Chatbots are increasingly becoming part of the workforce across many industries, as customers actively seek to self-serve where possible and receive assistance on-demand.
Sophia Warwick, solution architect at Sutherland Global Services
Chatbots are increasingly becoming part of the workforce across many industries, as customers actively seek to self-serve where possible and receive assistance on-demand. In the mortgage industry, there has been lots of talk of robo-advice, but not a great deal has happened yet and the sector lags behind others.
We know that chatbots can help brokers by gathering preliminary intelligence about a customer’s financial circumstances. We know they can make customers feel more comfortable by acting as a sounding board for ‘stupid’ questions. The potential for chatbots, however, goes far beyond what we have seen already.
Yet before we can get to that point, the industry needs to understand what the right type of artificial intelligence application for chatbots is. That way, mortgage companies can ensure they are channelling resources at the right kind of technology, and not wasting it chasing the latest trend or buzzword.
Fixed flow vs machine learning
We might know what we want chatbots to help us achieve, but knowing what type of artificial intelligence (AI) we need to make that a reality is more difficult. There are two types of chatbots: fixed flow and machine learning. Think of the former more as a dynamic flow than fixed, as it doesn’t necessarily follow a linear progression.
Fixed flow chatbots are trained and controlled by humans, relying on predetermined equations as a model. While they might not sound as innovative as their machine learning counterparts, for highly regulated environments such as the mortgage industry, fixed flow AI is undoubtedly the right option.
Machine learning might be heralded as the ‘smartest’ of the AI branches, but the needs of the mortgage industry simply do not warrant it. Machine learning chatbots should be applied to sectors where this is a long tail, where there are lots of variables and no standard approach – not where there is a clearly defined process to follow.
The travel industry is a good example of where machine learning bots would be more at home. There are millions of question variations a customer could ask a travel agent (human or bot) where there is no easily defined response; from ‘where is a good holiday destination for windsurfing’, to ‘can you recommend a good hotel that caters for coeliacs’. There are only so many variations a customer would ask a mortgage agent.
Unlike fixed flow bots, machine learning bots do not rely on a formula or equation that has been inputted by a human. Instead, they gather intelligence directly from data they are exposed to, and sift through these data-sets to return an answer they think matches best. They also become more accurate over time, as they are exposed to more data.
In an FCA-regulated industry you simply could not have a chatbot sifting through options and selecting one as it sees fit based on how much it has had a chance to learn already. Essentially, it would be like leaving a toddler to offer mortgage advice and if it got it wrong, the consequences could be disastrous for your customers and your business.
Bot potential
Fixed flow chatbots have huge potential in the mortgage industry that is yet to be realised, and can be used at any point within the end-to-end journey. They could act as a concierge for the mortgage process and direct a customer from Rightmove to a reputable broker by being linked to a Trustpilot rating.
Chatbots could be used further down the line after the mortgage is secured to respond to any questions a customer may have about - for example - a redemption statement, making a lump sum overpayment or remortgaging. Bots could also aid lenders in cross selling and upselling. By acting like a data scout and asking the customer if there’s a particular reason for the site visit it can suggest certain products.
Customers respond well to a personalised service, and chatbots are also capable of delivering on this by having authentication built in. This means that a customer could pick up a conversation where they left off, avoiding the formulaic ‘Hi, how can I help you today?’ and the time consuming conversation that follows.
If the mortgage industry is going to catch up with other industries in its chatbot capabilities, it needs to have a better understanding of the types of AI that sit behind bots, and what is most beneficial to pursue and develop. We need AI that is human assisted and subject to quality control, rather than that which is left to its own devices. Machine learning is too risky. The mortgage industry needs to step back, think about the processes that chatbots would be used for, and go with the (fixed) flow.
