From voice assistants to personalised recommendations on streaming services, machine learning is a branch of artificial intelligence that many of us encounter daily and makes our lives easier.
While the protection industry has taken steps to enhance efficiency, such as with pre-underwriting tools, could machine learning have a role in making the underwriting process quicker?
“Naturally, all insurers want their new business and underwriting processes to run efficiently, delivering outputs and decisions as quickly as possible,” says Luke Harrison, head of data analytics at technology provider iPipeline.
“Insurers will strive to develop the perfect harmony between platform, data and people. That is, a rules-based underwriting system which is able to make decisions in real-time using machine learning to define parameters and controls, and an expert team that examines more complex cases/disclosures, using the benefit of their experience to make decisions.”
The current picture
Most insurers have an underwriting philosophy, which is usually interpreted into a set of underwriting rules around which companies create a question set, says Chris Monaghan, senior marketing manager at UnderwriteMe.
“The underwriting process asks a number of health and lifestyle questions and rules drive decisions from this data.
“For around 75 per cent of cases, this leads to straight through acceptance and the ability to go on risk immediately.”
At Vitality, for example, the majority of underwriting is completed using a rules engine.
John Downes, director of underwriting and claims at VitalityLife, explains: “Underwriting rules engines are pieces of software that use decision trees to automate and replicate the human underwriting process.
“These rules engines have been in existence for around 20 years. They work by asking high-level application questions, the answers to which use a decision tree process to further clarify answers if necessary.
“For example, additional questions may be asked about the treatment received and the frequency of symptoms of a declared condition, the answers to which will be used to determine the acceptance terms.”
Meanwhile, applications with a more complex medical history or those for larger amounts of cover will require a human underwriter, Downes adds.
At Zurich, applications are likewise run through a rules engine that performs an initial assessment based on the responses given.
“It’s worth noting that underwriting experts have developed and shaped this automated process, which will continue to evolve,” says Nicky Bray, chief underwriter for Zurich’s UK Life Business.
“This approach means that for the majority of customers, we can offer an instant underwriting decision, meaning that they can secure cover quickly and easily.”
But in cases where customers share information with Zurich about medical concerns or histories, and where the insurer needs additional medical information, all cases are looked at individually.
“Machine learning certainly has potential to help refine our initial process, though for more complex cases, human judgement will always be needed. Especially for conditions like mental health, where the symptoms and circumstances for each customer will be very different,” says Bray.
Duncan Mosely, Guardian’s chief operating officer, says that like most others, the provider has automated underwriting that uses a rules engine. This accounts for around seven in 10 applications, the balance requiring human intervention and judgement, he says.
“Beyond our underwriting rules engine, machine learning is something we’re monitoring,” Mosely adds.
Meanwhile at MetLife, the majority of the provider’s protection range has limited medical underwriting.
“[We] recently launched MortgageSafe, which is offered with simple underwriting. Acceptance can be automated as it is rules based,” says Clare Lusted, head of propositions at MetLife UK.
As such, Lusted says the provider does not use complex models, machine learning or any other artificial intelligence as part of its medical underwriting processes.
Speeding up the human half of underwriting
When further medical evidence is required, Monaghan at UnderwriteMe says this comes back to insurers as either structured or unstructured evidence.
An example of structured evidence might be a blood test, where an insurer is looking to find out something very specific and easily measured.
Unstructured evidence might be the overall health record or a doctor’s report on a condition. This will usually be in the form of text and would need an underwriter to read, review and make a decision based on the evidence.
According to Monaghan, insurers can use natural language processing and machine learning to review both structured and unstructured data when further medical evidence is required, to reach quicker decisions and with less resource on the more complicated cases.
“The biggest reliance of third-party data for the protection industry is medical data. This is where machine learning can bring the most benefit to advisers and consumers alike.
“The technology can help read and learn from the structured medical data and can also turn unstructured data into ‘readable’ content to help the decision-making process.”
Is the protection industry ready for machine learning?
Some machine learning algorithms require vast amounts of data to perform while others work on small, curated datasets, Monaghan says.
Eoin Lyons, CEO of Opal Group, which builds technology and platforms for providers, says that while “more than enough” data is available, the major challenge is accessing it.
VitalityLife's Downes says in other markets such as the USA, there are multiple external data sources such as prescription records and clinic summaries, which can be accessed by insurers that have enabled a degree of machine learning.
But this type of access to information is not available to UK insurers, according to Downes.
Meanwhile, Mosely at Guardian says there could theoretically be a role for machine learning to be used in the future.
“Especially with the growth of electronic collection of medical evidence and certainly for test results received as part of the underwriting process.
“Ultimately the process of underwriting requires the pooling of risk, so machine learning or further automated processes could be used more widely.
“But there will always be some need for human judgement with complex scenarios that are not routinely encountered.”
A threat to those in the protection industry?
A common concern around new technology is its potential capability to replace humans in the workplace.
But Downes says it is unlikely that underwriters will ever be replaced, although the role is likely to change over time.
“Regular ongoing maintenance of any machine learning model will need to be done and an underwriting team together with a data science team would be essential for that, as well as for the ongoing underwriting philosophy development.
“Widespread deployment of such models, however, seems some way off at present.”
Harrison at iPipeline likewise says that in the longer term, the use of machine learning will likely change the focus of underwriters and how they deliver value to insurers.
“Their skills will be deployed to help drive and manage any unwanted bias present in algorithms. We’re seeing changes in the actuarial profession, with a greater focus and emphasis placed on data science. It’s hard to imagine that underwriting won’t develop in this way too.”
Bray at Zurich agrees that machine learning and the shift towards more automated ways of working is not about replacing jobs.
“Instead it’s about revolutionising existing roles, so that more time is invested in customer engagement and complex cases so we can deliver tailored solutions to meet customer needs.
“Underwriting roles have and will continue to evolve to include analytics and understanding data to develop things like machine learning. These shifts give us the capability to offer cover to more people quickly, providing them with important financial protection.”
Lyons at Opal describes an "ideal world" where data-driven and processing tasks that do not require execution by humans are automated.
“You could see an increase in people working where they are needed, at the point of engagement, ie technology and innovation, proposition development, sales.
“I have cited the travel industry as an example in the past. There was significant digital adoption for booking flights and accommodation during the 2000s. This disrupted the business model for intermediaries, ie travel agents.
“Travel agents now deliver more complex trips and higher value trips where there is margin and their expertise can actually add value – a good case study for protection.
“Automate mundane tasks, make the user journey easier, grow the market and add value where there is margin.”
Chloe Cheung is a features writer at FTAdviser
