Overcoming hurdles in automated underwriting
Technological advances are progressing the automation of nearly all steps in the underwriting process from pricing to claims handling, and all the way to portfolio analysis and steering. Associated challenges need to be properly managed.
Article information and share options
Automation can make underwriting processes easier, more efficient and effective
Electronic business platforms with advanced data analytics, or new algorithms that digest large swaths of data, can provide swift pricing, in particular for standard insurance business. At the same time, automation may help to apply more complex models in predicting future claims, yielding more efficient risk pricing. In addition, such automation may lower costs, thereby also improving consumer access to insurance across markets.
Challenges to implementation remain
Common challenges in automation are potential flaws and biases. For instance, automated personalised engines can help customers understand what they are buying, but the engine may categorise the customers wrongly. This can often be traced back to biases in the training data used to build the algorithms (ie, the initial data used to train models).1 Curating the input data carefully to flag such biases could be one way to go about it, with an emphasis on data quality rather than quantity.
However, if machine learning and other complex algorithms are involved, it becomes more difficult to understand and explain the model’s actual workings and output. Such automation will also obscure individual accountability. Whereas in the past a person might have clearly been in a specific role with associated responsibilities, the higher reliance on algorithms makes the attribution of mistakes to algorithms, the companies which order them, or their users more challenging.
From an operational perspective, an important factor in the implementation of automated underwriting is governance and organisational framework. For a long while already, many large IT projects have been in the headlines for delays and being over-budget.2
Oversight and managing limitations will be key to success
More recently, government decision making based on automated systems has been in the spotlight.3 This points to the important role of oversight, and of requiring human intervention in key parts of the process. Carefully controlling the degree of disintermediation of humans, most notably by letting human underwriters oversee part of the process, can help reduce the risk, while remaining cognizant that humans have biases too. Re/insurers need to ensure that humans still have the skills and ability to override machines’ decisions, and also to perform complex underwriting tasks for which automation is not (yet) fit for purpose.
Further Information
References
1 “AI Regulation is Coming,” Harvard Business Review, September-October 2021.
2 Humphrey, S.W., “Why Big Software Projects Fail: The 12 Key Questions”. In: Software Management, Reifer, D.J. (ed), IEEE Computer Society, 2005. Office scandal: What the Horizon saga is all about. BBC, Jul 22 (last consulted on 31.01.2022).
3 For an example of such a scandal, see two recent ones: At the beginning of 2021, the Dutch government resigned following thousands of families wronged by social benefits, and despite a parliamentary report flagging the shortcomings. Mid 2021, a UK judge overturned the conviction of 59 post masters accused of stealing money when in fact, the reporting software for accounting was full of bugs, with the judge stating that their conviction was “an affront to the public conscience”. See “Dutch government faces collapse over child benefits scandal,” The Guardian, Jan 14 2021; “Post Office scandal: What the Horizon saga is all about,” BBC News, Jul 22 2022.
