Knowles Knowledge Centre•Unconscious Bias

You Cannot Inspect Your Own Bias,
But You Can Audit Your Record

A practical framework for the one kind of self-knowledge this subject actually offers: the gap between the rule you say you follow and your last twenty decisions.

Introspection does not reach an unconscious association, and a test score does not tell an individual what they will do. What is available is your own record. Write down the rule you believe you apply, score your last twenty applications of it against that rule, and the discrepancy is evidence about you that nobody can wave away.

Evidence-led guide•Singapore context•Updated October 2026

Explore the five-stage method

About this guide

Published byKnowles Training Institute
FrameworkKnowles Own-Record Audit — an editorial synthesis, not an official or accredited standard.
Evidence baseSkillsFuture Singapore • IAL • WEF • ILO • OECD
YOUR OWN RECORD SAYTHERULE SCORETHEM FIXTHERULE FACETHEGAP PULLTWENTY
Four professionals discussing workplace skills around a meeting table in a Singapore office overlooking Marina Bay

In a workplace context, that definition matters more than any account of how bias works. Almost nobody disputes that unconscious associations exist. What goes wrong is the next step: people are invited to look inside and find theirs, which is the one place the answer is not kept, and they come away either untroubled or vaguely guilty, and in both cases unchanged.

This guide explains what an own-record audit is, how it relates to Singapore’s Critical Core Skills, why awareness alone reliably fails to move behaviour, and what an organisation should do about it.

The practical answer

The practical answer at a glance

For practical workplace use, auditing your own record can be organised into five connected stages:

1

Say the rule

Can the employee state, in writing, the rule they believe they apply?

2

Pull twenty

Have twenty of their own real past decisions been retrieved?

3

Score them

Has each one been marked against the stated rule, not against memory?

4

Face the gap

Has the discrepancy been accepted as a finding rather than explained away?

5

Fix the rule

Has either the practice or the stated rule been brought into line with the other?

About this framework

This five-stage structure is the Knowles Own-Record Audit used in this article. It is an editorial synthesis for workplace application—not an official government framework, accreditation standard or universal definition.

There is no single agreed account of what unconscious bias training achieves

“Unconscious bias training” is an umbrella term rather than one defined intervention.

Banaji and Greenwald
The Associations Are Real; the Diagnosis Is Not

Mahzarin Banaji and Anthony Greenwald established that people hold measurable associations they do not report and would often reject on reflection. Their own position on the use of such measures is more careful than the training industry’s: they have cautioned against treating an individual’s score as a diagnosis of that individual or a prediction of what they will do. That caution is the starting point here. If your score cannot tell you about you, your record has to.

Nisbett and Wilson
We Report Reasons That Were Not the Cause

Richard Nisbett and Timothy Wilson argued that people confidently give accounts of why they judged as they did which demonstrably were not the operative causes — and that the accounts are constructed from plausible theories rather than retrieved from the process. The consequence for this subject is severe: asking somebody why they chose as they did produces a reason, reliably, and the reason is not evidence.

Argyris
The Rule You Espouse Is Not the Rule You Use

Chris Argyris distinguished the theory somebody espouses from the theory they actually use, and observed that people are generally unaware of the gap and skilled at defending it when it is pointed out. He treated the gap itself as the object of study rather than as a failure of sincerity, which is exactly the posture this method requires. Nobody is being caught out; a discrepancy is simply the normal condition.

Kahneman, Sibony and Sunstein
One Decision Shows Nothing

Daniel Kahneman, Olivier Sibony and Cass Sunstein argued that unwanted variability in judgement — between decision-makers, and in the same decision-maker across occasions — is a large source of error, and that it is invisible in any single case and only appears across a set. This is why the method asks for twenty. A single decision can always be justified; twenty have a pattern, and the pattern is the finding.

Associations
Confabulation
The gap
Aggregation
Repair

The recurring pattern matters more than the labels. Across every serious treatment of this subject, the same five themes reappear: the associations are real, our accounts of our own reasoning are not reliable, there is a gap between the rule espoused and the rule used, the gap is only visible across a set of cases, and the only honest responses are to change the practice or to change the stated rule.

This also explains why the advice in this area so often conflicts. Guidance written for awareness, guidance written for procedural design and guidance written for representation are answering different questions — and only the second has a reliable record of changing what anybody does. A useful workplace framework should therefore identify observable behaviour and business context—not merely raise consciousness.

The five stages of the Knowles Own-Record Audit

01

1. Say the rule

This stage produces the standard everything else is measured against, and it has to be written before any records are retrieved. The rule is not a value and not an aspiration. It is the specific thing you believe you do: who you put forward for the interesting work, what you weigh in a shortlist, how you decide who speaks first.

It includes:

  • Writing it as a rule, not as a value.
  • Making it specific enough to score a case against.
  • Writing it before looking at any records.
  • Keeping it to one decision type, not your whole conduct.
  • Stating what would count as departing from it.
  • Resisting the urge to write the rule you wish you applied.
  • Dating it, so the comparison is honest.

An employee demonstrates this capability when they can produce one written sentence, dated before any audit, that a colleague could use to mark twenty cases without asking them any questions.

For example, a manager may write: “I give the client-facing work to whoever is next in the rotation, unless the client has asked for somebody specific.” That is scoreable. “I try to be fair about opportunities” is not, and is the version almost everybody writes first.

02

2. Pull twenty

Retrieve twenty of your own actual past decisions of that type. From the calendar, the allocation sheet, the email record, the system — anywhere except memory.

Twenty is the number because one decision can always be justified, and twenty cannot all be exceptional.

Pulling them means:

  • Taking the last twenty in sequence, not a selection.
  • Retrieving them from a record rather than from recollection.
  • Including the ones you already know will look bad.
  • Recording what was decided, not why.
  • Noting the alternatives that were available at the time.
  • Stopping at twenty rather than hunting for a better sample.
  • Refusing to annotate any of them yet.
  • Taking the last twenty in sequence is the whole discipline of this stage. A sample chosen by the person being audited is not a sample, and the selection will be made on exactly the associations the audit was meant to surface.

The instruction not to record reasons is deliberate. A reason supplied now is constructed now, and it will be constructed to fit the rule you wrote down twenty minutes ago.

The result is twenty rows of what happened. It is a dull document and it is the only evidence in this method.

03

3. Score them

Mark each of the twenty against the stated rule: followed, departed from, or not applicable. No commentary, no explanation, and ideally not by you.

They include:

  • Marking each case against the written rule only.
  • Having somebody else score them where that is possible.
  • Scoring departure as departure even where it was justified.
  • Keeping justified departures as a separate count.
  • Resisting the reclassification of awkward cases.
  • Counting the total rather than discussing individual rows.
  • Writing the two numbers down before interpreting anything.

Having a colleague score the rows is worth real effort, because the scoring is where the audit is usually lost. A person marking their own cases knows the circumstances of each one, and every departure acquires a reason as it is read.

For example, the manager with the rotation rule may find eleven of twenty followed it, six departed with a recorded client request, and three departed with nothing recorded at all. The three are the finding. They are not a scandal; they are the gap, and until the rows were counted there were none.

04

4. Face the gap

The discrepancy rate is the result. This stage asks for one thing and it is not analysis: accept the number as a fact about your own practice before accounting for it.

It includes:

  • Stating the discrepancy as a number, out loud or in writing.
  • Not explaining the individual cases at this point.
  • Noticing which direction the unexplained departures ran in.
  • Accepting that the pattern is yours regardless of intent.
  • Declining to compare yourself favourably with anybody else.
  • Declining to conclude anything about your character.
  • Keeping the number where you will see it again.
  • Somebody facing a gap asks:
?
?
?

What is the number?
Which way did the unexplained ones run?
Would I have predicted this yesterday?
What am I about to say to make it smaller?
That last question is the one that matters, and the answer is usually available within seconds. The defence arrives faster than the finding, it is sincere, and Argyris’s point is that being sincere is not the same as being accurate about yourself.

It is worth being clear what this stage does not establish. A discrepancy is not proof of prejudice, and this method cannot tell anybody what their associations are. It establishes something narrower and more useful: that the rule you believed you applied is not the rule your record shows, by a stated amount, in a stated direction.

05

5. Fix the rule

There are exactly two honest responses, and both are available. Bring the practice to the rule, or amend the rule to describe what you actually do and defend that in the open.

Fixing means:

  • Choosing between changing the practice and changing the rule.
  • Writing the amended version down, dated.
  • Removing the discretion where the practice will not hold.
  • Telling whoever is affected what the rule now is.
  • Re-auditing after the next twenty, not after the next one.
  • Accepting a smaller gap rather than demanding none.
  • Declining the third option, which is to leave both as they were.
“

An audited rule you have amended and can defend is worth more than an admirable rule your record does not support. The second is not a standard; it is a sentence.

This fifth stage is what separates the method from being told about bias. The first four stages produce an uncomfortable number, which on its own changes nothing and in some people produces a durable sense of having already done the work. Only the fifth stage touches a decision, and where a practice will not hold, the honest fix is usually to take the discretion out rather than to resolve to try harder.

How the five stages work together

The stages run in order, and the fourth is where the sequence usually stops if it is going to.

Consider an illustrative Singapore workplace scenario: a department head believes she allocates stretch assignments by readiness, and has been told by two people that the same handful of names always get them.

The team must:

01


Say the rule:

write down that stretch work goes to whoever is next ready, measured by their last completed assignment.

02


Pull twenty:

retrieve the last twenty allocations from the project log, in sequence, without selecting any.

03


Score them:

have a peer mark each against the written rule, with no explanations attached.

04


Face the gap:

accept that seven of the twenty departed from the rule with nothing recorded, and that six of those seven ran the same way.

05


Fix the rule:

either allocate strictly by the readiness measure, or state openly that proximity also counts and let that be examined.

A session explaining how unconscious associations work would address one part of the situation. A statement of commitment to fairness would address another.

The integrated method shows why two years of good intentions had changed nothing measurable: the belief was sincere, the record had never been counted, and nobody can correct a pattern they have not yet been shown to have.

Further worked examples appear in the companion guide on identifying the gap between your stated rule and your last twenty decisions.

Is this the same as unconscious bias training?

No. The terms overlap, but they are not identical.

Term

Primary emphasis

Examples

Awareness training

Understanding that unconscious associations exist

Concepts, examples, discussion, a shared vocabulary

Implicit association testing

Measuring associations at group level

Reaction-time measures, aggregate comparisons

Blind processes

Removing identifying information from a decision

Anonymised sifting, screened auditions, redacted papers

Own-record audit

Finding the gap in one person’s actual practice

A written rule, twenty cases, a score, a number, a repair

Researchers have a name for the confusion these overlapping labels create. The US Institute of Education Sciences calls it the jangle fallacy: assuming two things are different because they carry different names. “Unconscious bias,” “implicit bias,” “subconscious bias,” “hidden bias” and “cognitive bias” describe substantially overlapping territory with different emphasis, and the last of them is a genuinely wider category.

The practical consequence is twofold. Awareness training reliably delivers a shared vocabulary and a shared factual baseline, which is a real and underrated outcome: it makes a later procedural change discussable without the change reading as an accusation. What it does not reliably deliver is a change in what anybody decides.

Conversely, this method should not be applied where the exposure is not in anybody’s discretion. A shortlist can be audited, scored and repaired perfectly and still be drawn from a pool nobody widened; a promotion rule can be followed exactly and still encode something decided years ago. An own-record audit finds the gap between your rule and your practice. It cannot find a problem with the rule everybody is following faithfully.

An understanding of how associations form, for example, remains genuinely useful, and is what stops a discrepancy being read as an accusation of prejudice. This method supplies the evidence. The two are complementary, not competing.

For a fuller decision guide, see record audits vs awareness training, implicit association tests and blind processes.

8

How does this relate to Singapore’s Critical Core Skills?

SkillsFuture Singapore identifies 16 Critical Core Skills that support workers’ ability to develop and use technical competencies. Ten reach a stage of this method. One works against the stage it is nearest to, and that is the useful part:

Thinking Critically

Creative Thinking, Decision Making, Problem Solving, Sense Making and Transdisciplinary Thinking.

Interacting with Others

Building Inclusivity, Collaboration, Communication, Customer Orientation, Developing People and Influence.

Staying Relevant

Adaptability, Digital Fluency, Global Perspective, Learning Agility and Self Management.

Conceptual overlap — not an official equivalence

Say the Rule→ Communication

Face the Gap→ Self Management

Fix the Rule→ Decision Making

The overlap is real, and one tension deserves stating plainly: Learning Agility is the framework’s nearest skill to the fourth stage and it assumes the thing the fourth stage denies. Learning Agility describes drawing lessons from experience — from experience as the person recalls it, which is precisely what Nisbett and Wilson showed to be constructed rather than retrieved. The frameworks should not be presented as interchangeable. SkillsFuture’s Critical Core Skills are an official Singapore framework; the five-stage structure is a simplified editorial model.

This crosswalk is not a claim that a Knowles Training Institute programme is SSG-accredited, WSQ-approved or SkillsFuture-funded.

Why does this matter now?

More allocation and selection decisions leave a complete machine-readable trail, which makes an own-record audit cheap for the first time — and more of those decisions are now assisted by systems trained on the records nobody had audited.

World Economic Forum 2025 employer survey

Analytical thinking was the most widely cited core skillIdentified as essential by around seven in ten surveyed employers

Employers also gave prominence to:

Resilience, flexibility and agility
Leadership and social influence
Technological literacy
Creative thinking
Curiosity and lifelong learning

International employer survey — not a Singapore-specific forecast.

Singapore’s Critical Core Skills similarly combine transferable cognitive, interpersonal, adaptive and digital capabilities with technical competence.

A session on how associations form will not tell you that seven of your last twenty allocations departed from your own stated rule.

9

Can this be learned?

Yes, and the arithmetic is trivial — but accepting a number about your own practice without immediately accounting for it is not something most people manage alone. Development normally requires a cycle rather than a single event:

1

Relevant models and worked audits

2

Practice on the participant’s own real records

3

Feedback on the score, not on intent

4

Reflection on what was said to make the gap smaller

5

A second audit of twenty later decisions

6

Support from managers and colleagues

Evidence from Singapore’s Institute for Adult Learning

762worker-learners

43educators and facilitators

8organisations

The Institute for Adult Learning’s Adult Learning Collaboratory examined future-oriented teaching methods across 762 worker-learners, 43 educators and facilitators, and 8 organisations, studying 86 learning interventions.

The work focused on emergent scenarios — which ask learners to respond to situations that evolve, change or remain uncertain — and generative dialogue, which asks learners to challenge assumptions and build understanding collectively rather than absorbing information individually.

The findings do not prove that every course using scenarios will produce the same result. They support contextual practice, dialogue and reflection over information delivery alone.

Which decisions deserve this treatment?

Apply it to recurring decisions you make with discretion and that leave a record—not automatically to everything. The method cannot help where everybody is following the rule faithfully and the rule is the problem, and it cannot tell anybody what their associations are. It finds one thing: the distance between your stated rule and your own last twenty cases.

1

Do you make this kind of decision repeatedly?

2

Is there a record of the last twenty?

3

Do you have genuine discretion at the moment of deciding?

4

Could somebody else score the cases against a written rule?

Example: the same few names get the interesting work

The gap may be seven unrecorded departures from a rule nobody had written down—not ‘bias’ in general.

The response may require an amended rule, a removed discretion, a conversation somebody has been avoiding, or a second audit in six months.

10

Common mistakes in working on unconscious bias

1. Looking inwards for the answer

Introspection does not reach an unconscious association, and the attempt produces either reassurance or guilt. Both feel like progress and neither is evidence about any decision you have made.

2. Using a test score as a personal diagnosis

The researchers who developed these measures have cautioned against this specifically. An aggregate instrument can establish that associations exist in a population without telling any individual what they will do on Thursday.

3. Choosing which cases to audit

A sample selected by the person being audited is not a sample. Take the last twenty in sequence, because any other method of choosing will be shaped by the thing you were trying to see.

4. Recording reasons alongside the cases

A reason written now is constructed now, to fit the rule you wrote down earlier. Record what was decided; leave why for after the score exists.

5. Scoring your own rows

You know the circumstances of every case, so each departure acquires a justification as you read it. A colleague scoring against a written rule reaches a different and more useful number.

6. Measuring attitudes instead of decisions

A survey of how people feel about fairness records a mood. Counting departures from a stated rule records what happened, which is the only thing anybody can be asked to change.

7. Stopping at the number

An uncomfortable finding that produces no amended rule and no removed discretion leaves the practice exactly as it was, and often leaves the person convinced the work has been done.

Turning the method into a working habit

A practical cycle contains six steps:

1Write the rule downOne scoreable sentence, dated, before any records are opened.
2Pull the last twentyIn sequence, from a system, with no selection and no reasons.
3Have somebody else score themFollowed, departed from, or not applicable — nothing more.
6State the numberSay the discrepancy out loud before explaining any of it.
5Choose which to changeAmend the practice, or amend the rule and defend it openly.
4Re-audit after twenty moreCompare the two numbers rather than the two intentions.

This cycle prevents the organisation from treating bias as a matter of consciousness. It also creates a clearer basis for deciding whether a discretion should exist at all.

Organisations that decide facilitated development is the right response can review the Understanding Unconscious Bias Training Course in Singapore, a commercial programme option for facilitated practice across the five stages.

Course format, facilitator suitability, customisation and expected workplace application should be evaluated against the organisation’s identified needs.

Frequently asked questions

?
What is an own-record audit?

It is the comparison of a written rule somebody believes they apply against their own last twenty actual decisions of that kind, scored by somebody else. It is distinct from awareness training, which explains how associations work, and from implicit association testing, which measures associations rather than decisions.

?
Why twenty?

Because one decision can always be justified and twenty cannot all be exceptional. A pattern in judgement is invisible in any single case and only appears across a set, which is why the number matters more than the care taken over any individual row.

?
Does a gap prove somebody is prejudiced?

No, and the method should not be sold as though it did. It establishes that a stated rule and an actual record diverge, by an amount and in a direction. That is a fact about a practice rather than a verdict on a person, and it is actionable in a way a verdict never is.

?
What if there is no record to audit?

Then start one, and the audit comes later. This is a common and unglamorous finding: the decisions that carry the most discretion are often the least recorded, precisely because they were never treated as decisions. Recording them is usually the single most useful thing an organisation can do here.

?
Does this apply to decisions made with AI assistance?

It applies more there, not less. A system trained on an organisation’s past decisions reproduces the pattern in those decisions, including the part nobody had counted. An own-record audit of the period before the system was introduced is the only baseline against which its output can be read.

Conclusion

Unconscious bias is best approached as an evidence problem rather than a consciousness problem.

People need to write down the rule they believe they follow, retrieve their own last twenty cases from a record, have somebody else score them, state the discrepancy before accounting for it, and then change either the practice or the rule. An understanding of how associations form remains valuable, and it is what keeps a number from being heard as an accusation.

For employers, the first step is not to book an awareness session. It is to pick one discretionary decision, ask whoever makes it to write down the rule, and count the last twenty.