Almost a decade ago, Daniel Kahneman published ‘Thinking, Fast and Slow’ about a lifetime of research, with colleague Amos Tversky, into biases and shortcuts people take in making judgements. It was the topic of conversation for months afterwards with investors and entrepreneurs as they discussed how best to integrate the many (more than a hundred now) biases and shortcuts that reliably lead people astray.
In the years since, however, I’m not sure that decision making has appreciably approved anywhere by taking this information into account. Kahneman has been asked how much he has improved his own decisions in light of his findings – not very much at all is the answer. Biases are hard for their owners to detect and counteract in the moment as a sort of bias bias or bias blind spot. We know in the abstract about the things that might affect our decisions but can’t correct for them in the moment. In this light, Kahneman’s new book, ‘Noise’, on research into random variability in decision making elicited some trepidation. Hearing yet more about how terrible your brain is at doing fundamental things without a method to improve is a counsel of despair.
As a bit of definition, bias is where judgment is consistently and predictably clustered around a point of error whereas noise is the unpredictable and arbitrary wrongness of a judgment in any direction. Both compound to negatively affect good (or at least better) judgment. Usefully, the authors of Noise offer some helpful guidance here on improving decision making. In respect of bias, the appointment of a ‘Decision Observer’ is suggested. This is based on the observation that people, with some training, are better at recognising bias in others than themselves. Obviously, having a Decision Observer on tap critiquing and improving decisions isn’t feasible much of the time but it is potentially practical in particularly impactful decisions and for strategy adoption for boards and for individuals.
With respect to noise or scatter in judgments, a trichotomy is suggested. Level noise is exemplified by generally lenient vs severe trial judges. Take any judge on the spectrum of lenient to severe and, additionally, occasion noise comes into play – the time of day, the weather, how their team is faring, mood can all cause noise or random variability in decisions. Finally, there is the largest contributor to noisy decisions, pattern noise. Even taking into account the level noise (a lenient judge say) and the factors around occasion noise, decisions are still noisy and have a lot of scatter. This matters because the errors in each case don’t necessarily cancel out. The example of insurers with noisy underwriting giving a huge payout in one example and very little in another of a similar type of case leads to a financial loss for the insurer in the first and the loss of a customer in the second. Following from this, one method of reducing the noise within each case is to aggregate independent but informed decisions. Another is to use algorithms or rules – these perform better on judgments that are predictions about the future than individuals precisely because they are rules based and not open to bespoke details or particulars in their operation. A consequence of both these approaches is that this makes for a conflict between the desire to take each case on its own merits thereby making a unique, localised determination and the need across all similar decisions to be consistent, fair and avoid errors. It has me thinking about where the balance lies and how to achieve it.
In the latter part of Noise, the use of carefully structured decision making in hiring, as an exemplar, is discussed in depth and evaluated with respect to how well the selected candidate does in the role. The approach seems to have broad utility for improved selection among candidates of interest in many other realms such as venture investment, deciding among collaborations, weighing potential acquisitions, etc. The approach is explored in some depth and it is something I may expand on in a later post.
Overall, the authors make clear that this field of research is at an earlier stage and less developed than that on biases. Nonetheless, the map of the territory is assuredly drawn with as yet unknown terrain identified and, most importantly, possible routes through marked.
Sean Mulvany 