Pareto Analysis Builder

What this tool does

CM8-241 builds a proper Pareto analysis from raw counts. You enter observations — a category, how many times it occurred, optionally what each occurrence costs — and the tool ranks the categories, draws the classic descending bars with the cumulative curve, splits the vital few from the trivial many at your chosen cutoff, and prints a report that shows its arithmetic in full.

It works for anything you can count in categories: scrap and defect causes, customer complaints, downtime reasons, warranty returns, invoice errors, support tickets, late deliveries. One file can hold several analyses side by side — name each one ("Scrap causes — March", "Complaints Q2") and the charts follow the largest one in the current view.

Everything runs inside this single file. There is no account, no upload and no network request of any kind, so your defect rates and cost figures never leave the computer you are using.

The Pareto principle, honestly

The observation behind the tool is that in most processes, a small number of causes account for a large share of the effect — popularly "80% of the problems come from 20% of the causes". That is an empirical tendency, not a law. Your split may be 70/30, or 90/10, or occasionally close to even. Nothing is wrong with your data when the numbers refuse to be 80 and 20; the point of the analysis is not the ratio but the concentration — the fact that effort aimed at the top two or three categories usually removes more of the problem than effort spread across all of them.

And when your data shows no concentration — a cumulative curve that climbs in a nearly straight diagonal — that is a finding too. It usually means the categories are cut wrong (too coarse, or overlapping), or the problem genuinely has many independent causes and a ranking exercise will not prioritise it for you.

How the analysis works

Four steps, all visible in the Ranked categories table so anyone can check them:

1. Categorise — every observation is assigned to exactly one category.
2. Count — entries with the same category name (within the same analysis) are added together, so you can log the same cause from different areas or weeks as separate rows.
3. Rank — categories are sorted descending by the chosen measure: raw occurrences, or total impact when unit impacts are entered and the setting is switched.
4. Cumulate — the running total of shares, top down, gives the cumulative curve.

A category belongs to the vital few if the cumulative share before it is still under the cutoff — in other words, the vital few is the smallest set of top categories that together reach the cutoff, including the category that crosses the line. Everything after that is the trivial many: "trivial" only in the sense that fixing any one of them moves the total very little.

Entering data

One row is one count of one category — not necessarily one incident. If the shift log says "porosity × 21" for the casting line this month, that is one row with a count of 21. If the same defect was also counted on a second line, enter it as a second row with its own source; the ranking adds them together and the vital-few detail table still remembers where each came from.

The analysis name keeps separate exercises separate. Never mix two different questions under one name — "Scrap causes" and "Complaint reasons" have different totals, and a percentage computed across both means nothing. When more than one analysis is present, the tiles, the Pareto chart, the curve and the tables all follow the largest one; the monthly chart counts every dated entry.

Count or impact — which ranking

Ranking by raw count answers "what happens most often?" Ranking by total impact — count × cost per unit — answers "what hurts most?" They are different questions and they can give different orders, and each misleads in its own way. A cosmetic blemish that occurs two hundred times a month but costs a minute of polishing will dominate a count ranking while barely mattering; a cracked housing that occurs four times but scraps a finished assembly each time will sit near the bottom by count and near the top by impact. If you only ever rank by count, you optimise for the frequent-but-cheap; if the unit impacts are guesses, an impact ranking dresses the guesswork up as analysis. The honest sequence is: rank by count first, put defensible unit impacts on at least the biggest categories, then switch the ranking and see whether the order changes. If it does, the impact order is usually the one to act on.

Note that a category with no unit impact contributes nothing to an impact ranking — it falls to the bottom regardless of how often it occurs. The impact-ranking tile tells you how many categories carry values; do not trust the impact order until the big categories are priced.

Category discipline

The analysis is only as good as the buckets. Three rules keep it honest:

• Mutually exclusive. Every observation must have exactly one home. If inspectors could file the same defect under "porosity" or "casting defect", the counts split and both categories rank lower than the real problem.
• Consistent spelling. The tool matches category names ignoring case and surrounding spaces, but "wrong dim" and "wrong dimension" are two categories. Agree on the names before counting.
• No “other” bigger than about 10%. An "other" bucket is fine for genuine one-offs. When it grows past a tenth of the total, it is hiding a real category — open it up, re-slice, and re-count. The same applies to any category so broad ("machine problems") that knowing it is on top tells you nothing about what to fix.

The cutoff is a decision aid, not dogma

The 80% line is a convention for drawing a boundary, nothing more. Its job is to force a decision about where focused effort stops and background management begins. If the top two categories already hold 75% of the total, stopping there is perfectly sensible even though the line says 80. If reaching 80% takes six of your eight categories, the concentration is weak and the cutoff is telling you so — lowering it to where the curve visibly flattens gives a more honest vital few. Change the cutoff in the settings and watch the tiles, the bar colours and the tables move together; the arithmetic is always the same, only the boundary moves.

Acting on the vital few

A Pareto chart tells you where to dig, never why. The Vital-few detail table is built as the hand-over to that digging: for each vital-few category it shows the count, the impact, the source or area where it occurs most, and the latest note — enough to brief an investigation. Take the top category into a root-cause method next: a Five Whys exercise for a single clear failure chain, or a fishbone (Ishikawa) diagram when several factors plausibly contribute — both exist as companion tools in this series. Resist the urge to launch actions on all the vital few at once; the top one or two, properly investigated and fixed, beat five started and none finished.

Re-running it monthly

A Pareto analysis is a snapshot, and its real power shows on the second and third run. Keep the same category names, start a new analysis name each period ("Scrap causes — April"), count the same way, and compare: after you act on the top cause, its bar should shrink and something else should take the top spot — that rotation is what improvement looks like. The occurrences-by-month chart tracks the top category through time for exactly this purpose. If the same category stays on top for three periods despite corrective action, the action treated a symptom, not the cause — back to the investigation.

FAQ

My split is 60/40, not 80/20. Is the analysis wrong? No. The exact ratio varies by process and by how you cut the categories. Concentration of any degree is usable information; the 80/20 figure is folklore about the typical case, not a pass mark.

How many categories should I have? Enough to be specific, few enough to count reliably — five to twelve is typical. Two categories make the ranking trivial; thirty make the counts thin and the chart unreadable.

Can I compare two analyses in one file? You can hold them side by side, but the charts show one at a time — the largest in the current view. Percentages only mean anything within one analysis, because each has its own total.

Should near-misses or minor cases count the same as serious ones? In a count ranking, yes — one occurrence is one occurrence. If severity matters, that is exactly what the unit impact field is for: give the serious category its real cost and rank by impact.

What period should the counts cover? Any period, as long as every category is counted over the same one. Mixing a month of one defect with a year of another is the fastest way to a confident wrong ranking.

Saving your work

Entries, settings and the report header are written to this browser's local storage as you type, and the toolbar shows the time of the last save. That storage belongs to one browser on one computer: another browser, a private window, a second machine or a clean-up tool that clears site data will not have it.

Treat Export .json as the real save — one file containing everything, which Import .json restores anywhere. Export CSV gives you the register for spreadsheet work. Reset asks twice, then erases everything this tool has stored. There is no undo.

Accuracy & disclaimer

The arithmetic here — share, cumulative share, count × unit impact — is simple and the tool does it faithfully. Everything that matters sits underneath it: whether every occurrence was counted, whether the categories overlap, whether the counts cover the same period, and whether the unit impacts are estimates or evidence. A Pareto analysis ranks what you counted; if the counting was biased, the ranking is your bias drawn as a chart.

This is an analysis and prioritisation aid. It does not identify causes, and it is not a substitute for investigating the categories it puts on top.