Blog

Making sense of Machine Learning

author by Graham Sheppard time Jul 15, 2022

Just as it’s doing with almost every aspect of contemporary life, new technology is rapidly changing the world of insurance. Insurtech, so recently a radically transformative innovation, is well on its way to becoming the norm.

Helping our clients leverage the power of the latest technology is a fundamental part of what DOCOsoft does. But we’re not here to supplant the hard-won skills and experience of real-life human beings, simply to empower and augment them.

When we talk about artificial intelligence (AI) and machine learning (ML), as we inevitably do with increasing regularity, we recognise that some people in our industry have questions and concerns around exactly how much of such things they want or need in their professional lives. So let’s take a step back and look at what AI and ML can - and should - be doing in the world of property casualty insurance claims.

Let’s kick off by defining terms. AI and ML are often used more or less interchangeably, but there’s an important distinction to be drawn between the two terms.

AI is what happens when computer systems have been programmed to process information in a way that broadly replicates how our own human brains work, for example through pattern recognition, cumulative learning, problem solving, or decision making.

Machine learning is essentially a subcategory of AI – one of the things it makes possible. It works by building mathematical models of data which then enable a computer to learn by itself, based on its ability to recognise patterns in the data it processes, rather than simply being ‘taught’ or programmed with rules by a human being.

While ML cannot happen without a computer having first been endowed with AI capabilities, over time, machine learning can effectively make the computer system within which it operates ‘more intelligent’.

Taken to its logical extreme, the prospect of intelligent computers that become more and more intelligent, and which take decisions based on their ‘experience’, could just conceivably result in them concluding that human beings should be overruled, or even – as science fiction has imagined - ‘deactivated’!

Happily, current applications of AI in the world of insurance claims technology present no such dangers. The strictly defined parameters within which AI and machine learning operate leave the systems on which they run a long way short of becoming our masters rather than our servants.

But these new technologies can be very effective servants. In the personal lines insurance space, a great many decisions can safely be left to computer systems with AI and ML capabilities – with the obvious proviso that it’s essential to ensure that the way they operate is not likely to discriminate unfairly or prejudicially against certain categories of policyholder.

Within a context of appropriate human oversight, there is virtually no end to the clever things AI-equipped technology can be enlisted to do. For example, AI software can now recognise and process human speech and handwriting with previously unimaginable precision.

It can also recognise complex or subtle patterns within data in just seconds - patterns that might elude human detection almost indefinitely, and certainly for long enough to prevent a timely and effective intervention in response. This ability is priceless in an age when most of us are in constant danger of being overwhelmed by an unprecedented overload of information. Big data is no use to anyone without rapid, powerful pattern recognition capabilities.

More prosaically, AI and ML can play an important role in getting the basics done faster and more efficiently. We’re already putting machine learning to use in everything from process analytics, claims trend spotting, fraud prevention, and compliance. If you can identify any activity within your processes where human involvement is more likely to act as a drag than an accelerant, we can likely create an AI tool that does the job better and faster. Come to that, our AI analytics can identify those activities for you.

Famously, AI-intensive U.S. insurtech business Lemonade quotes the case of a customer whose (low-value personal lines) claim was approved and paid in just seconds, after he had answered a handful of questions via an app and recorded a report on his mobile phone. In the three seconds Lemonade claims it took to settle the customer’s claim, its AI Claims Bot, reviewed it, cross-checked it against the original policy, ran no fewer than 18 anti-fraud algorithms, approved it, paid it electronically, and notified the claimant.

We’re obviously not suggesting that many – or even any – big-ticket P&C insurance claims could or should be settled at comparable speeds, but if AI and ML can strip low-value or no-value human touchpoints out of the claims process, this will minimise unnecessary delay, improve customer satisfaction, and free-up human decision makers to deploy their expertise where it adds most value. And that’s what we’re doing right now at DOCOsoft.

AI was at the forefront of our minds when we designed and constructed DOCOsoft’s new Straight Through Processing (STP) service. This focuses on what can and cannot be automated, and when. Our STP service creates process efficiencies and removes system inefficiencies, building on existing market initiatives and the future change agenda.

In essence, the STP module allows customers to set and control their own varied risk appetite around various financial thresholds, per bureaus, per class of business, and more. The system will alert Adjusters to a claim’s STP eligibility allowing them to opt the claim into STP going forwards but will also automatically fail and exit the claim from STP should a new transaction exceed your configured risk appetite.

DOCOsoft has been actively researching the potential of AI in the world of claims management for some time now. Over the last few years, our research and development activities in this area has led to DOCOsoft to successfully creating its own DOCObots. These are not physical robots, of course, but digital co-workers. They are a new part of DOCOsoft’s technology stack that makes it easy to carry out repetitive low-value tasks automatically.

A bot cannot read and understand the significance of an expert report and take appropriate action, but it can process a diary task, reassign claims, clear static claim tasks, run and email reports, categorise emails and upload documents to a DMS. Any task that has the form of if ‘x’ then do ‘y’ can be carried out by a bot.

To find out more about how AI and ML can help your claims team work smarter and more efficiently, contact graham.sheppard@docosoft.com


Back to top