Mar 31, 2021 7:01:00 AM
The Next Wave of Customer Service Chatbots
How it all started
It was 2016 when people started to talk about chatbots. It was the year when Facebook launched their bot platform for Facebook Messenger and paired it with the upcoming hype around machine learning and natural language processing – the possibilities seemed to be endless.
Yet, it was difficult to find use cases where chatbots were really offering smart and helpful solutions. You could see a lot of examples and concepts, but it was difficult to spot real, useful chatbots in the wild.
One use-case however seemed to be very powerful: Applying these technologies in the field of customer service, making day-to-day touchpoints between customers and business faster and more efficient. The most prominent example was the idea of booking a flight with a chatbot:
I want to book a flight from BER to LAX for May 15.
It looks like a great example for NLP capabilities. Understanding the intent of "booking a flight" and automatically extracting the entities "from", "to", "when". But let's be honest: No one has ever booked a flight that way. And in practice a chat interface might never be the right interface to skim through prices, departure times and luggage preferences to find and book a flight. Still, people approach chatbots with these examples in mind.
Chatbots Are Designed Around Wrong Assumptions
Our experience shows that your customers don't approach you with a clear intent in mind. They contact you because they have questions or problems. Chatbots need to be able to understand these. But they can't because they are missing a skill that is the base of our human language: Asking questions. Asking to make them understand the context of the conversation. And asking to make sure that they talk about the same.
Chatbots in 2021 are broken. They don't ask questions, they don't understand. What's left are frustrated customers that just see your chatbot as another way to put them off.
Let's use a typical ecommerce example: You just submitted an order, but you realise that you have a misspelling in your delivery address. When you called customer service, you would first explain the situation ("I just ordered at your shop") and then state your problem ("I have misspelled my address").
The agent would then ask you questions to better understand it. In this case, if it affects your invoice or delivery address and based on further context around the order status, the agent can provide the best solution:
- Change the address as the package is not yet prepared
- Inform the carrier if the package is already on its way
- Or creating a new order if it cannot be changed
This is how a natural conversation looks like.
Customer service is not user intent driven. Customer service requires real conversations. It requires an agent who knows the context of the conversation, who can ask detailed questions to clarify the situation and who can come up with the best solution.
Do you know what we call such conversations? We call them meaningful conversations.
Customer service needs meaningful conversations. That's why we built the Solvemate Contextual Conversation Engine™️ that is different from any other chatbot.
Solvemate's engine is built from the ground up to provide excellent customer service and to engage customers in actual conversations. It's not built to fulfill a hype. Precisely, it's built around these 3 core principles:
- Easy setup and maintenance, fully in the hands of customer service departments
- Context-based understanding of user problem to provide the best solution
- Engaging natural conversations to make interactions meaningful
The Next Generation of Conversational AI
The next generation of Conversational AI an approach that goes beyond a simple intent matching. It needs a smart way to take contextual information into account and to ask questions to make sure it got the user right.
It starts with the customer writing a message to the bot. The bot will take this as the general direction of the conversation and then take over, asking smart questions that narrow it down, question by question. Once the bot really understood the customer correctly, it provides the right solution - whether it's a simple answer or a whole multi-step business flow that kicks in.
As a company you start building up the bot by collecting the 20 most important contact reasons and you sort them into broad categories. Based on this, you start adding context: how do solutions differ for different customer groups (e.g. private and business), what is needed to qualify for a specific solution and how can you further break down the solution categories. Once live, you refine it based on customer's feedback and you add more solutions to the bot.
As a result you build up a model that represents the inner workings of your service strategy. It's unique to every company.
Whenever a customer texts you, you can't just match it against a list of simple intents. Like your service agents, you need to match it against a full set of knowledge and guide the user along the resolution. And that's the big shortcoming of NLP-first approach in customer service: They promise to overcome this by talking about all kinds of magic machine learning techniques and conversation models. But in the end, they make you build static decision trees that you link to very broad intents and this is in no way smart or natural.
For you this means maintaining a fragmented chatbot structure that is broken down in dozens of static flows that can't learn from each other.
The Solvemate Contextual Conversation Engine™ is just the first step for Solvemate into this new era. User authentication, in-depth integrations into your existing services, powerful automation flows and multi-language capabilities just mark the beginning of a new way to think about your service operations.
Go with a bot that engages your customers in meaningful conversations on all of your service channels. Give your customers the appreciation they deserve.
Goodbye frustration, hello Solvemate!
Jürgen is the CTO and Co-Founder of Solvemate. Before founding Solvemate, Jürgen worked as a developer and coach in e-commerce and is now a partner at the consultancy etribes in addition to his work at Solvemate.