We learn from our mistakes

Not just bias: silencing noise for fairer and more effective business decisions

The book by Daniel Kahneman, together with Olivier Sibony and Cass R. Sunstein, reveals how noise clouds the judgements and decisions of people and organisations

by Alessandra Cattani *

28 August 2024

Aggiungi Il Sole 24 Ore
ai preferiti su Google

4' min read

Two summers ago in a whatsapp group, a colleague shared the photo of the cover of Rumour, a flaw in human reasoning, the latest book by Daniel Kahneman, the first psychologist to be awarded a Nobel Prize in Economics for his theories contributing to our understanding of how human beings make decisions.

I rushed to buy it thinking I would read it with all the attention it deserved. But, as it was summer, I did not, caught up in lighter reading. The news of Kahneman's passing prompted me to pick up the book again and discover a perspective capable of innovating the analysis and cultural management of organisational behaviour.

It is an eye-opening read. Written with Olivier Sibony, professor of economics, and Cass R. Sunstein, legal scholar and co-author with Richard Thaler ofNudge, the book explains how the phenomenon of noise clouds the judgements of people and organisations and what to do about it.

After more than 50 years of studying biases and the way they lead to errors in evaluation and judgement, known as cognitive biases, Kahneman became aware of another type of error, something he had not thought of before, namely noise. It follows that to the impact of biases on decision-making, already at the centre of managerial debate, is added that of noise.

What is the difference between bias and noise?

Bias determines erroneous judgements that, in most cases, systematically go in the same direction, e.g. when a company's managers are always over-optimistic about turnover growth forecasts. Conversely, noise determines errors that go in different directions. An example that helps to understand the difference is the scales: if it consistently underestimates the real weight by 1 kg, it is biased but without noise; if it marks a different weight each time, it is noisy.

Noise is therefore the unwanted variability in professional judgements. The key to the definition lies in the adjective 'undesirable'. Sometimes, in fact, variability in judgements is not a problem, indeed, it is even desirable. But not when it comes to professional judgement. If two doctors make two different diagnoses of the same patient on the basis of the same examinations, at least one of them must be wrong. The same if two judges assign different sentences to perpetrators of the same crime, or two insurance experts settle two different sums for the same claim. The issue becomes even more complex if it is the same insurance adjuster who settles different sums in relation to the temperature of the environment, the mood of the day, or the successes or failures of his sports team. These are judgements where variability is undesirable: there is a correct answer and we would like the answer to be the same.

Where is the noise most commonly found?

Wherever human judgement is exercised, noise is found. In the corporate world, the noise rate is outrageously high, even in situations that would require great accuracy of judgement. Examples? Forecasts of sales of a service, of turnover growth, of the time it takes to develop a product. Another surprising example are performance evaluations: it seems that only a quarter of the evaluations are related to actual performance, the other three quarters to noise. Whether it is level noise, i.e. that some evaluators are on average more generous than others, or occasional noise, i.e. that one day the evaluator may be in a better disposition than on other days, these phenomena together make up about three quarters of the performance evaluation.

Can noise be reduced? And are there methods to do so?

The good news is that yes, it is feasible and specific measures can be taken. The bad news is that noise is hardly predictable and not easily visible or explainable; therefore, it is often neglected even when it causes major damage.

Usually, when discussing reducing errors in organisations, it is the bias that takes centre stage: the focus is on identifying the most common biases and how to counteract them. For example, if projects are consistently behind schedule, it could be because one is too optimistic about deadlines. Once the cause is identified, the error can be corrected. But errors do not always result from bias. In most decisions there is no such obvious directional error, but many different types of errors. That is to say, there is noise. It follows that eliminating, or at least reducing, bias from a set of judgements is possible. Reducing the noise, on the other hand, is very difficult.

The strategy to correct the error cannot therefore be to identify the cause. What is needed is to implement a series of decision hygiene measures, i.e. specific procedures designed to prevent an unspecified series of potential errors before they occur. It is similar to when we wash our hands: we do not know exactly which germs we are avoiding, but we know that it is an effective preventive practice against various germs. Similarly, we may never know what injury or error we have avoided by applying decision-making hygiene, but given that noise is an invisible adversary, and given the damage it can cause, it is still worth fighting it.

Two examples of decision hygiene applicable in organisations

A good way of containing the noise is to aggregate several independent judgements: instead of assigning the judgement to one person, or putting several people around the table to talk about it, we can ask them to make a judgement independently and then aggregate all the collected judgements. The point is to avoid the noise of group dynamics: who speaks first, who speaks with more conviction, who with more eloquent expressions or gestures.

Another good measure of decision hygiene that we can adopt is to prefer relative judgements and scales to evaluations on an absolute scale. For example, in performance evaluations, to say that a person is a '2' or a '3' with respect to a skill remains quite subjective, even when one has a description of specific behaviours. But if one asks: Is Tizia's ability to connect with customers better than Caio's? the answers of different evaluators are more likely to converge. Relative judgements tend in general to be less noisy than absolute ones.

*Partner of Newton SpA