Conversational AI and automation have undeniably changed the way businesses engage with their customers, bringing huge gains in efficiency, cost-savings and customer satisfaction. We’ll look at how this tech works to achieve this.
Bot creators need to design bots to answer queries quickly and properly without frustrating customers.
Webio places a strong emphasis on conversation design and believes that how you say something makes all the difference to the outcome. Customers who are in debt tend to be anxious and often emotional, so Webio’s bots are taught how to respond in a way that reassures customers that they have been heard and that resolving their query is important.
Building a bot involves in-depth research and analysis of how your customers interact with you; the terms they use when stressed, common queries, what frustrates them, etc. It involves tapping into hundreds of thousands, even millions, of real-life examples to understand customer intents and how to resolve them. This informs the routing design and triage.
Watch this video to see how Webio's AI works for debt collection
Routing Design for Query Resolution and Triage
The route to the best possible query resolution starts with intent recognition (more about intents below). In the design phase, once you have classified the possible customer intents (what they want to do/are saying), you match each intent with a resolution path. This could be, for example, to a self-service option for quick transactions, to an agent or to an AI chatbot.
Triage involves assessing how urgent an issue is. For instance, a customer asking for an account balance is not as urgent as a customer who has lost their bank card. An AI chatbot or self-service option can handle this action with ease. If a query is critical and time sensitive, it will take priority, especially if the customer needs to be helped by an agent.
Webio's visual drag-and-drop interface
Natural Language Understanding is the AI technology that interprets the messages coming from a person. It takes the natural, unstructured language and converts it into a code that it can then work with. This is the fundamental starting point to conducting a conversation between a customer and the AI chatbot.
Large Language Models (LLMs) are used as the basis for generative AI. They are built using an exceptionally large resource base, including the Internet. However, LLMs like the one ChatGPT uses, are generalists and are not accurate or secure enough for industry specific uses.
Certain industries, such as credit and collections, need to have a more precise and controlled chatbot. This is where Custom Language Models (CLMs) come in. They are often fine-tuned open-source models built with custom data and labels.
Webio’s CLM is trained on millions of real-life customer conversations (with personal details stripped) which makes it a laser sharp CLM for use in conversational AI for customer engagement in debt collection. In essence, Webio’s CLM allows the AI chatbots to speak the language of credit and collections.
[For more, see: Give Chatbots the Edge with Custom Language Models]
Taking it a step further, the AI is taught to use NLU to spot certain words and phrases and label them as a particular ‘intent’. These labels are then used by the bots and agents to better understand what a person is actually saying and wants to do.
In Webio’s platform, the agent can see these intents (called Propensities) in a panel on the agent interface screen. The agents or bots can then take the appropriate action which has been set up in the intent-mapped to-resolution route design phase.
Examples of Intents (Propensity Labels) are:
|Income & Expenditure
The labels are also used for reporting purposes which then give management insight into who is contacting them and why.
Detect Customer Vulnerability Using Intent Recognition
One of the most important benefits of intent recognition is being able to pick up the emotional state of the customer. This is particularly valuable if a customer is showing signs of vulnerability. The AI then sends the user to the best place for them to get help, which is directed by the resolution routing mapped out for vulnerable customers.
The AI looks for phrases that indicate vulnerability such as:
“My wife is sick.”
“I have no money for food.”
Entities are useful pieces of information that can be recognised and extracted by the AI from unstructured natural conversations. Examples are dates, amounts, names, phone numbers, addresses and post codes.
This forms part of the conversational nature of Webio’s engagement platform. A customer does not have to answer each question separately but can include all the details in a single answer. The system is smart enough to ask an open-ended question and understand what a person means.
Entity gathering is useful for ID&V where you can verify a customer in a single step. This way of garnering data is quick and provides a smoother and less frustrating customer experience.
With propensity guidance, the AI system predicts the conversation’s outcome in real-time as the conversation is happening. With this data, the system then routes the customer, within two to three utterances, down the best path to resolve the query. For example, it may send the customer to an agent or to an Income & Expenditure Form.
Webio’s Propensity Studio gives businesses a high level of automation at scale while still remaining personal to individual customers.
Webio is busy building AI that is not only the engine behind the chatbots that converse with customers, but it is also a virtual assistant that comes alongside an agent to help them. For instance, the AI can suggest responses based on its understanding of the conversation, or it can pull data from a backend system to immediately hand the agent accurate data to inform the query resolution.
Sometimes, the quickest and most effective way of helping a customer is providing a way for them to help themselves. Webio's Self-Service Studio is an online portal giving customers the option to make payments, set up promises-to-pay, arrange payment plans and check accounts. Additionally, conversational AI chatbots are on hand to guide customers through the process and answer any questions they may have.
Sometimes, you need to gather information from a customer in order to resolve their query. A good example of this is an Income & Expenditure form. Webio offers conversational forms that can be accessed from within the conversation in the messaging app, like WhatsApp and SMS, and the information is scored and used straight away. Customers are sent to the form by an agent or an AI chatbot.
The conversational design makes for a good customer experience and information is gathered 5x faster than if an agent were asking the questions one-by-one.
Since we live in an always-on digital world, businesses need to engage with their customers through the digital channels they are using every day. In fact, customers today expect to be able to engage with a business in five different channels – WhatsApp, SMS, webchat, Viber and Messenger are popular examples.
Webio’s conversational AI is channel agnostic, which means you can design it once and deploy it on any digital channel.
Customers prefer using messaging to contact a business as it is immediate, less awkward and can be used any time.
[For more, see: Debt Collection Made Easier with Multichannel Messaging]
APIs are coded rules or protocols that enable software solutions to exchange data. Webio has built its own API (using standard Rest API architecture) that clients can use to connect the Webio conversational AI platform to their own software or to another solution, like Salesforce for example. Basically, Webio’s API can connect with any solution that can be integrated with.
Webio also supports webhooks that connect into backend solutions to retrieve information, e.g. to get real time data on, say, bank account balances.
[For more detail, see: What is the Difference Between an API and a Webhook]
API integration brings in real-time accurate data that the AI chatbots and agents can use to inform and personalise customer conversations and help resolve queries simply and quickly. Integration ensures that there is only a single version of the truth with push-pull reporting and status feeds.
As it stands now, generative AI that is built on open source LLMs cannot be trusted to interact with customers about sensitive and private matters such as debt – they hallucinate (fabricate information) about 15% of the time, are not 100% secure and they do not have the precise industry knowledge needed.
However, Webio is building generative AI for use in other areas, such as conversation summarisation and code building.
For more, see:
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.