This page has an average rating of %r out of 5 stars based on a total of %t ratings
Reading Time 3 Minutes Reading Time 3 Minutes
Created on 21.09.2020

Voicebot: how virtual assistants are learning to speak

PostFinance’s chatbot has been providing a valuable service for some time now. Its job is to provide simple written answers to simple customer queries. But now a new virtual assistant is on the horizon: a voicebot that can talk. Dominic Spalinger from Business Development explains how the pilot project is progressing, and why it’s not possible to use everything from the existing chatbot.

How much does a credit card cost? Where is the nearest branch? How can I open an account? Since 2017, PostFinance customers have been able to ask the chatbot these sorts of written questions. That was when the text-based virtual assistant was first introduced, and is still being enhanced today.

Bots can help us improve the quality of the customer support we provide because it gives our agents more time to deal with more complex queries.
Dominic Spalinger, PostFinance Business Development

You can’t have a smart bot without people

“We use incoming customer queries to come up with new responses on an ongoing basis, and feed these to the chatbot,” explains Dominic Spalinger from Business Development at PostFinance. This is also known as training the chatbot or the relevant NLP services . It involves a lot of work: every single answer is noted down by former Contact Center employees who have moved to the chatbot content team, where the bot is essentially fed with content. The chatbot has now evolved into a useful assistant, especially since personalized conversations based on customer data became a possibility this year.

From chatbot to voicebot

The chatbot now has a colleague – the voicebot. “As before, the vast majority of queries received by the Contact Center aren’t received by e-mail, but over the phone,” explains Spalinger. As with the chatbot, the idea of a voicebot is to answer simple support questions in an automated manner, except this time, the answers are spoken. “We’re right at the beginning. With the voicebot, we want to be able to deal with calls better and faster, which will in turn create added value for the customer,  for instance with a 24/7 service or shorter waiting times.” Yet the Contact Center staff themselves could also benefit because they would have more time to deal with more complex support/sales queries rather than basic issues requiring the same stock answers, which would improve the quality of the support provided.

Dominic Spalinger has been working at PostFinance since 2015 and runs an interdisciplinary team that tackles all sorts of projects in the retail area. In addition, as a product owner, he also does work on the voicebot.

How the bot learns to talk

Besides the technological and organizational issues involved, Spalinger sees one of the biggest challenges in developing this voicebot as coming up with the actual content. “We’ve been able to use the same technology we used for the chatbot to some degree, but not the same content.” Because customers speak differently to how they write, and the queries are often different, the training data the system uses to recognize the customer’s query is practically useless. What’s more, the team needs to do more to break up the responses and make them more succinct. This is because, unlike a chatbot, a voicebot cannot provide any images or links to help the customer. “When communication is oral, it’s important the customer receives as simple a response as possible that they can understand straight away.”

Order documents with the voicebot

In the current pilot phase, customers who have provided a voiceprint can order account statements and interest statements via the voicebot by phone in German. This is possible for private and savings accounts, but not gift savings accounts. The customer calls the Contact Center and chooses the option to order the documents via the virtual voice assistant. The voicebot goes through the process step-by-step by verifying the customer’s identity and asking them the relevant questions, e.g. the account or year the customer would like to order the document for. What’s more: the voicebot understands Swiss German.

What happens next

The voicebot is still a pilot project that we are using to assess customer acceptance and its added value so we can decide on what  steps to take next. One thing’s already clear, though: developing a voicebot in the financial services sector is groundbreaking work. “There’s no project like it for us to draw on,” emphasizes Dominic Spalinger. “This is why we are still using trial and error and working with our customers on the project, who, with their queries, help us to optimize the voicebot on an ongoing basis.”

This page has an average rating of %r out of 5 stars based on a total of %t ratings
You can rate this page from one to five stars. Five stars is the best rating.
Thank you for your rating
Rate this article

This might interest you too