Algorithms for Insurance Age

Everything you were afraid to ask about algorithms

In case you hadn’t noticed, algorithms are making news. This summer’s row about examination grades saw students take to the streets to protest against an algorithm that, they argued, had downgraded their GCSE and A-level results. In the US in 2016, there was an outcry when an investigation found that an algorithm used to guide judges on sentencing was attributing a much higher chance of reoffending to black subjects than white subjects.

All of this may seem a long way from insurance broking, but if you work for a reasonable-sized broker, by the time you’ve typed your key code into the office entry system, used the coffee machine, looked at Google news and logged onto your broking system, you’ve already relied on dozens of algorithms, all of them working behind the scenes – and there are more coming. As technology advances, algorithms become more commonplace. So, by default, brokers and MGAs become more reliant on them.

Back to basics

To understand algorithms and their impact on businesses, there are a couple of basic misconceptions that need to be addressed. First, an algorithm is not Skynet in The Terminator. It will not rise to consciousness and turn on its broker creators. An algorithm is simply a series of machine-led instructions to carry out a process such as calculating a price. Once upon a time, we would have called these instructions ‘a simple computer programme’; now we call it an algorithm. Even an algorithm with the potential to ‘learn’ does not learn in the same was as a human. It still follows instructions. An example of a machine-learning algorithm might be one processing large amounts of claims data for trends which can then be fed back into product pricing models to help eliminate unwanted risks.

Second, algorithms are not new: it could be argued that much of the blame for the financial crash of 2007 can be pinned on them. Where trading algorithm gave a hedge fund a strong rating, other algorithms bought into the fund, which in turn prompted the first algorithm to buy more, and so on, creating a feedback loop that led to overpriced stock and a flawed assessment of risk. The rest, as they say, is history.

Does not compute

The key thing for a broker to understand is that an algorithm is only as good as its programming and the data it’s fed. It’s the programmer who creates the algorithm’s internal logic and instructs it what weight to place on one data set versus another. In the case of this summer’s exams, too much emphasis was placed on socio-economic factors and schools’ past performance, hence the algorithm’s skewed results.

Imagine you’re presenting a risk to an underwriter who’s using an algorithm-driven system to generate a rate. From your first-hand knowledge of the client, you consider the risk to be attractive, however the underwriter quotes a premium higher than expected. This could be the result of either flawed data or the algorithm focusing too much on something like a postcode.

Testing is the way to iron out problems – ideally automated testing in which different combinations of data in huge volumes are fed into the algorithm to see the range of results produced. If it’s a pricing algorithm, then a human will need to sift through results looking for exceptions, although this creates room for further human error. The greater the volume of data processed, the greater the potential for a human operator to miss an exception.

Widening usage

The most common use of algorithms in insurance is in the calculation of rates and premiums. But already major broking houses are looking to deploy them in other areas. Several large firms are building intelligent algorithms to be plugged into customer relationship management systems. These algorithms will search customer data looking for cross-selling opportunities and extracting trends in consumer behaviour. Others will analyse customer interaction with chat-bots to streamline customer service models.

Claims management is an area of focus for insurers. Algorithms are increasingly used to assess the level of reserves needed for any given claim based on the claim data. At the extreme end of the spectrum, insuretech Lemonade, is actually paying claims by algorithm without any human interaction.

Algorithms will increasingly control and automate large swathes of the insurance industry. Today, the challenge for brokers is not to be a slave to them, but to utilise them efficiently while recognising that they may need to explain their workings to a regulator. Algorithms may seem shrouded in mystery, but the results of their activity can be all too public. Just ask a few hundred-thousand students.

Author: Jim Campbell, Business Development Director, Covernet

ENDS