Pareto analysis
Rank categories by one measure and compare their cumulative share before choosing where to focus your investigation.
Use Pareto analysis when you need to prioritize a set of recorded problems or contributions. Decide what each category contains and what you are measuring before comparing them. High counts identify where to investigate; they do not establish causes or tell you which solution will work.
Category totals for one population, period and measure, with a consistent unit such as complaints, hours or GBP.
An editable chart with descending category values, each share and cumulative share, plus the scope and proposed follow-up.
On this page
How to use it
- Choose a question, period, population and one measure. State the source and unit, and account for missing or excluded records.
- Define mutually exclusive categories so each record contributes once. Aggregate records before entering one nonnegative total per category; use a consistent rule for Other.
- Enter Category | Value on each line. The chart sorts values from largest to smallest, then calculates value divided by total and the cumulative share. Equal values keep their input order.
- Read the actual concentration. In this example, two of four categories account for 75%, while three account for 90%. Crossing 80% is not an automatic action rule.
- Inspect the underlying records, assess severity and feasibility, and choose an investigation. Compare later results under the same definitions.
Keep its limits in mind
The 80/20 pattern is a heuristic, not a required result. A low-frequency event may still have severe consequences. The worksheet calculates from the totals you enter; it does not classify records, detect overlapping categories or verify causes. Do not mix counts, money or currencies.
Investigate the main complaint categories
Fictional example. Qinghe recorded 100 valid complaints in one month. One operations staff member can follow up. Each complaint is assigned once to its main reported issue; the category label is not proof of a root cause.
On a small screen, scroll across the diagram to read it.
Measured concentration: Stockouts and late delivery total 75 complaints out of 100. These are two of four categories, so this example is not an 80/20 split.
A starting point for investigation: Packaging brings the cumulative share to 90%. That does not automatically make it a higher priority than a rarer, more serious complaint.
Next step: Check the selected complaint records for supporting evidence before choosing an improvement, then measure again using the same scope.
Three reasons to postpone the ranking
Use Pareto analysis to see how recorded contributions are concentrated under one measure. Choose counts, effort or money and keep that unit throughout. Before sorting, define the period, valid records and category assignment. The chart can identify a useful place to investigate without establishing a root cause.
Pause if one record can appear in several categories but the total is presented as unique complaints. Pause if some bars measure money while others count incidents. Pause if missing or unclassified records are quietly excluded to produce a cleaner pattern. Explain exclusions and reconcile the total first. “Other” is not automatically an invalid or unknown record.
Keep severity separate from frequency. A low-count issue may need urgent attention, while a large category may contain several unrelated explanations. The familiar 80/20 phrase is not a target that the dataset must match. Do not change boundaries or combine categories just to make the percentage look familiar.
Choose concentration, explanation or a redesign
Select the next method by what the ranking leaves unresolved. A category label tells you what was reported, not what caused it or which change would help.
| Method | Start with | Output |
|---|---|---|
| Pareto analysis | Comparable category totals with reconciled scope | Rank and cumulative shares |
| Logic tree | Several possible investigation paths | Branches, definitions and evidence gaps |
| 5 Whys | One bounded incident branch | Supported links and competing hypotheses |
| ECRS | A verified current process and required controls | Redesign candidates to test |
Worked continuation: 75% is the result, not 80/20
Return to the fictional Qinghe month: 100 valid complaints, each assigned once to its main reported issue. For traceability, label the existing groups C001–C045, C046–C075, C076–C090 and C091–C100. These identifiers only label the supplied category totals; they add no individual complaint narratives or verified causes.
The category shares are 45/100, 30/100, 15/100 and 10/100. Cumulative shares are 45%, 75%, 90% and 100%. The first two categories are two of four, or half the categories, and account for 75% of records. This example therefore does not show 20% of categories contributing 80%. Packaging takes the cumulative total to 90%, but that arithmetic does not authorize ignoring the last ten complaints.
The ten Other records are valid complaints already assigned to that category; their detailed issues are not supplied. Keep them visible and inspect severity. If further records cannot be classified, record that separately and explain whether the analysis total changes. Do not relabel the existing ten as missing evidence or claim that a stockout complaint proves an actual stock or system failure.
| Identifiers | Main reported issue | Count | Cumulative share |
|---|---|---|---|
| C001–C045 | Stockouts | 45 | 45% |
| C046–C075 | Late delivery | 30 | 75% |
| C076–C090 | Packaging | 15 | 90% |
| C091–C100 | Other | 10 | 100% |
Review the investigation before choosing a remedy
A defensible request is to inspect ten stockout and ten late-delivery records while separately screening rare severe issues. That is the example’s proposed investigation, not work already completed. With one operations staff member available, agree capacity and selection rules before assigning it. A convenient selection of ten does not automatically represent every complaint in its category.
For each inspected record, retain the original report, the classification reason, relevant operational evidence and what remains unknown. Check how the main issue was selected when several were mentioned. A complaint about late delivery may involve different events and cannot be converted into a universal handoff diagnosis. Take a specific supported branch into further investigation instead of using the largest bar as a solution label.
At review, distinguish records inspected, explanations supported and actions approved. If a change is tried, use the same definitions next time and report the exposure needed to interpret counts. Fewer complaints could accompany fewer orders rather than better service. Retain serious exceptions, scope changes and unresolved records. The ranking has helped when it directs a checkable next question, even if no remedy has yet been selected.
Check your work
- Categories cover the intended population without double counting; exclusions and Other are explained.
- All values use one measure and unit. Counts and amounts are not combined, and missing values are not replaced with zero.
- The total is positive, values are finite and nonnegative, and category names are unique.
- The chart shows descending values and the final cumulative share is 100%; displayed percentages may differ slightly due to rounding.
- The 80% line is a reference. Severity, cost and rare risks are checked separately before prioritizing actions.
Framework references
References explain the framework and its scope; they do not validate the fictional example's results.
Put the framework to work
Copy the outline and add your question, evidence, and open questions.
Read a practical guide →Practice with an example
Reproduce historical retail RFM results, compare filtering rules and customer lists, or paste aggregated customer results with definitions and investigation notes.
Work through an RFM analysis with public retail data →