it is not the smallest. Check that “Other” is smaller than the biggest bar. If it is not, your coding is too vague.

  4. Work out percent and cumulative percent. Divide each count by 1,640.

    Defect Boards Share Cumulative
    Solder bridging 820 50.0% 50.0%
    Tombstoning 410 25.0% 75.0%
    Solder balls 164 10.0% 85.0%
    Misalignment 82 5.0% 90.0%
    Insufficient solder 66 4.0% 94.0%
    Other (4 codes) 98 6.0% 100.0%

  5. Draw it. Bars use the left axis (counts). The cumulative line uses the right axis (0 to 100%).

  6. Read the vital few. Draw a line at 80%. Here the first two bars give 75%, and the third takes you past 80%. Bridging and tombstoning alone are three quarters of the scrap.

Agnieszka and Mateusz now look at stencil aperture design and the reflow preheat profile. They stop discussing placement accuracy, which is only 5% of the loss.

After the fix, they make a second chart with new data. If the first two bars did not shrink, the fix did not work.

Three ways it goes wrong

  1. Counting when you should be costing. Six hundred missing cable clips at a few cents each can beat four cracked structural welds on the count chart. Add cost or severity, or at least mark safety items. One chart by count and one by cost is cheap insurance.
  2. A giant “Other” bar. When operators have no good code, they pick “Other”. If it is 35% of the total, the chart is hiding the answer. Fix the code list first.
  3. A chart with no second chart. The team draws a Pareto at the start, makes a change, and closes the project. Nobody draws the new one. Closed is not solved. The after-chart is the evidence.

Our recommendations

  • Use one time period and one unit (boards, not a mix of boards and panels).
  • Chart by cost as well as count.
  • Keep “Other” under about 10 to 15% of the total. This is common practice, not a rule.
  • Pareto shows where to look, not why. After you pick a vital few, use 5 Why or a fishbone on that one bar.
  • Always plan the after-chart before you start the fix.

Try it

The Pareto tool will let you paste a column from Excel and get the chart, the 80% line and a one-page export, all in your browser. It is coming soon: Pareto chart tool.

From the book

Pareto analysis supports Chapter 20 (Lean, OEE, KPIs) and the problem description in Chapter 10 (8D). See the book page.