Skip to main content

Noisy underwriting? It’s when you give two underwriters a less than straightforward case to assess and they arrive at different decisions – maybe very different. Or if you asked an underwriter to assess a risk that he or she processed a while back, and the decision second time round is different from the first.

Why might underwriting decisions might vary between individuals or within the work of the same underwriter? Different skill levels? Different intelligence levels?  Variations in mood? (Might Monday morning decisions differ from Friday afternoon ones? Or be influenced by a recent argument with a spouse/partner or friend or colleague?) Variations in training received? Recent experience, like having seen an early claim involving a particular risk type? Bias, such as when an underwriter views a risk, maybe a lifestyle one like alcohol or cannabis, in a way based on values that are different from colleagues’?

The answer is, basically, all of the above.

This variability in decision-making is termed ‘noise’ by three eminent academics who recently published a book of the same name.1 One of them, Daniel Kahneman, is the Nobel Prize-winning author of the seminal behavioural economics work Thinking, Fast and Slow.2 Another, Cass Sunstein, co-authored another best-selling book, Nudge,3 which discusses how different ways of presenting choices can improve decision-making and its consequences.

A degree of ‘noise’ is often inevitable and readily acceptable, but there are situations in which high degrees of variability occur and with potentially far-reaching results. For example, medicine is noisy: doctors frequently disagree on diagnoses and, indeed, even on what is the correct path to making a diagnosis. Interpretations of lab tests, X-rays and other scans vary. These noisy situations can lead to mis-diagnoses, missed diagnoses, inappropriate or unnecessary treatment and even early deaths. Court sentencing is noisy, even if there are guidelines for judges. Lawyers and their clients may be disappointed (or pleased) that their case will be heard by Judge X.

There are other examples, and one of them is insurance underwriting. The authors of Noise actually use this as a case study, although the insurance in question is liability cover, underwritten by groups of underwriters located in several centres in the United States. There was a remarkably wide variation in individual underwriters’ decisions: in an experiment, members of the underwriting team were each given a complex but typical case and asked to come up with an appropriate premium for the risk. The decisions ranged from under US$145,000 to over US$255,000; in fact, half of those decisions were outside that numerical range. The insurer’s executive team found that decision variability was five times greater than they had thought and that it was costing hundreds of millions of dollars.

The spectrum of liability risks is wider than those in life and disability underwriting, and so those assessing and pricing such cases are likely not to have the sort of detailed guidelines provided in the underwriting manuals that we are all familiar with; thus the scope for decision noise is much greater and hence the huge variability in the risk pricing decisions.

In life and disability insurance, accepting that there is probably less variability compared with liability underwriting decisions, does noise matter, especially if the portfolio is not producing mortality/morbidity losses? Is it OK that over-pricing is pretty much balanced by under-pricing? Well, surely it does. Consider equity between policyholders. Customers deserve to be charged the right price for the risk. And over-pricing may result in applicants going ahead with a rival insurer, or not going ahead at all; being put off buying necessary cover is a loss for all parties involved.

In keenly competitive situations involving substandard or unusual risks there is always the chance that if a case is won from rivals, then the successful insurer has made a mistake in its assessment and pricing. Reducing the amount of noise potentially would reduce the likelihood and size of any losses in respect of the cases won.

Maybe more worrying is when two or more underwriting centres consistently differ in their decisions. Consumers applying to an individual insurer deserve equal and fair treatment regardless of which underwriting centre their application lands up in. This is a real issue where underwriters work from home and without colleagues at adjacent desks to interact with, to question and from whom to get second opinions. After all, it has been known for underwriting units on either side of the same office to differ overall in their approach to a certain type of risk.

There is an international angle in this too: an insurer’s underwriting philosophy and practice in the US and in India may be very different, and justifiably so given the big contrast between those market environments. But if that company’s underwriters in the UK and Australia had diverging opinions on similar risks, would that not be something worth looking into?

Hopefully, variability in decision-making is reduced by regular internal auditing of a sample of cases. But then the auditing underwriter is a human, with all that entails… And how often is underwriting in two or more geographical areas compared?

Reducing noise should improve profitability and, arguably just as important, improve outcomes for customers. So how to address it?

Firstly, find out how much noise there is. Do a ‘noise audit’, which will lead in turn to where the noise is coming from, how it arises and what the consequences are. Then consider what to do about it. Is it about training, personnel selection, communication, managing teams or provision of underwriting tools and guidelines? Or…? And what would be the costs of remedial actions and how would the benefits stack up? Maybe personal bias is one of the most difficult challenges, because overcoming it depends on at least reducing the influence of an individual’s values and beliefs, some of which may have become ingrained over many years.

This discussion highlights how fallible humans are and, implicitly, the benefits of automated underwriting that enables decisions to be made cheaply, in a split-second, time and time again on identical risks. But, at least for the time being, underwriting engines are for processing the unexceptional risks that make up the bulk of life and disability insurance applications, whereas humans are better (innate variability notwithstanding) at assessing the complicated risks that form the minority.

However, artificial intelligence (AI) has arrived in the world of underwriting and will march farther into our domain. AI will learn from previous case outcomes – even in respect of complex risks – and recommend decisions to human underwriters. Importantly, AI has the virtue of processing information in a dispassionate way. For example, it won’t be influenced by the way that risk information is presented in a physician or hospital report – maybe sympathetically versus unsympathetically, or positively versus negatively (due to physician bias?) – thus eliminating another source of decision variability.

In time the decisions on the majority of the more complex cases will be automated as well, with a consequent improvement in underwriting consistency and a reduction in ‘noise’. Maybe AI won’t entirely eliminate noise, but the residual amount will be minimal. Automation with AI is the future, and don’t under-estimate the speed at which that future is approaching. In the meantime, have a think about how ‘noisy’ your underwriting operation is, the effect on the business and what you might do about it.


  1. Kahneman D, Sibony O, Sunstein CR. Noise. London: William Collins, 2021
  1. Kahneman D. Thinking, Fast and Slow. New York: Farrar, Straus and Giroux, 2011
  1. Thaler RH, Sunstein CR. Nudge: Improving Decisions About Health, Wealth and Happiness. New York: Penguin Books, 2009