Automation with a human touch: Introducing AI at insurer Hiscox

CIO Ian Penny runs through the ways AI and ML can streamline processes and reduce the drudgery of mundane tasks

Hiscox is a specialist insurer. Rather than offering generic mass market insurance, its customers are mainly small businesses that cannot obtain cover from a generalist. Hiscox custmers include photographers who use drones, and people whose business takes them away from home for more than half the year.

Hiscox is willing to take risks the other insurers aren't. This makes machine learning (ML) and artificial intelligence (AI) particularly attractive options for the firm. After all, when policies are individualised rather than general, additional work is required to calculate the premiums. Algorithmic processing can potentially take some of the strain. And because each customer's needs are unique, using technology to streamline customer services has an obvious appeal.

But it's not about replacing people with machines. Ian Penny, CIO of Hiscox, is adamant that automation technology is an aid rather than a replacement for the company's staff.

"Machine learning for us is about human assistance, about making our colleagues more productive and able to make better decisions, but ultimately the buck stops with us, not the machine," he told the audience during Computing's AI and Machine Learning Live event last week.

The company is deploying AI in a number of areas, including customer support, internal processes, and the creation and pricing of policies.

Take the last example. One of the products Hiscox offers is cyber insurance. Unlike, say, flood or shipping insurance, cyber is not backed up by decades-worth of data and actuarial charts. AI and ML can be used to fill in gaps in this sparse data as well as identifying common traits. It can also enable continuous improvement, ensuring products and services adapt to fit the market as it changes.

AI can be used to find gaps in the market, too, and to recommend policies based on profiles - 'you bought this, you may also like that'. However, Penny insisted Hiscox is not going with the full retail model.

"We are not looking to be the Amazon of the insurance industry, we're looking to find that 'twist of lime' the makes it very sticky for the customer base and meets their needs, and machine learning and AI gives us some of that edge," he said.

Specialised insurers are dependent on skilled underwriters who price the policies and present them to brokers. Much of this work involves processing unstructured data based on emails and paper, and AI and ML can ease the burden by standardising some of these interactions.

It's freeing up our underwriters and our claims handlers, so they have more time to work on the things that are interesting

"It's freeing up our underwriters and our claims handlers, so they have more time to work on the things that are interesting. It's reasonable to think they will value the work more and so do a better job. There's nothing worse and doing mundane tasks day after day after day," Penny said.

In the contact centre, incoming calls and emails can be directed (using natural language processing combined with information from customer profiles and records), to the operative best qualified to deal with them. Analysis of these communications can also be used to spot market trends.

Customers don't want to communicate with chatbots. They contact Hiscox rarely and when they do they invariably have a specific query or request so want to speak to a real person. Automatically triaging communications can remove friction here.

"We like to think our customers also have a significant benefit as well," Penny said. "They are busy people, small business owners and they don't really want to buy insurance at all; they want to build their business. The more we understand them and offer them product set that is personalised to them has appropriate coverage the better, and we can avoid the awkward conversation around ‘Oh sorry, that that wasn't insured'."

Form filling can be much reduced too, for both customers and brokers, if more information can be compiled automatically. "Instead of 100 questions, they'd much rather have five or ten".

Clean data is the foundation of everything

All of this requires clean, timely and reliable data, and Penny's team spends a great deal of effort on ensuring this is so.

"Clean data is the foundation of everything. We need to know good looks like," Penny said.

"If you don't know the provenance of the data, how accurate it is, applying robotic automation and machine learning gets you faster decisions which are just as inaccurate as human decisions."