‘Our new chat experience meets the consumer where they are’
AI is everywhere. My toothbrush has AI. My washing machine has AI. My vacuum cleaner has AI. I’m pretty confident my AI smart speaker is smarter than I am, and it definitely knows more about my partner’s taste in music. So, if AI can clean my teeth, wash my clothes, hoover my house and knows what an “Ibiza Chillout Classic” is, surely AI can help us help advisers sell more General Insurance?
That’s the question we’ve been wrestling with at Source for the past few months, and our first product with generative AI at its core, is almost ready. We’re about to embark on testing a new AI chatbot with a select group of advisers, that will hopefully make it easier than ever for customers to get a referred home insurance policy.
The core of our new referral platform is a conversational interface - simply type your message to our chat bot, answer its questions, and we’ll build the quote in the background. We all know the problem; consumers are short on time and need solutions that fit in around their busy lives. With our new referral platform, your customer will be able to chat with our system via a web interface or via WhatsApp, fitting the quote process around their schedule. No more long form filling, no more drop downs and no more scrolling through endless options and choices.
We’re transforming the process from “Oh no, I need to sit down tonight to do the insurance now!” into “I’ll do the insurance now, while I’m watching Strictly”. Our new chat experience meets the consumer where they are, takes a break when they need to and still has the all-important human element ready to step in if things get difficult.
Sounds simple right? Well, we’ve certainly tried to make it as simple as possible for the adviser and the customer, but behind the scenes this outcome has been anything but easy to achieve!
If you’re following the news on AI (or reading the excellent blog posts by our IT Director, Chris) then you’ll know that errors and inaccuracies, also known as hallucinations, continue to plague generative AI. The steepest learning curve we have encountered embedding a language model into our quoting process is learning, and accepting, that it doesn’t do the same thing every time you ask. No matter how specific or detailed your prompt, there is no fighting the inherent randomness within the system.
We spent a lot of time testing different models, looking for the optimum balance of obedience and flexibility. We wanted the conversation to feel real and that means dealing with vague input, letting the customer go backwards if they make a mistake, and dealing with the language models’ often hilarious attempts to inject “personality”. The ability of large language models to synthesis and classify text makes them superb for handling a conversational interface, far superior to a rigid chat workflow. Let’s face it, nobody wants to get a quote from the WhatsApp equivalent of “Press 1 for a bungalow”, but that still doesn’t mean that you can let AI run the show.
Ultimately, what worked best was to allow the language model to handle the processing of the text, but to have code in the background checking the input and output and making sure that everything was being done correctly. This approach, combining traditional symbolic code with neural network-based machine learning, is often referred to as “neurosymbolic AI”. In our opinion, it’s the best way to leverage the benefits of AI, whilst minimising the risks of a hallucination taking the system off the rails. (Chris also likes how saying “neurosymbolic AI” in meetings makes him sound clever, and we haven’t fixed that problem, yet). - I think we could drop this part as, I love the humour but don’t want it to detract from the previous statement which is powerful and more serious? What you think?
So, what does this all this mean for advisers and customers?
It means we’re moving forward with the latest technology without forgetting the things that are really important - reliability, trustworthiness, and creating a great outcome for the customer. It means we're putting quality at the heart of our process, not just rubber-stamping something because it means we can put “AI” on our website. Whilst other people are messing about creating digital clones of their board members, we're working on robust tools that create new ways for us to do business and support our users.
Most importantly, for you as a adviser, this means that soon you will be able to safely refer business to an AI that will work on your behalf with your customer at a time and place that suits them - even if that is in the middle of the Pasa Doble semi-final of Strictly.
Source Insurance operate a fully advised home insurance panel, alongside their referral offering. To find out more visit www.thesource.co.uk.
This article was created in partnership with Source Insurance
