Knowles Knowledge Centre•Implementation

How to Build an Unconscious Bias Development Plan That Changes a Decision

What an organisation does before, during and after a bias programme determines whether anything changes. In this field the session is not merely the smallest part of the plan; on its own it has a poor record.

A development plan is not a course booking. It is a sequence that establishes which decision the pattern came from, secures the record needed to audit it, creates conditions for a rule to be written and twenty cases scored by somebody else, spaces that across real decisions, and reviews whether a second twenty looks different from the first.

Most plans fail at the second and third stages — not because the session was poor, but because no record exists to audit and nobody will score somebody else’s rows without hearing the reasons. A method whose evidence is a count cannot work where there is nothing to count.

Implementation guide•For Singapore HR and L&D teams•Updated October 2026

About this guide

Published byKnowles Training Institute
PurposeDesigning a bias plan around an auditable decision
Use afterA capability gap has been diagnosed and scoped

Why bias development plans fail

The common pattern is a well-delivered event surrounded by nothing.

No baselineWithout a count of how one decision was actually made before the intervention, there is no way to tell afterwards whether anything moved. Session evaluations fill the vacuum and answer a different question — and in this field they answer it unusually favourably, because the content is interesting.
No recordThe decisions carrying the most discretion are frequently the least recorded, precisely because nobody treated them as decisions. A plan that assumes a retrievable last twenty, and finds none, has no instrument at all.
No third-party scorerIf participants mark their own rows, every departure acquires a justification as it is read, and the count comes back close to perfect. The scoring has to be done by somebody who does not know the circumstances, and arranging that is the stage plans quietly drop.

An awareness session reliably produces a shared vocabulary. Only a counted record produces a number, and only a number can be compared with a later one.

The five-stage sequence

Each stage has an owner and a completion test. A plan missing any stage will produce well-informed people whose next twenty decisions look exactly like their last twenty.

1

Baseline — before anything is booked

Pick one decision that produced the pattern. Establish whether a rule can be written for it, whether its last twenty instances are retrievable, and whether faithful application of that rule would have removed the pattern at all. Those three answers are the baseline, and the third one quite often ends the plan.

Set the target in the same terms. “Unrecorded departures from the written rule fall between this twenty and the next” is a baseline you can return to. “Greater awareness of bias” is not.

Owner: L&D with the line manager. Complete when: one decision is named, a rule is written, the record is confirmed retrievable, and the rule’s sufficiency is tested.

2

Practice — during the session

The session should spend most of its time on participants writing their own rule and working through their own retrieved rows, not on the mechanisms of implicit cognition. Evidence from Singapore’s Institute for Adult Learning supports contextual practice, dialogue and reflection over information delivery alone.

Some explanatory content is worth including, for one specific reason: it is what stops a counted discrepancy being heard as an accusation of prejudice. That is the honest job of the awareness element, and it is a supporting role rather than the programme.

Owner: the facilitator. Complete when: each participant has a written dated rule and their own twenty rows retrieved, unscored.

3

A scorer who is not the decision-maker — the stage most often skipped

Name, before the programme, who will score whose rows. The scorer marks each row as following the rule, departing with a recorded reason, or departing with nothing recorded, working from the written rule alone and declining to hear the circumstances. The point is structural, not personal — a scorer who knows the context will score generously however rigorous they intend to be.

The single highest-value action is an agreement that the scorer does not discuss the rows with the decision-maker until both numbers are written down.

Owner: the line manager. Complete when: every participant’s twenty rows have been scored by somebody else, and the two counts are recorded.

4

Spaced application — across real work

Schedule a second audit after at least twenty further decisions, which in most cases means months rather than weeks. The comparison is the only outcome measure this method has.

Choose the same decision type, scored the same way, by the same scorer where possible. A second audit of a different decision proves nothing about the first.

Owner: the line manager, supported by L&D. Complete when: a second twenty has been scored on the same rule, and the two numbers can be put side by side.

5

Review — against the baseline, not the evaluation form

Put the two counts beside each other. The question is narrow: did unrecorded departures fall, and if the rule was amended instead, is the amended rule one the organisation is willing to state publicly?

Record what did not change as well as what did. In particular, record every case where a discretion was found not to hold and was left in place anyway, and who decided that. That is a finding about the organisation rather than about the training, and it is the most useful thing this plan can produce.

Owner: L&D. Complete when: two counts exist side by side, with any retained discretion named.

A worked plan on one page

The same sequence applied to one audited decision, with owners and timing attached.

Stage

Action

Owner

Timing

Baseline

Name the decision; write the rule; confirm the record; test whether the rule would remove the pattern

L&D + manager

Weeks 1–2

Manager brief

30 minutes: who scores whose rows, and the agreement not to discuss them first

L&D

Week 3

Practice

Facilitated session; most of the time spent writing rules and retrieving participants’ own rows

Facilitator

Week 4

Scoring agreement

Scorer and decision-maker agree the rows are marked against the rule alone, with no discussion

Manager

Week 4–5

Application 1

First audit scored and both counts written down; no amendment decided yet

Employee + manager

Week 6

Application 2

Practice or rule amended, in writing, and whoever is affected told what the rule now is

Employee + manager

Week 10

Review

Second twenty scored on the same rule by the same scorer; counts compared

L&D

Week 14

Note the shape: one day of facilitation inside a plan measured in months, because twenty more decisions take that long. The waiting is not slack; it is the instrument.

Getting the rows scored by somebody else

This stage is skipped more than any other, usually because it is asked for vaguely and lands as extra work.

Make the ask small and specific. “Review each other’s decisions” will not survive a busy quarter, and produces a sympathetic conversation when it does. “Mark these twenty against this one sentence, and do not ask me about any of them” takes twenty minutes and will.

Give the scorer the rule, not the context. They do not need to understand the team’s circumstances. They need one written sentence, twenty rows, and an explicit instruction that a departure is a departure even where it was obviously reasonable.

Check the scorer is genuinely insulated. A colleague on the same team knows why each exception was made and will score accordingly, with complete sincerity. Somebody from another function, or a peer in a different office, will not. If no such scorer can be found, the audit will overstate compliance and should be reported with that caveat.

If the rule would not remove the pattern, stop the plan

Where faithful application of the stated rule produces the same distribution — because the pool was narrow, or because the rule encodes a precondition few people meet — the honest position is that this is not a development need at all, and the work should be re-scoped to whoever owns the rule and the pipeline. Running a bias programme anyway tells a group of people that an organisational design is their personal failing, which is both inaccurate and the most damaging use this subject is put to.

Frequently asked questions

?
How long should a plan of this kind run?Long enough for twenty further decisions of the same kind to have been made and scored — which for most allocation or selection decisions means six months or more. Shorter plans have no second count, and without a second count there is no evidence of anything.
?
What should the plan measure?Two countable things: unrecorded departures from the written rule in the first twenty and in the second, scored the same way by the same person. Session evaluations and attitude surveys are useful operational data but say nothing about whether a decision changed.
?
Can this work for a large group?Yes, but it scales by decision rather than by headcount. Auditing one rule across a function is workable and usually locates the exposure; auditing two hundred individuals produces two hundred numbers and no finding.
?
What if the audit comes back clean?Then record that, and look upstream. A clean sheet against a written rule is a real result: it means the discretion is being exercised consistently, and that any remaining pattern comes from the rule, the pool or a decision further back. That is a more valuable finding than a gap, and it is reached only by having counted.
?
Do we need external facilitation to run this plan?Not necessarily. The audit itself is internal by nature, since it uses the organisation’s own records. External provision is worth considering for the explanatory element, or where no internal scorer can be found who is genuinely insulated from the circumstances — not by default.

Conclusion

A bias plan succeeds or fails on whether anything was ever counted.

1Baseline
2Practice
3An independent scorer
4Spaced application
5Review

These are not administrative wrapping around a course — they are the only part of the plan that produces evidence.

The most useful discipline is to establish that a retrievable record exists before booking anything at all.

If you cannot say where the last twenty instances of the decision are kept, the plan is not ready, and no provider will tell you, because the session can be delivered either way.

Before building a plan, confirm the exposure is in a discretion rather than in a rule using how to identify the gap between your stated rule and your last twenty decisions, and check which audience level the behaviour belongs to in the record audit matrix for employees, managers and leaders. Where facilitated practice is the right component, the Understanding Unconscious Bias Training Course in Singapore is one option to evaluate against the plan you have written — not the other way round.