What is decision hygiene?

Decision hygiene is the set of procedures that reduce error in judgement without knowing which judgements were wrong, in the way handwashing prevents infection without identifying the pathogen. It is the structural answer to variability, and it works because it asks nobody to think differently.

Why hygiene rather than correction#

Bias has a direction, so it can in principle be corrected. Noise has no direction: the same case decided by two qualified people produces two answers, and neither person experiences themselves as varying. Nobody can tell you which of their decisions were the noisy ones, including afterwards, which rules out any remedy that depends on identifying the error first.

That is also why debiasing training mostly fails and hygiene mostly does not. One requires a person to notice something they cannot notice. The other changes the procedure and leaves the person alone. See do debiasing programmes work.

The practices#

• Structure the decision into separate assessments. Break the judgement into components, assess each on its own evidence, and hold off the overall view until the end. This is the Mediating Assessments Protocol, and the holding off is the part that gets dropped.
• Judge independently, then aggregate. The single most reliable intervention available, and it requires only that the order be reversed. See collective judgement.
• Use a common relative scale. Ratings mean different things to different raters unless everybody is anchored on the same cases. Ranking against shared reference points removes a large slice of noise for no cost.
• Sequence the information. Early impressions contaminate later assessments, so withhold what is not needed yet. An assessor who knows the outcome cannot unknow it, and an interviewer who has read the CV is no longer scoring the interview.
• Use a rule where a rule does as well. The clinical versus actuarial evidence says consistency is what the rule contributes, and consistency is the thing noise destroys.

Finding the noise first#

Hygiene is worth spending on in proportion to how much noise exists, and organisations are consistently wrong about that. The measurement is a noise audit: give the same set of real cases to the people who normally decide them, separately, and compare. It is the one measurement of judgement quality that can be run without waiting for outcomes, and the results are usually larger than the people involved expect.

The machine version, and its cost#

An automated system is perfectly consistent. Replacing a noisy human process with a rule or a model removes noise outright, which is a real hygiene gain and the strongest available argument for automating certain decisions.

It also preserves bias exactly. Noise is variability around a point, bias is where the point sits, and consistency does nothing to the second. A model that is systematically wrong in one direction is now wrong in that direction every time, without the variation that used to produce occasional right answers and occasional complaints. The hygiene gain is genuine and it should be recorded alongside what it does to the other half of the error.

What this does not establish#

The practices come from Kahneman, Sibony and Sunstein's synthesis and the underlying noise research, and no first use is claimed for any of them here. The evidence for individual practices varies: independent aggregation is well supported, structured assessment protocols are argued from practice, and the claim that a full hygiene programme improves organisational outcomes has not been tested as a programme.

Related SuperSkills research#

• What is a noise audit
• What is collective judgement
• Models of judgement
• How do you build judgement in an organisation