Webio's Chief Strategy Officer, Paul Sweeney, gave a talk at the European Chatbot & Conversational AI Summit 2023 where he unpacked how conversational AI can help companies engage with financially vulnerable and indebted customers.
Here, Paul delves into how this AI chatbot technology not only helps indebted customers but it also benefits the companies themselves.
Hello everyone, I’m Paul Sweeney, the Chief Strategy Officer at Webio and I was invited to give this presentation at the European Chatbot & Conversational AI Summit in Edinburgh 2023. I'd just like to go over that presentation with you and share some of the things I shared with the crew on the day.
Conversations about money can be difficult, and many people don't share the full extent of their debt with even their closest partners, which can lead to anxiety and unrealistic promises to creditors. Avoiding these conversations can result in accumulating debt, late payments, fines, and credit rating impacts.
Financial illiteracy and shame also exacerbate the problem, as people may not understand loan terms or obligations related to bills and repayments and fear judgment from others.
Additionally, lower reading skills and disabilities can create additional barriers for people in addressing these financial issues.
The pandemic has increased customer interactions, with many companies switching to messaging instead of phone calls, leading to a shortage of skilled staff. To increase productivity, companies need to automate conversations, including digital collections and messaging.
Understanding the emotional state and vulnerabilities of customers is essential to establish long-term relationships, especially as recovery time for customers falling behind on payments has decreased.
Customer retention is critical, so companies must prioritize customer interactions and treat customers as valued over the long term. Conversational messaging is now a critical part of the change agenda.
The Webio stack was designed with a focus on credit and collections companies, with a guiding principle of specialised APIs to provide data for these conversations. Access to essential information, like balance and payment due dates, is necessary to effectively automate these conversations.
Webio has developed a specialised bot framework that addresses the unique challenges of credit and collections conversations, allowing for the creation of specific rules and policies tailored to this niche area.
The company has built its system on a custom, Large Language Model, similar to popular models like ChatGPT, or GPT-4.
At an enterprise level, accuracy and relevancy in responding to customer inquiries are crucial. Webio has been developing an architecture that allows for data to flow through a custom model, enabling ChatGPT-like conversations without relying on third-party information. By using their own data, training, and labels, companies can maintain control over their data journey and ensure compliance and auditability.
Here are some examples of conversations that illustrate how a conversation about debt might happen.
Consider a message like:
"I'd love to pay but my wife is in hospital."
While the first part of the message is positive, many systems might overlook the second part, which reveals a potential health vulnerability. It's important to understand the context to know what the next question should be.
Similarly, a statement like:
"I can't pay this."
This can mean different things at different stages of the conversation. Initially, it might indicate a lack of available funds, but later on, it could be due to technical issues with the payment gateway or credit card.
Another example is:
"I'm just off my holiday and don't have any money."
This statement doesn't necessarily mean that the person has spent all their money on vacation; they might be on a payment holiday and simply haven't received their expected income yet. In this case, they will likely be able to pay once they receive their income, but they may need to renegotiate the repayment schedule.
So even short snippets of conversation can be highly ambiguous, and it takes a deep understanding of credit and collections to correctly parse them.
To achieve the value of using conversational technology, it's important to start with simple solutions and add complexity over time.
For example, one customer reduced their high failure rate in the loan application process by implementing a date entity that recognizes any date format, resulting in more automated pass-throughs to authorisation.
To automate larger chunks of conversation, entities such as names, addresses, phone numbers, and custom entities can be added after achieving success with simple solutions. This raw data can be used for identification and verification checks and to capture contact details from customers who haven't provided them.
Budgeting, traditionally a one-to-one conversation, can also be digitised using an automated conversational process review, which is more efficient and cost-effective than traditional methods.
The concept of short-term memory allows capturing customer information at the beginning of a conversation and using it later on, thus avoiding repetition of requests for information. This can also be applied to the handoff process between an AI agent and a human agent. The human agent can assess the conversation and determine the next step, and then pass it back to the AI agent to continue the process.
The gradual process of automation should be approached by starting with automation that involves people and then progressing to automation that involves AI agents. The ultimate goal is to design custom interactions that meet the specific needs of customers, and the Webio stack can provide the means to design these interactions.
Implementing well-designed automation can significantly improve collections, with rates increasing by 3 to 5 times using digital messaging and bots.
This impact varies depending on the specific area of the process being targeted, with some parts seeing a 20% increase, others a 100% increase, and still others a 300% increase.
Overall, the benefits of automation are immense. We have achieved automation rates of around 75-77% using bots, which is a massive improvement.
In Q1 2023, the buzz seems to be all about ChatGPT and Large Language Models (LLMs). This is a massive transformational change that occurs once every 10-20 years. The last big leap of this kind was the jump to the Cloud in around 2000, followed by mobile interaction and, some argue, social media.
The AI and Machine Learning revolution brought by LLMs will transform businesses in many ways. For example, it will make conversations easy to translate into multiple languages, handle text, voice, and visual inputs, and offer new interaction capabilities that may have been previously inaccessible or too costly to develop.
I anticipate that in the near future, people will be able to easily take a picture of a bill, send it in, and have it automatically scanned to understand its contents. They will then be able to receive a response about what's happening with the bill.
Currently, one can drop a document into a specialist service and ask it questions about the document, and it will provide answers. These advancements represent significant transformations in how we handle data.
I believe that Large Language Models will become available at the enterprise level for every company and will be equipped with the necessary safety features. However, it's still important to ensure human review before sending anything out to customers unless there is a well-designed architecture that can accurately detect issues like false positives and hallucinations.
I believe it's crucial to implement a reliable middleware layer to ensure the LLM operates within appropriate boundaries. Simply using them without proper guardrails is not feasible.
The focus is on providing intelligent assistance to everyone in the organisation, from customer service agents to sales reps and developers. Currently, ChatGPT is being used by various teams, including marketing and sales to generate content for marketing materials, emails, and blog posts. And developers are using tools like GitHub Copilot to parse their code.
Webio's mission for the last five years has been to provide an intelligent assistant for credit and collection situations through a specific stack. The solution is tailored to meet the needs of those in this space dealing with challenging conversations about money.
The presentations at the European Chatbot & Conversational AI Summit this year showcased excellent examples of conversational messaging in various industries, including insurance, health, and large-scale retail. Conversational AI and AI chatbots are delivering benefits to many different sectors.
If you have any questions or are interested in conversational AI, feel free to contact me on LinkedIn or contact us to discuss how conversational AI can transform your customer engagement.
Take care everyone!
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. We promise.
Talk to us about Conversational AI
