WorkflowQuality Operations & CAPA

Root Cause Analysis Methods

Root cause analysis that fits the method to the event, proves or rules out every fishbone cause from a cited record, and runs each proved cause down to the system that let it happen.

Follow one granulator investigation from method to approval ↓
Watch with transcript →
The problem

Two seal deviations closed at the worn part, and the third ran with no alarm

An investigator has a fixed clock to say why a batch left its validated range, and an inspector will read the answer. The method is whichever one that investigator knows best, the fishbone is drawn in a meeting, and the minutes call a cause ruled out with nothing on file to show why.

The work ends at the first physical cause. The seal was worn, so the seal was replaced. Nobody asks what let a weeping seal stay in service, or why a software update moved the high-torque alarm past what was validated, while the granulator next door carries the same limit.

Built forQA investigatorsProcess engineersQA investigation approversHeads of quality

The Problem Today

A cause cleared in a meeting has nothing on file behind it, and an investigation that ends at the worn part lets the failure come back.

164
Times FDA cited 21 CFR 211.192 in drug inspections, fiscal 2025 published

Pulled by hand, case by case

  • eBR steps
  • Historian
  • Work orders
  • Change log
  • One investigator, one favorite toolThe method follows the person instead of the event
  • The quiet alarm goes unaskedNobody tests why the protective layer did not act

The Solution

  1. A method chosen for the eventIs/Is-Not, fault tree, five-why or FMEA, and the reason it fits
  2. Six bones read against recordsEach candidate cause proved or ruled out, its record cited
  3. The inspector's questions firstA challenger caught the agents' draft naming only the seal
  4. Signed once each fact fitsThe cause test passes, the investigator signs, QA approves

Built-in controls

  • A changed root cause loses both signatures
  • Nobody approves an investigation they signed
  • Supplier free text never decides a cause

What You Get

Root causes that have already met the inspector's questions, with a record behind every cause ruled in or out.

  • New investigators reason like veteransThe method and each ruling arrive with the reason
  • Fixes that reach sister granulatorsSystem causes go to a CAPA for each similar machine
  • Verdicts the next case starts fromEach proved or ruled-out cause is a cited record to reuse
Features

Root cause analysis that proves the second cause its own first draft missed

A QA investigator works one wet-massing torque excursion on a high-shear granulator through each method, while the approver and engineering work on the same record.

demoEverything on these screens, from the plant and its clients to each person, batch and reading, is sample data.

  1. 01Method chooser

    Root cause analysis methods suggested for the event, each with its reason

    • Is/Is-Not, fault tree, 5-Why or FMEA. The chooser reads what kind of event it is, such as one asset that changed and an alarm that did not act, and names the method that fits it.
    • The investigator picks, and the reason goes on record. Adding FMEA asks for a reason code, so the report already says why each method was used before an inspector asks.
  2. 02Proved fishbone

    Fishbone diagram with each candidate cause proved or ruled out by a record

    • One reader for each of the six bones. Machine, method, material, measurement, people and environment are each read against their own records, so no bone is skipped because the meeting ran long.
    • Ruled-out causes stay on the bone. A cleared cause is struck through with the record that cleared it beside it, so the binder lot reads as two clean batches and a certificate.
    • No verdict before the records are read. Each candidate shows as not read yet until its reader has opened the records, so last month's answer is never assumed for this event.
  3. 03Challenger

    A challenger puts the inspector's questions to the agents' own first draft

    • Questions from the inspector's side. Why the alarm did not act, why two earlier deviations closed with no system cause, and whether human error was cleared by records or by interview.
    • Its own draft attacked first. The first point is aimed at the agents' work: a limit set past the validated range is a cause in its own right, and the draft had filed it as contributing.
  4. 04Draft root cause

    A second root cause proved from change control and the alarm's audit trail

    • The alarm limit proved from its audit trail. The method reader opens the change record and the configuration audit trail, finds the update that raised the limit, and the draft names two root causes.
    • Contributing kept apart from root cause. A batch record screen that showed torque without its range is recorded as contributing, so the operator is not blamed for a limit software had moved.
  5. 05Cause test

    Is/Is-Not analysis and a cause test the surviving causes must pass

    • Cited cells in the problem specification. What, where, when and extent come from records, so a distinction such as the sister granulator running the same binder in range is a fact the report can show.
    • Surviving causes tested against each fact. A cause set that leaves one fact unexplained cannot be signed, so a partial answer such as the seal alone goes back to the bones.
    • Locked while anything is open. The test will not pass with a bone unread, a challenger point unanswered or a ruled-out cause missing its record.
  6. 065-Why and fault tree

    5 whys root cause analysis that runs each proved cause to a system cause

    • Each why cited to its record. The chain from a worn seal reaches the maintenance plan that let it keep running, so the fix lands in the plan itself.
    • No chain ends at the part or the person. The why-chain agent will not stop at human error or retraining; it stops at the procedure or system that allowed the event.
    • A fault tree for the alarm that stayed quiet. Each way a protective layer can fail is tested against a calibration, alarm log or audit trail record, and only the raised limit comes out true.
  7. 07QA approval

    QA approval held until each proved cause's records have been opened

    • Each record held open before it counts. The approver reads the audit trail, the change record and the validation report behind each proved cause, and approval stays locked until all have been read.
    • The cleared causes read as closely. The ruled-out and contributing list is opened with its records before approval, so what was cleared gets the same scrutiny as what was proved.
  8. 08Report section

    The investigation report's root cause section, built from each method's page

    • Each method's page in the report. The fishbone table, the cause test, the 5-Why and the fault tree go in as cited pages, so the written record of the investigation holds its conclusions and follow-up.
    • Reopening clears both signatures. A changed root cause marks the export unsigned until the investigator and QA sign again, so no stale approval travels with the file.

Investigators and engineers work the bones; the quality head owns the rules

The investigator works the case, engineering confirms what the readings mean, a separate approver checks it for QA, and the quality head owns the rules.

QA investigators

Take each investigation to a signed root cause before its clock runs out

  • The method fitted to the event, with its reason
  • Each bone read against its own records
  • The inspector's questions answered on the page
QA investigation approvers

Approve each root cause after reading the records behind it

  • A checklist of the records to open
  • The cleared causes with their evidence
  • Send-backs that teach the next investigation
Process and maintenance engineers

Answer the process and equipment questions the bone readers raise

  • Requests in fixed fields with a named owner
  • Trends and work orders cited where they are used
  • System causes that change the maintenance plan
Heads of quality

Own the method rules and approve what the agents learn from rulings

  • Learned rules waiting on their approval
  • Loosened controls held for their signature
  • Investigations an inspector can follow cause by cause

Nobody can approve a root cause they signed as its investigator

Each person signs under their own ID and password, with what the signature means and why.

QA investigator
Sees
The fishbone, the cause test and any challenger point still open
Signs
The root cause, stating that the cited records support it, and why
QA investigation approver
Sees
The records behind each proved cause and the list of causes ruled out
Signs
The approval as QA reviewer, or a send-back giving its reason
Process engineer
Sees
The torque and dry-mix power trends, and the question asked about them
Signs
A fixed-field answer that the readings are comparable, which can be withdrawn
Head of quality
Sees
The ruling a proposed rule came from, and any setting that loosens a control
Signs
Each learned rule before it applies, and any loosened setting
More screens

A send-back over a binder ruled out with no record became a standing rule

Around the fishbone: what was cleared and by which record, the seal's history, the open investigations, the rules taken from rulings and the signatures.

Twelve lines cleared, each beside its recordEach ruled-out and contributing cause with the record that settled it, from the drive's vibration and the binder's comparison batches to the calibration and room humidity.
A seal found weeping a month earlier stayed in service with the purge raisedThe work order that removed the seal at containment, the maintenance plan with no run-hour trigger, and the dry-mix power that crept up over the nine batches before the event.
Four investigations past their 30-day clock, in one queue across the plantsOpen investigations across four plants with class, method, stage, investigator and days left; 5-Why can be proposed in bulk for minor ones while the approver samples the choice.
Rules taken from reviewers' rulingsEach rule names the send-back or ruling it came from and waits for the quality head's approval; the newest asks for a system cause behind each equipment cause.
The approval names the two causes it covers and why it was givenThe investigator's signature and the QA approval with meaning, reason and session, filtered from the full trail of people and agents and exported as CSV.

demoSample data, from the working solution.

What the challenger found that the meeting-room fishbone had let through

An alarm raised above the validated range

A recipe import during a software update moved the high-torque alarm from 340 to 380 N·m, and the change control had compared screens and reports only.

Two seal deviations with no system cause

Both earlier deviations on the granulator ended at the seal, while the plan had no run-hour trigger and let a weeping seal run with the purge raised.

A supplier's certificate on its own

The binder lot was cleared by the same lot running in range on two other batches, with the certificate of analysis as the second record.

Human error cleared by records

Signed batch record steps with second-person checks and the operator's requalification ruled out a slip, which EU GMP Chapter 1 requires before concluding human error.

demoSample data, from the working solution.

The challenger argues against the agents' own draft and may remove no cause

Each agent proposes from the GMP records it reads; the investigator and QA decide.

Method chooser

Reads the event's features and suggests the method or methods that fit, with the reason.

  • Locks no method: the investigator chooses
Bone readers

One per bone, each reads its own records and proves, rules out or keeps open each candidate with a cited passage.

  • Rules out a cause only with a record
Cause tester

Checks that the surviving causes explain each fact on the timeline and in the Is/Is-Not.

  • Never passes with a bone unread or a point open
Why-chain agent

Runs a five-why from each proved cause to a system cause and builds the fault tree.

  • Never stops at human error or retraining
Challenger

Argues as an inspector would against each draft, including the agents' own first one.

  • Removes no cause
  • Cannot sign
Request agent

Writes structured requests to the supplier, maintenance and process engineering, and reads the replies.

  • Free text never changes a verdict
  • Contacts no outside party unless the investigator sends it

What QA investigators ask about root cause analysis

What is root cause analysis?

Root cause analysis finds why a deviation happened, down to the cause that stops it returning once it is fixed. Here the method is chosen for the event, each candidate cause on the fishbone is proved or ruled out by a cited record, and a five-why carries each proved cause to the system that let it happen.

Which root cause analysis method fits which event?

Is/Is-Not with problem analysis fits a sudden change on one asset where a comparison exists. A fault tree fits a protective layer that failed to act, FMEA a weakness across steps, products or a transfer, and a five-why a simple single cause, or a proved cause that still needs its system cause.

How is a cause ruled out?

Only by a record. A setting that requires a cited record for each ruled-out cause is locked on, and a cleared cause remains on its bone, struck through beside that record, so a reader of the report sees what cleared it.

Can human error be the root cause?

Not until the process, the procedure and the system around the operator have each been cleared by records, as the EU GMP guide asks. The people bone is read from the signed batch record steps and the training record, and no why-chain ends on retraining.

What sets this apart from other root cause analysis tools?

Agents open the batch record, historian, maintenance, calibration and change records bone by bone, a challenger questions the draft, and a cause test refuses a root cause that leaves a fact unexplained.

What happens when a supplier answers in free text?

Requests go out in fixed fields, such as the retained sample's K-value and water, and only those fields can move a bone. A free-text note waits for the investigator, and nothing reaches an outside party unless the investigator sends it.

Send us the investigation that
closed at the worn part.

Send one closed investigation with its batch record, historian data and work orders, and we will read each bone against its records.