Joyce Lopulalan, Vera van der Steen and Tim de Boer about Voicebot technology

Generative AI (GenAI) is hot, but companies are also increasingly realizing the dangers associated with it. And how complex it is to generate the right quality of answers. This is even more true when GenAI is deployed in the form of a voicebot. What does this entail? What applications is this technology suitable for? And how do you ensure the sound, safe and ethical use of a voicebot?

December 3rd, 2024   |   Blog   |   By: Conclusion

Share

Joyce Lopulalan, Vera van der Steen and Tim de Boer about Voicebot technology

What exactly is a voicebot?

Scripted chatbots have been around for a long time now, think of the chatbots used in customer service. Chatbots that work with GenAI, such as Microsoft's Copilot, are emerging. These bots can formulate an answer based on different sources, such as documents in a database. The step to consult GenAI not only via the keyboard, but also via speech, is still rare. But that's what a voicebot does. The big difference with a GenAI chatbot is that a voicebot adds personal interaction and facial expressions to the conversation.

What makes deploying a voicebot so complex?

A voicebot is more than a chatbot through spoken language, says Joyce Lopulalan, business manager at Future Facts Conclusion. "Questions via a keyboard are very direct, to which a straightforward answer is often possible. But if you, for example, have a voicebot conduct telephone conversations, you have to add human interaction to that." Moreover, while it is not unusual in a chat if it takes a second for the answer to appear, such a silent gap in a telephone conversation is not natural.

Her colleague Vera van der Steen, AI and ML engineer at Future Facts Conclusion, developed the first Conclusion voicebot. She explains: "It all comes down to high-quality prompt engineering. This means that you tell a voicebot how to answer the questions: what context should be taken into account? What is the desired level of accuracy? Etcetera." For example, the voicebot must be able to engage in small talk. But he also needs to do proper sentiment analysis and adjust his tone. If a customer calls with a complaint and is quite angry about it, the tone of voice should be empathetic and sympathetic. If a voicebot gives a factual answer in a neutral tone at that moment, the customer doesn't feel heard."

Another point of attention is legislation and regulations, including the AI Act and of course the GDPR. "We've seen a lot of predictive models in recent years that people wrongly label as a risk," Joyce says. "You must therefore know exactly what you are doing and mitigate risks in advance."

It all comes down to high-quality prompt engineering. This means that you tell a voicebot how to answer the questions.

Joyce Lopulalan, Vera van der Steen and Tim de Boer

What are the areas of application?

A voicebot has many possible applications. Tim de Boer, also an AI and ML engineer at Future Facts Conclusion, recognizes that the most obvious application of such a bot is receiving inbound telephone traffic in a call centre. But it can also be applied for outbound telemarketing campaigns, Tim says. "Tesla is doing that for example. People who look at cars on the website and leave their phone number will be called by a Tesla voicebot the next day to discuss the car in question further and possibly schedule a test drive. The voicebot knows exactly which pages the potential customer has looked at and can have a relevant conversation based on that knowledge."

Another application is having information read out to people who need to prepare for a meeting. Tim: "Take a sales manager who, while driving to a customer, has a voicebot read out a summary of the customer history: which products and services has the customer purchased in recent years? Have there been any complaints about that? What details are included in the conversation reports? If something comes up that he wants to know more about, he can interrupt the bot and ask further questions about that topic. This is of course much more efficient than having to consult the CRM system yourself and retrieve that information from all kinds of different places."

A similar application could of course also be made for doctors preparing for a conversation with a patient. Or even for surgeons who want to check certain information about the patient during an operation. "This is because you can link a voicebot to any software, whether it is a CRM system or an electronic patient file," says Tim.

What are the benefits?

The business benefits of a voicebot can be found in several areas, says Joyce. "Firstly, it can reduce workload because organizations are still faced with a tight labour market and a shortage of qualified personnel. But a voicebot can also boost customer satisfaction. Queues for inbound telephone traffic are a thing of the past. In addition, a voicebot will always respond sympathetically and empathetically, even at times when it will be difficult for people not to show their irritation." There is a reason why voicebots are already used in the care of people with dementia on a regular basis. People with dementia can ask the same question dozens of times a day. A voicebot will continue to answer in the same calm tone every time, something that is almost impossible for humans to maintain.

Other advantages of a voicebot are its perfect memory, the fact that you can support the bot with countless sources and that you can easily make the bot speak multiple languages. If video calling becomes possible in the not-too-distant future, you can even let the bot include the context of the image in the conversation.

How do you go about it?

As the examples described show, there are various possible applications of voicebots. You can use the technology in almost every industry and every department. Future Facts Conclusion has developed its own voicebot, for which we cloned Joyce's voice. The bot is therefore called the Joycebot.

To discover whether a voicebot or other forms of AI can add value, Future Facts Conclusion uses a proven discovery process that can test the issues at play within the organization and the added value of solutions with AI, in only six weeks. This way you can quickly identify where the low-hanging fruit is. The risk assessment is performed using the GenAI Framework, a framework that covers all aspects of a company's maturity in dealing with technology (AI Readiness), up to and including legislation and regulations. "This ensures that you don't skip any steps that are necessary to deploy AI solutions safely and compliantly," Joyce explains. Of course, this framework is continuously updated with relevant laws and regulations, such as the AI Act. If a customer then wants to get started on the low-hanging fruit, there will be a roadmap that helps build a minimum viable product in a period of three months.

The Joycebot is an experiment by Future Facts Conclusion, which ultimately aims to have an animated conversation at any time of the day with answers to relevant questions. At this moment the Joycebot is still a work in progress. If you want to train the Joycebot and experience how it works, call the freephone telephone number +31 20 259 59 87 and ask the Joycebot a question. For example: ‘This and this is my biggest business challenge. How can a voicebot help with this? If you like the answer, we will be happy to visit to take the next step.

Emerging technologies trend report 2024 #2

The Conclusion companies continuously monitor emerging IT trends and explore new technologies that may be relevant to our clients. In our semi-annual trend report, we highlight several technologies that, in our view, deserve special attention.