Stan Chang, Director Group Buyer Underwriting, talks about how artificial intelligence has revolutionised the world of risk underwriting.
I’ll confess. I’m old school. Old school in a prosperous, century-old credit insurance company that was built on a bedrock of pride, people, expertise and a gold star service. Back in 2016, my approach to economics, finance and risk management was anchored to classical theories and regression models that was thrust into a daunting Brave New World. Enter the new cool of Big Data, Artificial Intelligence (AI) and Machine Learning (ML) that consultants and technocrats prophesize will transform and disrupt. Digitalisation has beckoned the age of the fourth Industrial Revolution, they say.
Stan Chang, lol竞猜(中国)联赛赛事官网
Why does it matter? lol竞猜(中国)联赛赛事官网 sells credit insurance (and surety, collections and information). We indemnify businesses when they suffer a bad debt in their trade receivables ledgers. On our books is 780 billion Euros of aggregated trade credit lines that support trade volumes large enough to put lol竞猜(中国)联赛赛事官网 in the top 10 economies of the world, measured by GDP.
The naked simplicity of what lol竞猜(中国)联赛赛事官网 underwriters do is this: we make underwriting decisions on requests by our customers to insure credit lines on their buyers. We process 20,000 credit applications a day, give or take, and we monitor the credit exposures we have taken on board.
Now spread 780 billion Euros of credit exposures across a portfolio of 50 customer countries, and a few million buyers in 230 countries who trade in 700 trade sectors that span agriculture to aeronautics. With erratic trading conditions, shifting economic cycles, tumultuous geo-politics, fragmented information landscapes, disparate legal frameworks, and diverse customer profiles, the list goes on… all converging on credit risk, simple quickly becomes complex.
The point is, our 100 year old operating model thrives on a recipe of people, data and systems through thick and thin, and is funded by competitive premium rates that keep us razor sharp. So, old school or not, the spectre of AI either colliding against, or being warmly embraced by our business, or fading away into history for that matter, presents two irresistible questions. Are we sleep-walking through the oft-proclaimed fourth industrial revolution? And what do we do about it?
Fast forward seven years later, from when those nagging questions roused us out of our smug bed of success. As I write, we are rapidly scaling AI and ML proprietary tools - both functionally and geographically - that drive our credit underwriting activities. These new generation tools help our underwriters deliver incremental efficiencies and superior underwriting performance that gets sticky on the maturity curve of yesterday’s successful practices and tools.
Here’s how we arrived at this point. We identified high frequency activities in our operating model to which we could apply Big Data technology. It was a good place to start, as using new technology to do familiar things gets better traction and galvanizes internal followship. It’s not What we do, but How (and Why) we do it. On our short-list of high frequency activities were information gathering, information processing and automatic underwriting - on most days, performed many thousands of times in lol竞猜(中国)联赛赛事官网 across the world.
First stop, we started by building APIs to amplify our information gathering capabilities. OK yawn! But it was exciting then to automatically pull in tens of thousands of financial filings from company registries into our databases, replacing manual effort, and reducing dependency on third party processes. Emboldened by our small success, we then scaled APIs, webcrawlers and machine learning to browse the World Wide Web, progressively in multiple (human) languages on a 24-7 basis from hundreds of thousands of open sources, which include news publications, company announcements, analyst opinions, job advertisements, product campaigns, gazette data and social media. And crucially, we automatically match this information reliably to entities in our databases. That was a showstopper for a long while, so when the solution came, it was like a magic switch. Innovation doesn’t need, or want all the answers from the get-go, asking the right questions is far more important.
With the mountains of information we were harvesting, surely we would need a fantasy army of underwriters to trawl though it? We found our answer, not in India or Romania, but in the shape of Natural Language Processing software. NLP is a type of AI which we used to co-build customised models to mimic human reading of financial reports, and to curate the rich but chaotic internet information. As if I need to say, machines read in seconds what takes humans hours. Web data is far more complex and diverse than financial reports, but the sophistication of NLP and ecosystems that use machine learning, classification, knowledge graphs, taxonomies, image recognition and sentiment analysis is pushing out new frontiers in this space. What once would have taken an underwriter hours to research is delivered to their desktop before they’ve even made their morning coffee.
From information gathering to machine reading, our Big Data journey graduated to automatic underwriting (using machines to make credit decisions). After one-and-a-half years of collaborating, prototyping and testing, we are rapidly migrating our traditional algorithms and decision trees to an AI and ML platform that uses Neural Networks. Whether you are a quant, a credit manager or CIO, the holy grail of machine decisioning is to deliver consistent outcomes that produce lower claims (through superior predictive capability), higher premium revenue (from a higher acceptance rate), lower operating costs (via more automatic decisions), and increased customer satisfaction (more protection, fewer losses, faster response times). And we delivered.
So yes, combining old school values with new world science is how we are gradually morphing. Beyond AI and ML, there is a fertile ground of technology - cloud, data science, robotics, e-trading platforms, connected ERPs and eco-systems that bestow believers, more specifically do-ers, with handsome rewards. These rewards range from cost-savings, to productivity gains, customer delight, new propositions, and dare we dream, new business models with game-changer accolades one day.
All said and done, innovation is inextricably hardwired into human evolution, and the symbiotic relationship between man, machine and money will continue to shape future society and species. For now, I’d rather like to think of technology as a means to an end, performing tasks that are too difficult, too boring or too expensive for people to do well. Artificial Intelligence will herald a golden age for humanity if we learn to become better humans. But there’s my old-school thinking again.
Article originally published in Insurance CIO Outlook