Insights

Root cause analysis: methods, template and a worked example

How to find the real reason a problem keeps happening, which method to use when, and a template you can copy for your next investigation.

By Christopher Fagan, Managing Partner, PEF Consulting Partners

£13bnlargest programme delivered
19 yrssenior systems delivery
5 continentsdelivered across
£25msaving on a single programme

Delivered for and alongside

ShellBPCapitaBaringaOfgemSmart DCCInsightsFlui TechnologiesExxonMobilVodafoneConstellation SoftwareDeloittePwCEYNational GridElexon

Root cause analysis (RCA) is a structured way of working back from a problem to the underlying cause that, once removed, stops it happening again. It tests candidate causes against evidence, confirms the one that matters, and ends with a corrective action that has a named owner and a date.

Most organisations are good at fixing symptoms. A team clears a queue of failed invoices every Monday, the queue is back by Thursday, and after a few months the overtime looks normal. RCA is the discipline of asking why the queue exists at all, and refusing to stop at the first plausible answer.

What RCA is, and what it is not

A root cause is the deepest cause in a chain that you can actually act on. Fix it and the problem stops, and similar problems often stop too. A contributing cause makes the problem worse or more likely but would not produce it on its own. A symptom is simply what you can see: the backlog, the complaint, the red status report.

RCA is not a search for someone to blame. When an investigation ends at “human error”, it has usually stopped one question too early. The better question is why the system made the error easy to make and hard to spot, which leads to causes in process, controls, data or decisions that can be fixed.

It is rarely about a single cause either. Serious failures usually have one or two root causes and several contributing ones, and a good analysis names and ranks them all.

Where RCA comes from

There is no single inventor. Most of today’s methods came out of manufacturing quality and engineering safety, and were later taken up by healthcare, aviation and IT service management, where finding the root cause of recurring incidents is the core of problem management.

The 5 whys is associated with Sakichi Toyoda and the Toyota Production System, where Taiichi Ohno described asking why repeatedly to reach the real cause of a shop floor problem. The fishbone, also called an Ishikawa diagram after its originator, sorts candidate causes into categories. Fault tree analysis was developed at Bell Labs in 1962 to assess the launch control system of the Minuteman missile. Pareto analysis is named after the economist Vilfredo Pareto and was brought into quality management by Joseph Juran, who wrote about the vital few causes behind most of the effect.

Root cause analysis methods, and when to use each

No method is best for everything. Most investigations combine two or three: one to generate candidate causes, one to narrow them with data, and one to drill into what is left.

5 whys

State the problem, ask why it happened, then ask why of each answer until you reach a cause you can act on. Five is a rule of thumb, not a target. It is quick and suits contained problems with a fairly clear chain of events, but it follows one path, so it can miss parallel causes and tends to reflect what the room already believes.

Fishbone diagram

A fishbone diagram sets out candidate causes under headings such as people, process, technology, data, measurement and environment. It is a tool for generating and structuring ideas rather than proving them. Use it early, with the people who do the work, to stop the group fixing on the first explanation offered.

Fault tree analysis

Fault tree analysis puts the failure at the top and works downwards, using AND and OR logic, to the combinations of events that could cause it. It suits complex technical systems and safety cases, where you need to show every route to failure and, where data allows, how likely each is. It takes more time and skill than the other methods, so keep it for problems that justify it.

Pareto analysis

Pareto analysis ranks causes or categories by how often they occur, or by what they cost, so you can see which few account for most of the effect. It is usually the quickest way to cut a long list of candidates down to the ones worth investigating. It only works if the data is categorised consistently, which is worth checking before anyone draws a chart.

How to carry out an RCA, step by step

  1. Contain the problem first. Protect customers, revenue and compliance while you investigate. Containment is not the fix, so record it separately.
  2. Write a precise problem statement. Say what is happening, where, since when and how big it is. Leave out causes and names, because a statement that already contains the answer will steer the whole analysis.
  3. Gather evidence and build a timeline. Pull logs, reports, tickets and volumes before memories fade, set out what happened in order, and list everything that changed around the start date, such as releases, data loads, staff and rules.
  4. Generate candidate causes. Use a fishbone or a structured workshop with the people closest to the work, including the ones who are not usually invited.
  5. Narrow with data. Use Pareto analysis or a simple count to see which candidates explain most of the effect.
  6. Drill down. Apply the 5 whys to each surviving candidate until you reach something you can change.
  7. Verify the cause. A real root cause explains all of the evidence, not just most of it. The strongest test is to remove it on a small scale and watch the problem stop.
  8. Agree corrective and preventive actions. Corrective action fixes this instance; preventive action stops it recurring here and elsewhere. Each action needs one owner and a date.
  9. Check effectiveness. Set a date to confirm the effect has gone, and only close the analysis when it has.

A root cause analysis template you can copy

Copy this one page template into a document or spreadsheet. Fill it in as the investigation runs, rather than writing it up from memory at the end.

FieldWhat to recordQuality check
Problem statementWhat is happening, where, since when and how big, in one or two sentencesNo causes, no names, and a number someone can measure
EvidenceData, logs, reports, tickets and interview notes, each with its source and dateCould someone outside the team check it?
Candidate causesEvery plausible cause raised, grouped by category, with the evidence for and against eachHave the people who do the work contributed?
Root causeThe cause or causes confirmed by evidence, with contributing causes listed separatelyDoes it explain all the evidence, and has it been tested?
Corrective actionWhat will change to fix this instance and to prevent it recurringDoes the action remove the cause, or only the symptom?
OwnerOne named person accountable for each actionA person, not a team or a department
DateWhen each action is due, and when its effect will be checkedIs there a date for the effectiveness check as well as the fix?

Keep completed templates together. Patterns across analyses, such as the same weak control appearing again and again, are often worth more than any single finding, and they belong in your lessons learned log.

Worked example: billing exceptions after a system go live

This example is hypothetical. The supplier, figures and details are invented to show how the methods fit together.

A mid sized energy supplier moves its customer accounts onto a new billing platform. Three weeks after go live, the number of bills held for failed validation has risen more than tenfold. The operations team adds overtime, but the queue keeps growing and complaints about late bills start to climb. The hypercare team has been treating it as a capacity problem.

Problem statement. Held bills have risen from around 300 to around 4,000 a week since go live, mostly on two older meter types in two regions. Accounts on newer meters are largely unaffected.

Candidate causes. A fishbone session with billing operations, the data team and the platform supplier produces about twenty candidates, including unfamiliar exception screens, an exception process not yet updated, default validation tolerances, migrated data, double counting in the exception report and higher winter read volumes.

Narrowing. A Pareto chart of held bills by reason code shows a single reason, meter read outside tolerance, behind most of the queue. Seasonality is ruled out because the same weeks last year showed no rise. Staff inexperience is slowing the clearance, so it is a contributing cause, but it does not explain why the bills fail in the first place.

The 5 whys. Each answer below was checked against data before the next question was asked.

  1. Why are bills being held? Because meter reads are failing the tolerance check.
  2. Why are the reads failing? Because the expected consumption held for those meters is far too low, so normal reads look abnormally high.
  3. Why is the expected consumption too low? Because it was migrated from a legacy field that, for the older meter types, stored the figure in a different unit.
  4. Why was that not caught? Because the data mapping did not record units for that field, and the migration reconciliation compared record counts rather than values.
  5. Why did the reconciliation only compare counts? Because the migration acceptance criteria covered completeness, and nobody in the business owned the accuracy of fields that drive billing.

Verification. The data team corrects the expected consumption for a sample of 200 affected meters and reruns validation. Those bills release, and no new exceptions appear for them in the next billing cycle. The cause has been switched off on a small scale, which is the test that matters.

Root cause and actions. The root cause is a control gap: migration acceptance covered completeness but not the accuracy of billing critical fields. The actions are to correct the data for every affected meter, release and check the held bills oldest first, add value level reconciliation and business sign off to any further data loads, and update the exception process and training. Each action gets an owner and a date, with the effectiveness check set two billing cycles later.

Notice where the analysis ended. The overtime, training and tolerance settings all mattered, but the cause that would have produced the next failure was a gap in how the migration was accepted. To put a figure on what a queue like this costs while it clears, try the billing exception cost calculator.

When to use RCA, and when not to

Use it when a problem keeps coming back, or when a single failure matters on its own: an outage, a regulatory breach, a missed go live or a large write off. It also fits an audit finding that needs a credible answer, or a programme that keeps missing milestones while the explanation changes every month.

Do not use it when the cause is already known and the fix is cheap; just fix it. Do not let it delay containment, or run a formal analysis on a genuinely random, one off event that no reasonable control would have prevented. Scale the effort to the problem: a short 5 whys conversation for a minor issue, a formal investigation for a major incident.

Common mistakes

  • Stopping at human error. People make mistakes in every system. The useful question is why the system let the mistake through.
  • Starting with the answer. If the problem statement says “the new system is causing errors”, the analysis will find that the new system is causing errors.
  • Treating a theory as a finding. A cause that has not been tested against the evidence is still a candidate, however senior the person who suggested it.
  • Running it too late. Logs roll over, people move on and memories rewrite themselves. Start gathering evidence while the problem is live.
  • Closing on the action, not the result. An RCA is finished when the effect has gone, not when the corrective action has been raised.

How root cause analysis fits wider delivery work

RCA is not only for operations. When a transformation programme is in trouble, the visible problems, such as slipping milestones, red status reports and supplier disputes, are symptoms. The causes usually sit upstream: unclear scope, slow decisions, a weak contract or a plan nobody believed. A programme recovery health check is a quick way to see which of those are in play, and our guide on how to rescue a failing programme covers what to do next.

It also pairs well with forward looking tools. A gap analysis measures the distance to a target; RCA explains why you fell short of it. Where the same causes keep appearing across a business, the answer is usually a wider improvement effort rather than more investigations, which is the territory of operational excellence.

This is work PEF does with clients rather than for them. Our Managing Partner, Christopher Fagan, has worked on problems like this across 19 years of systems and transformation delivery, and traced and recovered a billing imbalance of more than £160m for a Big Six energy supplier after its Kraken migration. When the problem sits in billing, our billing integrity and revenue assurance work runs alongside your operations team; when it sits in a programme, our programme recovery service does the same.

Common questions

What is root cause analysis?

Root cause analysis is a structured method for finding the underlying cause of a problem so it can be fixed rather than managed. It starts with a clear problem statement, gathers evidence, tests candidate causes, confirms the cause that explains all of the evidence, and ends with corrective actions that each have an owner, a due date and a check that the problem has actually gone.

What are the main root cause analysis methods?

The most widely used are the 5 whys, the fishbone diagram, Pareto analysis and fault tree analysis. The 5 whys drills down a single chain of cause and effect. A fishbone diagram generates candidate causes by category. Pareto analysis shows which few causes account for most of the effect. Fault tree analysis maps every combination of events that could produce a failure, which suits complex technical systems.

How do you do the 5 whys?

Write a clear problem statement, then ask why it happened. Take the answer and ask why again, and keep going until you reach a cause you can act on, which is often around five steps but not always. Back each answer with evidence rather than opinion, follow each branch if an answer splits, and stop when fixing the cause would prevent the problem recurring.

What should a root cause analysis template include?

At minimum it should hold a problem statement with a measurable size, the evidence gathered and its sources, the candidate causes considered, the confirmed root cause with contributing causes listed separately, the corrective actions, one named owner per action, and dates for both the fix and the check that it worked. Recording rejected causes, and why they were rejected, saves time if the problem returns.

What is the difference between a root cause and a contributing cause?

A root cause is the underlying cause that, if removed, stops the problem happening again. A contributing cause makes the problem more likely or more severe but would not produce it on its own. Most significant failures have one or two root causes and several contributing ones, and a good analysis records both so that the actions address the whole picture.

Related reading: Fishbone diagram: template and example, Gap analysis: how to do one, SIPOC diagrams explained and RAID log template.

Got a problem that keeps coming back?

We work alongside your team to find the cause, prove it and put the fix in, whether it sits in billing operations, a system or a programme. Talk to us about a short diagnostic.

Book a confidential conversation

Sources: Taiichi Ohno, Toyota Production System: Beyond Large Scale Production, 1988 (English edition); Kaoru Ishikawa, Guide to Quality Control, 1976 (English edition); Joseph M. Juran, Quality Control Handbook, 1951

PEF TRACK RECORD£13bnLargestprogramme led5Continentsdelivered across19 yrsSystems delivery£25mLargestsingle saving
CF

Written by Christopher Fagan

Managing Partner, PEF Consulting Partners

Christopher has led systems and transformation delivery for 19 years, including the National Grid separation, the £9bn Energy Bills Support Scheme and the BP Pulse EV charging rollout.