Conversational AI in customer engagement can be used ethically for helping people escape the cycle of debt and its repercussions.
Collections is one of the most difficult and emotionally charged customer service environments, and customer communication in this industry has been broken for some time now.
Recently, there has been a growing interest in using artificial intelligence to improve debt recovery practices while still upholding ethical standards and genuinely helping those who are struggling with debt.
The sad scenario is, if a person owes money to one company, they probably owe money to five or six others too. And they have to go through the same stressful rigmarole with each one to try get things sorted.
However, if companies truly go digital, they don't have to employ the old legacy ways of doing things. They don't have to be stuck in a pattern of sending a letter once a month and calling every 30 - 60 - 90 days. On the other hand, if they could make the process digital, they could make it flexible and adaptable to their customers.
And what's more adaptable than a conversation? For example, say a customer owed an electricity company €100. In a conversation, they could negotiate and explore options. The result could be that instead of paying the full amount at once, they could pay it over three instalments. And now the utility company is happy because they are getting paid, and the customer is happy because they have money for basic needs that week.
What's more, the customer doesn’t get fined for a missed payment and there's no direct debit costs for bouncing. All these additional costs that are forced onto the people who can least afford them are like a tax on poor people.
During the Covid pandemic we "found out" that nearly everything could be done digitally. We saw companies with thousands of contact centre agents move entirely to messaging over SMS and Messengers with no phone calls at all, over night, and they did not suffer much from the sudden transition. This proved that switching to digital is not only relatively simple, but advisable.
Looking ahead, the path of large language model (LLM) usage in enterprises is leading to each one having their own custom language model (CLM) built for their own specific use case.
Ultimately a company is going to want to have its own AI, trained on its own data with its own labelling. They can bring in useful documents like service level agreements and knowledge bases and feed them into the AI to build a bespoke language model. And in the end, every company has a language model that perfectly services its unique needs.
If an enterprise doesn’t have a custom language model, it will find itself in a difficult place. It will end up having to say, "Hey Microsoft, or hey Google, how would you like to have this whole layer in the middle where everything's actually happening?" It's that crucial layer where all the company's intelligence and sensing is happening, and they will slowly empty themselves out to the tech giants.
An analogy: it's like GM in the 1960s where they didn't have enough platforms to build the cars on. They only had a single platform so all the cars ended up looking the same. So, what happens in the LLM space is you lower costs for a while but then you end up losing differentiation and niche advantage in the longer term.
Example: Webio’s conversational AI platform founded on a custom language model and a unified stack
Webio specialises in conversational AI for debt collection, and they understand that conversations about debt and finances are emotionally sensitive and involve sensitive private information. Given that credit and collection conversations are some of the most sensitive customer interactions, and there are many facets that need to be corralled together to make it work, then what Webio wanted to do is to have everything in their own stack.
For example, say a customer said they wanted to pay €30 towards their account. This would mean the company would need to validate who they are, check their account and then execute the action. All this requires integrations and API connections.
So, Webio built their own natural language understanding (NLU) engine, and they trained their own large language model, i.e. they built their own custom language model (CLM) specifically for the requirements of credit and collections. On top of that, Webio built their own dialogue manager. It all works in a unified framework and has a logical flow and has everything together in one stack.
Transparent and Explainable AI
Notably, having your own contained stack means that none of the customer data goes out the system. Ethical AI dictates that you must be able to say where your data has been trained, what it has been trained on, why you are getting the results you're getting, and you should be able to produce a report to demonstrate this. Having multiple different stacks makes it more difficult to comply with these standards.
Debtors and regulators should have access to information about the data used, algorithms employed, and the factors influencing repayment decisions.
Initially, it strategically made sense to go the way of using the large language models of the big players, but now the tide is turning. For example, Webio has trained their platform to understand if a customer is starting to display the characteristics of a vulnerable person.
Now this vulnerable person may be someone who can't pay a bill because they don't have the money or paying this bill will actually put them in a dire financial situation. And that's going to come back on the company as regulators will be monitoring this type of behaviour.
Intent recognition picks out words and phrases that are flagged as vulnerability indicators and then the AI triages the conversation to the best place for a resolution. There're a whole range of vulnerabilities that AI can look for, health, family, employment, etc. that if you were able to sense properly, you'd be able to intervene quickly and put that person in contact with an agent who would now be able to help them. Human should always be included to work alongside the AI as can provide a level of intuition, empathy and judgement that is beyond AI. (See: How AI Can Help Detect Vulnerable Customers)
And going a step further, the agent is then guided by AI, an intelligent assistant, with response suggestions and useful information, and prompts on how to be sensitive and to use the right tone. And obviously, this type of AI-aided interaction is not something that you would throw to general LLMs like OpenAI which cannot respond with inside knowledge related to a specific customer and industry.
Another useful skill of CLM training is the AI analyses data to identify patterns, trends, and potential solutions related to debt collection and customer vulnerability, such as historical customer data, payment patterns, and economic indicators. It can then offer financial guidance to customers, helping them with strategies for managing and repaying their debts.
Building AI solutions for collections can be more than just building software for more efficient process.
There's a sense of mission about what Webio is trying to do. Imagine if you move people to a situation where they didn't miss that one bill which could trigger a desperate action and ripple tension into their home life. These spiralling situations don’t have to happen. There is a better way of doing things where people don’t get stung by overdraft fees and late payment penalties. This burden can be lifted from customers if they are offered feasible options to repay their debt.
If companies used the AI technology for customer engagement in collections it would help to greatly reduce the stress on families who are struggling. For example, an indebted customer could work with the AI system to set up payment plans and find relief with the flexibility they offer.
Ethical Debt Recovery
AI can contribute to ethical debt recovery practices by promoting transparency, fairness, and empathy. AI-powered analytics can assess a borrower's financial situation and recommend suitable repayment options based on their ability to pay. This approach helps prevent aggressive or harassing collection tactics and fosters a more compassionate approach to debt recovery. By tapping into AI, collection agencies can find mutually beneficial solutions for both customers and the companies they owe.
Going digital and using AI simply opens so many opportunities to help people manage their finances better.
The source for this blog is a TADSummit podcast where Paul Sweeney was hosted by Alan Quayle and Giovanni Tarone. See the full episode: TADSummit Innovators: AI's Ethics & Transformative Potential in CPaaS
If you need to improve your customer engagement, talk to us and we'll show you how AI automation via digital messaging apps works.
You will love the Webio experience.