#1 Voice Bot Introduction

Hey! Welcome to the series — How to build a voice bot for your business. Let’s understand the main concepts around the voice bot world and why you should consider doing it in your company. Today we will check the different types of voice bots, we will see possible use cases and also some precious tips on how to design a good conversation flow.

Voicebot is a Chatbot with voice!

My name is Bruno, AI Engineer at Sogedes and I will show how you can implement a Voice Bot with real examples and implementation details. It is supposed to be a more technical series, so in the next blog posts we will cover the technology components and we will use Google Dialogflow to develop together a Voice Bot that can be integrated with some cool APIs and Sentiment Analysis model.

But before going into more complex stuff, let’s start with the basics and recap quickly what is a Voice Bot. Simply put, Voice Bot is a Chatbot with Voice. It is an Artificial Intelligence technology that is able to understand some input speech, for example from a phone call. It is able to process it using Natural Language Processing (NLP) and return a nice response, like in a normal oral conversation between people. Many times we call this Conversational AI, which englobes both Voice and Chatbot.

And there are 2 different types of voice bots: Open Domain and Closed Domain. You probably have contact with Virtual Assistants in your daily life, like Siri, Alexa and Google Assistant. Or maybe you already tried ChatGPT. This kind of bot can react to questions from many domains and they can do small talk, which means they are Open Domain, where the conversation can go to many paths. Usually this kind of technology is more expensive, since it requires a large amount of data and training to provide good results. Besides that, generative models like ChatGPT are usually used in Open Domain chatbots, which leads to less controllable and predictable responses (until today), something businesses want to avoid. Consistency in the response is key, but this is a fast pacing field: frameworks and tools to better control model output are being developed in light speed, like RLHF and Langchain.

On the other hand, voice bots specialized in solving a specific problem in a certain context are called Closed Domain Bots, also known as Domain Specific. They are more accessible, cheaper and require much fewer data to provide a good performance (normally Rule-based chatbots are used here). And that’s where I want you to focus your attention on. Many Enterprise Voice Bot applications are simple and focused on one domain, which opens a great opportunity to improve your business processes.

For example, let’s say many people call your Contact Center to book appointments. A voice bot could handle these requests automatically. Or perhaps your company provides phone support services when a customer has a problem. The employees workload could be reduced if a Voice Bot listens to the issue and automatically creates a ticket. Possibilities are endless, let me know in the comments how a Voice Bot could help your business.

Even if the use case sounds simple, there are actually many things you need to consider, especially if it is the first time you are implementing it. Here are some tips we use in our Company to develop successful Voice Bots for our customers.

1. Start with a simple case in a narrow context. You don’t want to automate your entire Call Center operation right away. Choose a process that can be easily automated and that could reduce employees workload.
2. Voice Bot is not a Chatbot. You may already be using a Chatbot on your website and you want to reuse it in your Contact Center by adding the Voice component. But that may not be a good idea. Many chatbots give extensive responses during the conversation, and that is something you want to avoid. The conversation experience can be different when you read a text or when you listen to it.
3. And that leads to the third tip. Responses should be as short as possible. We want the conversation to be dynamic, so that the customer experience is enhaced.
4. Fourth tip is also about building the customer experience. Avoid robotic responses, such as “Please say yes to proceed”. Always try to simulate what a real human would say. Try to make it cool and add some energy to the dialog.
5. The next thing you need is avoiding (when possible) open-ended questions. Even if it is a Closed Domain Robot, it is important to minimize errors. Try to guide the user through the conversation. For example, instead of saying: “Hello, how can I help you?”, try something more directive, like “Hello, do you need help with this or that?”. After you have this foundation in place, you can start to improve the bot to make it smarter. But that takes time and a lot more data.
6. Now try to think in a real conversation where someone ask your credit card number. I imagine most people does not have it right away. You would probably say: “Please, wait a minute so I check the number”. This kind of scenario will also happen in a Voice Bot conversation and you should handle these long pauses so you don’t lose the conversation flow.
7. And talking about conversation flow, something that also happens a lot in a talk is giving wrong information. Imagine you give the credit card number to the bot and you notice later you gave a wrong number. Next, you could say: “Sorry, I gave the wrong card, the actual number is 1234”. Now, the Voice Bot needs to understand the context and switch back to the previous state, so it retrieves the correct card number.
8. Your voice bot has its limitations. It is not a human. Always develop use cases where you can transfer the call to a human, or offer human help if the bot doesn’t understand what the person said twice in a row. Technology is good, but it can’t handle 100% of the cases.
9. And since we know that voice bots have their limitations, it is always important to validate the information retrieved from the call to make sure it is correct. For example, in a scenario where the caller wants to change the address, you should ask the user to confirm it after the Voice Bot extracts it from the speech.
10. The final tip is knowing that building a succesful conversation flow is an iterative process. It is about constantly getting feedback and improving the model on a regular basis.

That’s it. We’ve covered all the important concepts you should know before implementing a Voice Bot. And if you need help to build a successful project, here in Sogedes we offer the whole package as a Service for you, so you don’t need to worry about anything. We’ll help you design the best use case, we’ll take care of the technical implementation and we’ll make sure your customers are happy with it. Let us know your use case and we will be very happy to take a look at it.

And it gets even more interesting in the next blog post. We’ll look at the key components you need to build a voice bot and go deeper into technical details such as NLP technologies, Speech Recognition and Open source approaches. See you in the next post, bye.

Note: if you wanna try a Voice Bot demo we have available, you can call: +49 621 92109138

If you have any technical questions, just leave a comment or reach me on linkedin. Thanks! (https://www.linkedin.com/in/brunofcarvalho1996/).