A data-driven industry, it would seem insurance was an industry ripe for digital disruption. Yet while the potential was seen some time ago, in practice this was slow to take off, explains Naoise Harnett, partner at Pinsent Masons, who says it is "generally accepted" that insurance companies have been slow to innovate and digitise.
“The rhetoric from the insurers is that they have seen the light and will make the necessary investment,” he says. “However, insurers have typically created innovation labs which were generally operating on limited budgets and were not as productive as desired. Insurers are also fearful that regulators will punish them for unwittingly writing bad risks or because of new artificial intelligence underwriting methods rejects cover or charge inflated cover to a particular ethnic group or type of customer.” He says this can explain some of the reticence to embrace technological advances.
Despite this innovation inertia in the sector, insurance consumers expect digitisation and a streamlined easy-to-use interface with their insurer, Harnett says. “Customers are also starting to demand personalised products with tailored premiums.”
As a result, insurers have started investing heavily in artificial intelligence solutions but the general consensus is that insurers have not yet unlocked the full potential of artificial intelligence.
“The majority of applications focus on optimising existing services and processes. These efforts are already yielding tangible benefits,” says Harnett.
“However, insurers are lagging behind in leveraging artificial intelligence to discover new insights in operations and customer interactions. Therefore, while artificial intelligence may prove to be a game changer we have yet to see its full potential in the insurance sector.”
Irish companies are lagging behind, he says and they are not embracing insurtech in a way insurance companies in Asia or other European markets such as the United Kingdom or Germany are. "There are pockets of engagement by Irish insurance companies with insurtechs but the levels of adoption are not the same in those other jurisdictions mentioned," he says, adding that there is significant scope in the Irish market to further adopt and embrace artificial intelligence subject to compliance with GDPR and other applicable laws.
‘Waking up’
Prof Cal Muckley is Professor of Operational Risk in Banking and Finance at the UCD College of Business and a Fellow at the UCD Geary Institute. He disagrees with Harnett, saying in his research across all aspects of “fintech”, he can see increased activity as the insurance industry finally get to grips with the role machine learning and artificial intelligence can play.
"I am seeing plenty of activity and companies coming together to collaborate on new solutions, while training courses are on the increase. The industry is waking up to the applications of machine learning to insurance problems specifically in the area of pricing and the area of claims management," he says, adding that a Bank of England October 2019 report on machine learning in UK financial services found there is a greater uptake of machine learning in insurance than in other areas of the financial services.
“That isn’t terribly surprising given that insurance has always been about data – they were pretty much waiting for the developments in these kinds of technologies.”
This data isn’t exactly ripe for processing, however, Muckley admits.
“Because they’ve been storing data for a long time, databases often don’t integrate well together. One very pronounced constraint in respect of the application of machine learning models is tricky legacy database systems where the data simply is very messy and difficult to integrate together so it can be inputted into a machine learning model,” he explains.
Where this can overcome, however, the potential is enormous. According to Harnett, the main areas where artificial intelligence can be applied to insurance include to effect digitised new business and underwriting processes, to power the claims management process, and to enhance the consumer interface and improve the customer experience.
“It will also allow insurers analyse their datasets in a way they have not been able to do so before to assist with pricing, claims profile and anti-fraud measures,” she says.
One example of machine learning in insurance is “decision trees”. Muckley explains that these support tools allow insurers to separate customer into different categories – by age, gender etc.
“This is extremely helpful because it allows them to discriminate across customers in terms of pricing and in terms of claims management. It’s about trying to learn which of these customers are most alike in respect of predicting out-of-sample claims on the policies that have been underwritten,” he said.
What machine learning allows for is ensembles of decision trees – the same tree, but tweaked a bit and then repeated hundreds or thousands of times. “Looking at lots of trees at once really helps in terms of much more accurate out-of-sample predictions.”
While decision trees have been around since the 1980s, ensembles and a so-called “random forest approach” – many ensembles of decision trees – have been enabled by disruptive technologies.
“Old ways of doing things have been advanced and modernised by the new capability of data processing by machine learning and AI,” explains Muckley. “Earlier models were accurate but random forest approaches delivered much better accuracy, and we can see this in a very clear-cut way. When we go in and we do work, we take the historical data and the predictions made by the old approach and we revisit all that data and make new predictions using the new approach and we invariably outperform the old predictions. It is a massive step-change in the quality of predictions and machine learning certainly does give us a better grip on these things, then we can make better decisions on what the insurance policy should cost up front.”