A Pareto chart ranks categories from the greatest to the smallest impact and shows how much each category adds to the cumulative total. In maintenance, it helps reveal which failure families concentrate events, downtime hours, or cost. The result is defensible only when the team defines the question, prepares the work-order dataset, normalizes categories, and chooses a coherent unit before ranking the data.
The same set of work orders can produce three different priorities. Seal loss may lead by frequency, while cavitation leads by downtime hours and a rare structural failure rises when cost is used. A Pareto chart does not make the decision. It makes concentration visible so that engineering, reliability, and maintenance planning teams can decide with operational context.
What is a Pareto chart?
A Pareto chart combines two views. Bars show the value of each category, ranked from highest to lowest. A line shows the cumulative percentage of those categories over the total. The horizontal axis contains the categories, the left vertical axis uses the selected unit, such as events, hours, or currency, and the right vertical axis runs from 0% to 100%.
The U.S. Agency for Healthcare Research and Quality, AHRQ, recommends selecting a standard measure, ranking items by frequency, cost, or time, and plotting the cumulative line. The NIST/SEMATECH e-Handbook includes Pareto charts among tools for process monitoring and investigation. These references support the general mechanics. Applying them to maintenance depends on the quality and comparability of plant records.
An ordered table is enough to perform the analysis. The chart speeds up interpretation, especially with many failure families. A bar answers, “How much does this category represent?” The line answers, “How much have we accounted for by including this category and all previous ones?”
80/20 is a reference, not a maintenance law
The 80/20 expression became associated with the method because many distributions are concentrated. Plant data do not have to reproduce exactly 80% of the effect in 20% of the causes, and 80% cumulative is not a universal cutoff for action.
Three families may represent 55%, six may represent 82%, or the distribution may be nearly uniform. The chart must show what the measured data accumulate. The team chooses where to focus resources based on intervention capacity, expected return, risk, criticality, recurrence, and classification quality.
Forcing an 80% cutoff creates two errors. It may include a category only because the line has not reached a conventional number, even when no viable action or technical hypothesis exists. It may also abandon the tail, where a rare failure can carry an unacceptable safety, environmental, or operational consequence. Pareto prioritizes concentration. It does not authorize risk acceptance.
Define the question and unit before extracting work orders
The question determines the dataset and metric. “Which failures repeat most often?” requires frequency. “Which failures remove the most production capacity?” requires downtime hours or comparable production loss. “Which failures place the greatest burden on maintenance?” requires a defined cost scope. Combining these questions in one column produces a meaningless ranking.
Frequency
Frequency counts failure events, not operation lines, confirmations, materials, or duplicate notifications. It is useful for recurrence, administrative workload, and maintenance-process stability. A family with many short events may lead by frequency and remain low by downtime.
Define the counting unit. One shutdown with two orders, five operations, and eight confirmations remains one event if every record belongs to the same occurrence. Without an event key or consolidation rule, Pareto measures how the system was filled in instead of the failures.
Downtime hours
Downtime hours approximate operational impact but require a common clock. Notification opening, failure start, maintenance start, technical return, and production return are different milestones. State which interval is measured and how parallel work, material waiting, operating windows, and partial capacity are treated.
Adding the duration of two orders executed in parallel can double-count asset downtime. Treating labor hours as downtime mixes resource consumption with loss of function. Both indicators are useful, but they answer different questions.
Cost
A cost Pareto must state what it includes. Internal labor, external service, materials, freight, rental, product loss, energy, and opportunity cost are not automatically equivalent or available from the same source. In SAP, the cost scope and posting timing depend on the company’s accounting process, configuration, and settlement practices.
Choose a stable definition and a period in which postings are sufficiently complete. Planned and actual cost must not be added as separate components. If production loss is estimated, keep it separate from recorded cost and document the formula.
How to prepare the work-order dataset
Chart quality begins before the spreadsheet. Use a reproducible procedure:
- Set the boundary. State plants, areas, asset classes, order types, period, and statuses. Separate contexts that are not comparable.
- Select the reference date. Failure start, execution completion, and technical completion place the same event in different periods. Select one and use it consistently.
- Define the event. Create a key that consolidates records for the same occurrence. Preserve notification, order, and technical-object numbers for traceability.
- Remove cancellations and duplicates using explicit rules. Record the exclusion reason and retain a reconciliation count.
- Separate planned work from failure events. Inspection, preventive work, improvement, and planned shutdown orders do not automatically belong in a failure Pareto.
- Validate time data. Look for negative durations, open clocks, overlaps, and late entries reconstructed from memory.
- Validate costs. Identify open orders, pending settlements, reversals, currencies, and cost centers outside scope.
- Freeze the dataset. Save extraction date, filters, dictionary version, and record counts before and after each treatment.
Official SAP Asset Management documentation shows that maintenance notifications can be filtered by type, priority, dates, equipment, and functional location, and related to orders and malfunction data. These dimensions support extraction, but they do not resolve the taxonomy. Fields, types, statuses, and mandatory entries still depend on the configured environment.
Normalize categories without erasing the raw data
A Pareto with “bearing,” “broke,” “mechanical,” “lack of lubrication,” and “B-204” mixes a component, symptom, discipline, presumed cause, and tag. The bars can be ranked, but the categories are not comparable. Normalization must place every row at the same semantic level.
Choose a level appropriate to the question. For screening, use an observed failure-mode family, such as seal loss, bearing degradation, or electrical connection failure. For a second Pareto within the leading family, move down one level to seal type, damage mechanism, area, or confirmed cause. Do not label data as “root cause” when records contain only a symptom or initial diagnosis.
The normalization dictionary should include:
- raw value found in the record;
- normalized category;
- definition and inclusion examples;
- exclusions and boundaries with adjacent categories;
- technical owner of validation;
- version and effective date.
Keep the raw value beside the normalized one so engineering can review the rule and trace a bar back to its work orders. Keep “unclassified” visible. If it becomes large, the first action is to improve data capture. Redistributing unknown records through retrospective judgment creates precision the source does not have.
How to calculate percentage and cumulative percentage
After consolidating events and normalizing failure families:
- sum the selected unit for each family;
- calculate the total across all families;
- rank families from highest to lowest;
- calculate share:
family value / total × 100; - calculate cumulative percentage:
previous cumulative percentage + current percentage; - confirm that the final row reaches 100%, allowing only minor rounding differences;
- plot bars in the original unit and the line on a percentage scale.
Rank the data again whenever the unit changes. Do not keep the frequency order for cost bars. The cumulative line may still close at 100%, but it will no longer show cost concentration from highest to lowest.
Complete industrial example: eight failure families
Every number in this case is a didactic example. The monetary values are illustrative amounts in Brazilian reais, identified by the R$ symbol. They are not converted market data, a client result, an industry benchmark, a process limit, or PM Run performance.
Consider 12 months of consolidated events in a pumping area. After reviewing duplicates and normalizing records, the team obtained this dataset:
| Normalized failure family | Events | Downtime hours | Illustrative cost in Brazilian reais |
|---|---|---|---|
| Seal loss | 34 | 28 h | R$ 140,000 |
| Instrument failure | 23 | 16 h | R$ 92,000 |
| Bearing degradation | 18 | 72 h | R$ 310,000 |
| Electrical connection failure | 12 | 30 h | R$ 175,000 |
| Cavitation-related performance loss | 8 | 96 h | R$ 420,000 |
| Misalignment and coupling | 6 | 44 h | R$ 210,000 |
| Interlock actuation or failure | 4 | 12 h | R$ 65,000 |
| Cracking or structural failure | 3 | 58 h | R$ 380,000 |
| Total | 108 | 356 h | R$ 1,792,000 |
Pareto by frequency
| Rank | Failure family | Events | Share | Cumulative percentage |
|---|---|---|---|---|
| 1 | Seal loss | 34 | 31.5% | 31.5% |
| 2 | Instrument failure | 23 | 21.3% | 52.8% |
| 3 | Bearing degradation | 18 | 16.7% | 69.4% |
| 4 | Electrical connection failure | 12 | 11.1% | 80.6% |
| 5 | Cavitation-related performance loss | 8 | 7.4% | 88.0% |
| 6 | Misalignment and coupling | 6 | 5.6% | 93.5% |
| 7 | Interlock actuation or failure | 4 | 3.7% | 97.2% |
| 8 | Cracking or structural failure | 3 | 2.8% | 100.0% |
Seal loss deserves attention because it represents 31.5% of events. The first three families total 69.4%. A team seeking to reduce recurrence and service workload would examine subfamilies, repeat assets, assembly quality, inspection criteria, and confirmed causes in these groups. It should not conclude that “sealing is the root cause.” The family remains an entry point for investigation.
Pareto by downtime hours
| Rank | Failure family | Hours | Share | Cumulative percentage |
|---|---|---|---|---|
| 1 | Cavitation-related performance loss | 96 h | 27.0% | 27.0% |
| 2 | Bearing degradation | 72 h | 20.2% | 47.2% |
| 3 | Cracking or structural failure | 58 h | 16.3% | 63.5% |
| 4 | Misalignment and coupling | 44 h | 12.4% | 75.8% |
| 5 | Electrical connection failure | 30 h | 8.4% | 84.3% |
| 6 | Seal loss | 28 h | 7.9% | 92.1% |
| 7 | Instrument failure | 16 h | 4.5% | 96.6% |
| 8 | Interlock actuation or failure | 12 h | 3.4% | 100.0% |
The priority changes. Cavitation, fifth by frequency, rises to first and accounts for 27.0% of downtime. Structural failure, last by frequency, moves to third by hours. If the objective is availability, the technical agenda starts with suction-system and cavitation conditions, then bearings, and includes engineering analysis of the three structural events. Acting only on seals might reduce calls but addresses 7.9% of downtime hours.
Pareto by cost
| Rank | Failure family | Illustrative cost in Brazilian reais | Share | Cumulative percentage |
|---|---|---|---|---|
| 1 | Cavitation-related performance loss | R$ 420,000 | 23.4% | 23.4% |
| 2 | Cracking or structural failure | R$ 380,000 | 21.2% | 44.6% |
| 3 | Bearing degradation | R$ 310,000 | 17.3% | 61.9% |
| 4 | Misalignment and coupling | R$ 210,000 | 11.7% | 73.7% |
| 5 | Electrical connection failure | R$ 175,000 | 9.8% | 83.4% |
| 6 | Seal loss | R$ 140,000 | 7.8% | 91.2% |
| 7 | Instrument failure | R$ 92,000 | 5.1% | 96.4% |
| 8 | Interlock actuation or failure | R$ 65,000 | 3.6% | 100.0% |
Cost keeps cavitation at the top but places structural failure second. The three structural events total R$ 380,000 in this didactic example, more than 34 seal events. A cost-led decision may require design, inspection, material, support, or repair-scope review. Verify the hypothesis in work orders and in the field before changing the strategy.
Three legitimate decisions from the same dataset
- Reduce the number of events: build a second Pareto for seal loss and instrument failure, locate repeat assets, and select representative investigations.
- Recover availability hours: prioritize cavitation, bearings, and structural failures while separating diagnosis, waiting, repair, and return-to-service time.
- Reduce direct cost: assess cavitation and structural integrity before a broad program of small events, while maintaining recurrence controls.
None is the correct decision without a declared objective. A balanced portfolio can reserve capacity for recurrence while opening an engineering workstream for low-frequency, high-impact failures. Pareto makes the tradeoff explicit. Governance defines the portfolio.
Biases that distort interpretation
Data-entry bias
One team records “mechanical failure,” while another selects symptom, part, cause, and activity. The broad label becomes a dominant bar. Mandatory fields, long catalogs, defaults, and training shape the distribution. Compare completion quality by shift, area, and role before comparing performance.
Time-window bias
Twelve months can hide a short campaign, off-season, product change, or turnaround. Three months can exaggerate one event. State exposure, seasonality, and regime changes. For populations with different use, consider rates per operating hour, cycle, or ton without mixing rates and absolute counts in one chart.
Recurrence bias
Twenty events across twenty assets indicate a different pattern from twenty events on one asset. Review failure families by asset, functional location, and time between events. Short recurrence after intervention may indicate incomplete repair, weak diagnosis, persistent operating conditions, or duplicate data.
Severity bias
Low frequency does not mean low risk. Safety, environment, quality, and continuity need a parallel decision path. If a tail category contains an intolerable event, it rises through risk governance, not by manipulating the count.
Critical-asset bias
Combining A, B, and C assets may allow many events in light auxiliary equipment to hide one event in nonredundant equipment. Stratify by criticality, function, line, or class. Use the asset criticality matrix when consequence requires treatment regardless of statistical concentration.
When Pareto is not enough
The chart shows concentration. It does not prove mechanism, causality, risk, or action effectiveness. Combine it with other methods when:
- a specific occurrence matters: open a root cause analysis, RCA, preserve evidence, and test hypotheses;
- the causal chain is relatively linear: use the Five Whys without turning five into a quota or stopping at “human error”;
- unobserved modes must be anticipated: use FMEA with declared functions, modes, effects, causes, controls, and criteria. This destination is currently in Portuguese because no published English sibling was found;
- consequence drives the decision: cross the result with asset criticality and failure-mode risk;
- the distribution changes over time: monitor time series, exposure-based rates, and stability indicators;
- categories are broad: build a second Pareto inside the leading family while retaining order-level traceability;
- failures have multiple combinations: use barrier analysis, fault tree analysis, or another method suited to the question.
The U.S. Department of Energy RCA guide emphasizes that causal factors should reveal control deficiencies and guide corrective actions. Pareto selects an investigation area. RCA tests why the event occurred and which controls must change.
From chart to maintenance decision
- Form the hypothesis. Describe what the concentration may mean and which evidence is missing.
- Open the family. Stratify by asset, model, age, location, operation, shift, condition, and previous intervention.
- Select events for analysis. Choose occurrences with evidence and representativeness, not merely the most expensive order.
- Define the intervention. It may be data correction, design change, procedure review, training, condition inspection, or planned work.
- Update the approved strategy. Change a task list, plan, criterion, or interval only after the technical decision. The maintenance plan receives the task. Pareto does not set frequency automatically.
- Define a verification indicator. Compare rate, recurrence, hours, cost, or another outcome using the same basis and a suitable window.
- Rebuild the Pareto. Use the same dictionary and record scope changes.
Maintenance KPIs help track effect and stability. Frequency by family, MTBF by asset class, MTTR, backlog by criticality, cost, and schedule compliance answer different parts of the cycle. No single KPI demonstrates the success of an entire strategy.
The NASA reliability-centered maintenance guide shows that failure modes, causes, and effects vary with the system and operating context. A leading family points to where analysis should deepen. It does not replace FMEA, RCM, or engineering judgment.
Pareto, SAP PM, and field-data quality
In an SAP PM environment, analysis can relate notification, order, equipment, functional location, dates, type, priority, malfunction data, confirmations, and costs, according to company process and configuration. Extraction must respect which object stores each item and how notifications and orders represent the same event.
PM Run operates as an execution, mobility, and planning layer on top of SAP PM, while SAP remains the system of record. In Mobility, field teams execute orders and return operation confirmations, notes, measurements, and allowed evidence according to configured scope. In Planning, maintenance planning teams organize orders, capacity, resources, and schedules integrated with SAP.
This continuity can improve analytical input when classification, technical object, times, and entries are captured carefully and returned to SAP. It does not guarantee a correct Pareto, make decisions automatically, or replace the dictionary, reconciliation, or engineering analysis. PM Run makes execution data available so the company can build its own indicators and analyses. It does not promise a ready-made or universal Pareto chart.
Technical checklist before presenting the chart
- Are the question and unit written in the title?
- Are plants, assets, work orders, and period declared?
- Is there one row per event, without duplication by notification, order, operation, or confirmation?
- Is one reference date used?
- Are categories at the same semantic level?
- Are raw and normalized values preserved?
- Does “unclassified” remain visible?
- Is downtime separated from labor hours?
- Are actual cost, planned cost, and estimated losses separated?
- Were bars reranked after changing unit?
- Do shares total 100%, allowing for rounding?
- Were critical assets and severe events assessed outside the concentration cutoff?
- Does the decision name an owner, evidence, deadline, and verification criterion?
Technical references
- AHRQ, Pareto Chart, public guidance on standard units, ranking, and the cumulative line.
- NIST/SEMATECH e-Handbook of Statistical Methods, Pareto as a process monitoring and investigation tool.
- U.S. Department of Energy, Root Cause Analysis Guidance Document.
- NASA Reliability-Centered Maintenance Guide.
- SAP Help Portal, Find Maintenance Notification.
Frequently asked questions
How do you build a Pareto chart for maintenance?
Define the question, unit, scope, and reference date. Consolidate one row per event, normalize failure families, sum by category, rank descending, calculate percentage and cumulative percentage, and plot bars with a line to 100%. Review criticality and severity before deciding.
What is the difference between a Pareto diagram and a Pareto chart?
In practice, both terms usually refer to descending bars plus a cumulative percentage line. The name does not change the need for coherent units, comparable categories, and traceable data.
Must a Pareto follow the 80/20 rule?
No. The proportion is a concentration reference, not a law or mandatory cutoff. The cumulative percentage must reflect observed data. Actions consider capacity, return, risk, criticality, and evidence quality.
Should I use frequency, downtime hours, or cost?
Use the unit that answers the decision. Frequency supports recurrence reduction, downtime approximates unavailability, and cost supports economic prioritization under a declared definition. Build separate views and rerank every one.
Can different asset types be combined in one Pareto?
Only when the question and categories remain comparable. If criticality, function, exposure, or operating regime differ materially, stratify by class, area, or context.
Does Pareto identify root cause?
No. It shows concentration in recorded categories. A family may be a symptom, component, or observed mode. RCA, Five Whys, and field evidence are needed to confirm causes and controls.
How do I know whether an action worked?
Define the indicator, population, exposure, and comparison window in advance. Repeat the analysis with the same dictionary, check recurrence and impact, and confirm that change came from the intervention rather than a classification or scope change.
To connect work orders, field execution, planning, and data return to SAP PM, learn about PM Run’s mobility and planning layer. Classification, Pareto analysis, and engineering decisions remain the company’s responsibility.
