The case for human-machine collaboration in underwriting

Shamolie Oberoi   /    Content Marketing Specialist    /    2022-01-07

LinkedInLinkedIn

Legacy banks are in danger of - and in some cases - are already disintermediated from financial transactions. In the FinTech and neobank age, they’ve been left behind to play catch up on evolving customer expectations of high-quality products and services, delivered at speed, contextualized to specific needs.

Building on top, not ground up, brings the edge

Meeting an increasing demand for ‘anytime, anywhere’ finance requires  traditional banks to update their legacy IT infrastructure, business processes and operating models.  And building this infrastructure in-house requires immense investment both in terms of capabilities, money and time. Partnering with technology firms can give a head start to their transformation agendas, bringing in agility, scalability and digital dividends.

Things do seem to be moving in the right direction - research shows that over 60% of financial services firms have already embedded artificial intelligence (AI) and robotic process automation (RPA) to enable straight-through-processing. These technologies, alongside innovations in cloud, machine learning (ML) and the internet of things (IoT), have helped banks and lenders further improve process  efficiency with  minimal human intervention. 

Do humans feature in the future of underwriting?

Let’s take underwriting, for example. Almost 80% of the underwriting data is scattered across sources. Sifting through this manually results in redundancies and delays. On the other hand,  automated underwriting has a 95% straight-through processing rate and is up to eight times quicker than a human underwriter.

The numbers make a strong case for advancing automation in the banking sector. But does this mean it will soon push human underwriters off the map?

Not really.

Firstly, underwriters won’t use models they don’t trust. The financial services industry works on a trust deficit, requiring extensive background checks and borrower profile assessment to rein in risk. Automated decision making must thus be auditable, back to its very first step for underwriters to be confident of its accuracy. The focus must thus be on building transparent, responsible AI-based underwriting models that can make traceable decisions in standard cases as they  automate the redundant, time-consuming steps in the process.

Most importantly, they should be able to flag cases where humans need to take the final call. After all, there is a certain value to intuition and judgement born out of the human experience. And research backs this up - surveys have shown that while users are happy using technology to engage with information, they seek out human intervention when it comes to creative problem solving.

Underwriting requires a nuanced combination of hard and soft skills, and the latter - such as communication and critical thinking - can only be brought to the table by humans.The question, then, isn’t whether humans will be made obsolete by AI. What we need to be asking is this - how do we build an automated underwriting system that replaces redundant tasks and so  enhances the value of human decision making?

It’s all about playing to one’s strengths

The optimal underwriting system should work in a way that allows underwriters to focus on strategic tasks and offers recommendations where required. FinBox’s proprietary risk engine, for instance, is built to automate human judgement for a variety of use cases. It’s based on ML models that are trained on billions of data points, and it evolves at scale with every new or contextual use-case. It makes use of AI and ML only when needed - such as in use cases where fuzziness is tolerable (e.g. bank names) or when our predefined templates don’t recognize a piece of data.

According to the Harvard Business Review, firms achieve the most significant performance improvements when humans and machines work together. It’s clear then, that the most effective underwriting decisions can only be made when the analytical and quantitative abilities of machines join forces with abilities, attitudes, and skills that are uniquely human.