People who end up in debt can feel that their situation is spiralling out of control. They may feel shame about talking about their debts, and they might not be financially literate (so don’t understand how they got into the situation), and worst of all, while they struggle to deal with the problem it can worsen.
It’s complicated for the companies who collect debts too. They need to know the debtor’s income in order to know whether they can afford to pay their debt, and they need to have uncomfortable conversations with them to find the best way to pay off the debts.
So how can automation improve things?
Someone who knows a great deal about that is Paul Sweeney, Co-Founder and Chief Strategy Officer (CSO) at Webio Ltd. As Kane Simms said in his introduction for Paul, he’s “a legend in our space.” Paul’s presentation showed how it should be done on VUX World’s stage at The European Chatbot & Conversational AI Summit in Edinburgh in 2023.
How to engage with the topic of debt
As Paul says (edited for brevity), “The end customer is somebody who is either paying late or falling into debt. And if you’ve never been in that situation, what happens is you stop talking, you stop interacting, you will not fully reveal that to your partner. And I will posit that you don’t even tell yourself because your self-conversation isn’t engaged with this. And because of that, your anxiety ramps up, you’re in avoidance mode, you actually constrict your thinking, and the more it goes on, the more anxious you’ll become. And that leads then to withdrawal. You withdraw from the problem. And of course, what happens is that you’re actually damaging yourself.”
As you can see, the people who talk to debt collection agencies are often in an extremely vulnerable situation, and they might not want to engage with the problem.
When communicating with them, it’s extremely important to evaluate their situation to understand whether they can pay their debts. Perhaps they have other needs which should be addressed first. This means that conversations can be emotionally fraught and lengthy.
Volumes of calls to debt collection agencies have been increasing in recent years too, and as with all industries, there’s a shortage of live agents in debt collections. Conversational AI can converse with anyone, whenever they need it, so there’s great potential to use it here.
An intelligent assistant that knows its boundaries
Webio provides an intelligent assistant that their clients – the debt collection agencies – can fully control. Those clients need tight data security, and they need to be compliant with various regulations, but at the same time automated conversations thrive on data. This means that Webio’s platform needs to be able to learn from a client’s data, but also ensure that data remains secure.
They’ve been utilising bespoke LLMs (large language models) to parse context too. If a customer says “I’d love to pay but my wife’s in hospital,” a bot could easily misinterpret this utterance as the customer’s intent to pay. It’s likely the NLU would detect the words “I’d love to pay,” but less likely that it would detect that the customer’s entire utterance makes clear they can’t pay.
In an ideal world, NLUs would be trained on every conceivable thing a customer could say, but that’s neither possible nor desirable (as there is a point at which NLUs become unmanageable if they become too complex). LLMs could be a way to parse language to understand it better before it gets fed into an NLU.
It’s a matter of context. Talking about debt is not just a matter of agreeing on how much can be paid – there’s always a bigger story around the debts that needs to be understood first. The challenge is in helping customers to start talking about debt, and then ensuring that your assistant understands what they’re really saying.
Learnings from difficult conversations
As Paul says, they’ve found various ways to make these things happen.
Sometimes the tone of how your assistant talks makes a huge difference to how likely a person is to respond to it. They found that simply rewriting the tone of an opening prompt for one client increased uplift by 30%.
They’ve also achieved a 3x-5x return on engagement. This means that their messaging was more likely to have people pay off their debts than communication via email or letters.
They’ve now automated around 50% of live agent conversations, and up to 77% of debts are collected via automated messaging.
Everyone listen up
So what can we expect in the near future?
Paul predicts that companies should be prepared to adapt their architectures to accommodate large language models in the next 18 months. As we’ve been discovering LLMs can be utilised in many ways that not only save time but also increase functionality. Webio have started using them for multilingual conversations, for example.
Click here if you want to watch Paul Sweeney, the legend, giving his talk at The European Chatbot & Conversational AI Summit in Edinburgh in 2023, and he also recently appeared on the VUX World podcast.