Evidence constraints
What information can the AI rely upon?
AI IN MANUFACTURING · WORKFLOW EXPLORATION
How should AI actually change the way quality and warranty work gets done?
A defect is usually recorded as a problem to be resolved. A warranty claim is usually treated as a case to be processed.
But neither begins with a decision. It begins with a signal.
Someone notices something. Someone describes it. Someone classifies it. Someone tries to understand what happened. Eventually, someone has to decide what it means — and what to do about it.
AI can participate across almost every part of this chain.
Can AI identify defects?
How should intelligence change the way an organisation understands, connects and acts on what its defects are telling it?
WORKFLOW
DEFECT, QUALITY & WARRANTY ANALYSISTHE CURRENT REALITY
The request usually sounds simple.
“Can we use AI to find defects faster?”
A quality or service team wants to reduce the time spent reading reports. Someone suggests AI. A tool is evaluated. Data is connected. A model is introduced.
Perhaps it can classify incoming cases. Perhaps it can summarise service reports. Perhaps it can identify recurring words.
The organisation now has an AI capability. But the underlying problem may still be there.
A technician describes a failure one way. A dealer describes it another. A warranty team classifies it differently. A quality engineer sees something that isn't obvious in the original report.
A pattern only becomes visible after hundreds of cases. And by the time the organisation recognises the pattern, the product may already be in the field.
The organisation improved the processing of the signal.
It may not have improved its ability to understand what the signal means.
That is where the question changes.
There are two ways to approach this.
CHOOSE A PATH TO EXPLORE
Select a path to see how the starting question changes.
Let's look at the work itself.
SIGNAL → DESCRIPTION → INTERPRETATION → PATTERN → ACTION
What does quality and warranty work actually carry?
Follow the work from the first observation to the consequence that follows.
Where does the difficulty sit in your workflow?
SELECT ONE STAGE
SELECT A STAGE
Signal → Description → Interpretation → Pattern → Action
Choose the point where the work becomes difficult for your organisation.
The workflow is recognisable. The information around it changes with the physical and operating context.
INDUSTRY → WORKFLOW
Where this workflow appears
SELECT YOUR CONTEXT
The underlying chain is similar. The failures, evidence and consequences are not.
The sequence is simple. The understanding is not.
THE COGNITIVE TURNING POINT
The signal isn't the problem.
The interpretation is.
A defect report can be perfectly accurate and still be difficult to understand.
A technician may record: ‘Noise from unit during operation.’ Another may write: ‘Intermittent vibration under load.’ Another: ‘Bearing appears to be failing.’
These may describe the same underlying problem. Or they may not. That distinction matters.
Because the value of the workflow isn't simply in collecting more reports. It is in helping the organisation move from:
01what someone observed
02what the organisation believes is happening
03what it should do about it
The organisation's understanding of the defect is itself part of the workflow.
ONE USEFUL INSIGHT
A signal only becomes useful when the organisation can interpret it.
WHERE JUDGEMENT BECOMES VISIBLE
Recording more observations can improve visibility without improving understanding.
The work changes when a description must be interpreted, connected to other cases and judged consequential enough to act upon.
- What was actually observed
- What the organisation believes is happening
- What evidence justifies action
Signal ≠ Interpretation.
ILLUSTRATIVE WORKFLOW EXAMPLE
Different descriptions can point to one failure — or several.
EXAMPLE · NOT A DIAGNOSIS
- Three field reports describe noise, vibration and bearing wear in different language.
- AI connects the reports as a possible pattern without declaring a common cause.
- Quality and engineering compare operating conditions, product configuration and inspection evidence.
- A validated pattern leads to a focused investigation and accountable action.
Signal → Interpretation.
So the examination follows how an observation becomes an organisational explanation.
How we would examine it
DIMENSIONS · NOT A CHECKLIST
These aren't a checklist to complete. They are dimensions through which we examine how the workflow actually operates.
These questions explain the direction of the examination without turning the visitor's context into a public diagnosis.
And this is where AI requires more care.
AI PARTICIPATION · CONSEQUENCE-APPROPRIATE CONTROL
AI can help the organisation see more.
AI can read large volumes of service notes.
It can compare descriptions.
It can identify similarities that would be difficult to find manually.
It can surface recurring themes.
It can organise unstructured reports.
It can help people investigate a problem.
It can also be wrong.
Two descriptions may look similar without having the same cause.
A rare but important failure may be buried among common cases.
An AI system may infer a cause that isn't supported by the evidence.
A confident summary can make uncertainty disappear.
This matters enormously when the output could influence:
- a warranty decision
- a quality investigation
- a product change
- a recall decision
- a safety-related action
- a meaningful customer or financial consequence
The more consequential the action, the stronger the control around the interpretation should be.
Validation
How is an AI-generated interpretation checked against actual evidence?
Confidence & uncertainty
How does the workflow distinguish a useful signal from an uncertain inference?
Decision boundaries
What can AI surface or recommend, and what requires qualified human judgement?
Exception handling
What happens when the evidence conflicts or the case does not resemble anything previously observed?
Auditability
Can the organisation understand how an observation moved from signal to action?
AI should help the organisation see more. It should not make uncertainty invisible.
So what would actually change if we redesigned the work?
What could change?
THE WORK · NOT ONLY THE TECHNOLOGY
An intervention might change much more than the technology used to analyse defects.
SIGNAL · DESCRIPTION
Compress the work
Reduce repetitive reading, classification and information handling.
DESCRIPTION · PATTERN
Connect the work
Bring together observations that currently sit across service, warranty, quality and engineering.
INTERPRETATION
Improve the work
Give people better evidence when they need to interpret a failure.
PATTERN
Surface patterns earlier
Help the organisation recognise potentially meaningful patterns before they become obvious through volume alone.
INTERPRETATION · ACTION
Clarify the work
Make the boundaries between observation, interpretation, investigation and action explicit.
ACTION
Control the work
Ensure that consequential conclusions remain appropriately validated and accountable.
The objective isn't simply to process more warranty cases. It is to make the organisation better at understanding what its failures are telling it.
ONE QUESTION WORTH ASKING
How long does it take your organisation to realise that several different complaints are actually the same problem?
You may know the answer.
You may not.
Either is a useful place to begin.
FIND THIS WORK IN CONTEXT
One workflow, connected to the organisation around it.
Return to the wider context without creating a separate version of this workflow.
Relevant functions
Operations & Quality →Related workflows
Service & Maintenance Diagnosis →YOUR WORKFLOW
How does your Defect, Quality & Warranty Analysis workflow operate today?
The page above examines Defect, Quality & Warranty Analysis as a workflow. Now apply the same lens to your own organisation. Answer eight questions and get a structured read on its operating condition.
Take the workflow assessment →You've looked at the workflow.
Now tell us what matters most.
YOUR WORKFLOW · YOUR PRIORITY
What do you want to change?
You know your workflow better than we do. Start by telling us where it hurts.
Tell us what matters most. We'll use it to understand where the workflow may be carrying unnecessary work, uncertainty or judgement.
ALREADY KNOW YOU WANT TO TALK?
Bring the workflow as it operates today.
You don't need to redesign the process before we examine it. Bring the workflow as people experience it now — including where signals, interpretation, patterns and action become difficult.