Dr. Iain Brown is the Head of Data Science for SAS UK&I and Adjunct Professor of Marketing Analytics at the University of Southampton. For the last decade he has been working closely with the financial services sector, providing thought leadership on the topics of risk, AI and machine learning. During his time at SAS he has been involved in driving innovation in AI and the corresponding fields of machine learning, computer vision and natural language understanding through the delivery of numerous projects. He is also a contributor to SAS’ blog and an active member of the AI community on Twitter.
In a wide-ranging conversation about the applications of machine learning in the financial services sector, Iain offered some helpful advice around the integration of AI into business models. He also shed some light on the ML trends that financial industry insiders are excited about for the coming year. For more expert opinions on a range of topics around machine learning and AI, be sure to check out the rest of our interview series here.
Lionbridge AI: How did you become interested in machine learning?
Iain: My interest was originally piqued during my undergraduate degree where I was first introduced to the topic of Operational Research (OR). I had enjoyed mathematics and statistics up to this point but had always struggled to see the practical applications. The topic opened my eyes to how data and insight fundamentally affects the world we live in. Machine learning wasn’t the nomenclature at the time, so I’ve actually held jobs where I have been referred to as a statistician, data miner, analytics specialist and now data scientist, all while working on the same topic.
“Although the terminology has changed, the fundamental drive to transform a world of data into a world of intelligence has held constant.”
L: Which machine learning projects are you currently working on?
I: As a global organization, there are too many to count, but from a UK perspective we are actively engaged in a number of novel ML business applications. We are currently working with a large UK insurer on revolutionizing their pricing process through the use of tree based learning, which determines the exact price to charge for each insurance policy. Compared with their traditional linear based estimations, our work with machine learning is more accurate by an order of magnitude. We are also doing some great work on Data for Good projects with Wildtrack in Namibia, which uses computer vision to identify specific species of wildlife for non-invasive tracking, and VU medical centre in Amsterdam, who are using natural language processing to improve tumor diagnosis.
L: SAS has been building machine learning into its solutions since long before the recent explosion in interest around AI. In your experience, what are some effective ways of incorporating AI into your product and how has it impacted your services?
I: SAS have been delivering machine learning solutions, a cornerstone of AI, for 40+ years, so we are in a fortunate position to capitalize on this experience. In the more recent past we have focused on three ways in which we can incorporate AI to best serve our customers:
- Supporting our customers in the areas that will make them successful in AI: data, computing, and skillsets
- Embedding AI into our own solutions, making them more effective and enhancing user experience
- Providing our customers with extensible capabilities to help them meet their own AI goals and develop their own AI applications
These 3 pillars have helped to solidify our spot as market leaders in the ML and AI space.
L: AI has a vast range of applications within the financial sector. Do you have any interesting examples of your clients building AI into their business models that you could share with us?
I: We’ve worked on a variety of machine learning projects for customers, from traditional approaches like supervised learning in credit risk to more recent applications of natural language processing in fraud detection and customer authentication. One I have worked on recently of particular interest was the use of image detection in the insurance claims process.
If you’ve ever had an accident, big or small, you will know the lengthy and time-consuming process you need to go through with your insurance provider to either repair or write off your vehicle. We’ve worked across the insurance sector for a number of years, applying machine learning to enhance the decisioning process and reduce claim time from weeks to days. We’re now looking to extend these efforts through the collection and processing of data upfront. Instead of needing to phone your insurance provider, wouldn’t it be more convenient to take photos of your car in an app and have the decision made instantaneously? Computer vision makes this a reality. Images from past claims are used as training instances to derive a damage severity model which can be scored in the app, while new images enable insurers to immediately determine levels of damage and repair costs. In this instance an image is truly worth a thousand words!
L: You wrote an article about AI in insurance, where you mentioned that ‘the advanced algorithms that typically empower AI only astound when fed the right data’. What role is training data playing in your work and how do you identify high-quality, relevant data, both for your clients and your own solutions?
I: When I’m asked this question, I often think of the Sherlock Holmes line from The Adventure of the Copper Beeches – “Data! Data! Data! I can’t make bricks without clay.” This sentiment is as relevant now as it was in 1892.
“Good quality data is the foundation for successful conclusions. Unfortunately, the hype around AI sometimes drives organizations to focus further downstream without putting the right data strategies in place.”
In most projects, this involves pulling data sources from disparate places within an organization, engineering features from these sources and munging the data prior to training a model.
L: More generally, how do you expect AI to grow and develop within the financial industry over the coming year?
I: A recent IDC study indicated that “By [the] end of 2018, 75% of enterprise and ISV development will include cognitive/AI or machine learning functionality in at least one application, including all business analytics tools”. We see this as indicative of most of the big financial services organizations at this time. Personally, I’ve also seen an increased focus on moving beyond the simple automation of backend systems (robotic process automation) towards intelligent process automation (IPA) for customer interactions.
L: Are there any use cases of AI that you’re particularly excited about?
I: Before we get too far ahead of ourselves, we should remember that AI is still in a fairly immature and weak phase. There is a long journey ahead before we see truly general artificial intelligence. Having said that, combining computer vision with natural language generation (NLG) looks like an exciting next step. The concept here is not only to have a machine recognize and classify images, but also to add the ability to explain what the image is displaying in a human-like manner. NLG has strong applications in truly contextual chat bots, which we still haven’t fully realized.
L: Finally, do you have any further advice for anyone looking to integrate AI with their services?
I: Start with the why and try to avoid the hype. Focus on the practical applications and the business benefit that can be derived, then look to determine whether the data, skillset and capabilities are available. The key question we always ask at the start of an engagement is not of what is possible, or what can be done, but rather what should be done, and the ethical implications associated with it.