About Dominic Spalinger

Dominic works at PostFinance and has been working in the chatbot area for three years. His responsibilities include managing the squat team in Retail Banking. This is responsible for the implementation of all innovations in retail banking. He also manages a project himself as project manager and product owner, which deals with chat and voicebots.

Progress of the chatbot project

To the PostFinance text-based chatbot

The PostFinance chatbot has been implemented for some time now. As a result, a change in technology was obvious at a certain point. This transition turned out to be quite complex. Integration in a corporate environment in particular is not trivial and represents a lot of work. PostFinance now has an entire chatbot team. Initially, it was a classic project team, which has now evolved into an implementation team of around ten employees. An additional team of five employees is also responsible for Conversational Design. Your task is to write the bot dialogs.

The idea of implementing a chatbot originally came from senior management as a strategic project. As a result, Dominic was chosen to drive this idea forward. Dominic emphasizes three indispensable aspects when implementing a chatbot project:

• Business case: You need a good business case that you can sell and a strategic reference. It is also important to ensure sufficient support from management so that the project can be set up properly right from the start.
• Iterative approach: The implementation of chatbots and voicebots is a highly complex matter. It’s easy to get lost in the details during conceptualization and implementation and suddenly not know what to do next. Accordingly, it is important to start small and build on iterative steps.
• Choice of technology: Initially, it is essential to decide on a suitable technology for the respective use case. Questions such as: Should it be a rule-based or AI bot? Do I want a text-based bot or a voicebot?

Development and approach

When you start a chatbot, you don’t have any content yet and don’t know exactly what customers will ask and what the bot should answer. Dominic and his team were faced with exactly the same situation. Almost nobody had any knowledge of this technology. In order to build up knowledge, knowledge was first purchased. Consultants were hired and a knowledge transfer was initiated to build up knowledge internally. As a next step, PostFinance launched the project by implementing a bot on a subpage. The team itself has defined what a customer might ask.

With this approach, a lot of time was spent defining intents and writing the trend data ourselves. In other words, Swiss Post has defined for itself what customers ask and how they ask these questions. When the bot was live, they quickly realized that customers weren’t asking what they thought they were asking and certainly weren’t phrasing it that way. With this realization, the trend data and intents were unusable. So the team decided to move away from synthetic data and let customers ask what they want to ask.

The bot was developed in three weeks of iterations over the course of a year. This process can be described as follows: The bot went live and customers asked questions. If these could not be answered, they were recorded and trained and then implemented again from the beginning. This approach is therefore a trial and error approach. In this procedure, Dominic considers it important to promote cross-company exchange. On the other hand, certifications are also of great benefit and bring a lot of new and in-depth knowledge into the project.

Key figures

The first Post Finance chatbot went live in October 2017. After that, intensive attention was paid to content development and more and more content was implemented. After a year, there were around 3500 sessions per month on postfinance.ch.

Today, Post Finance has integrated the bot on several platforms, such as mobile browsers and online banking. Various new integrations of the bot were carried out between 2017 and 2020. The number of chats has risen massively in recent years. At peak corona times and the associated corona credits, there were even up to 55,000 chat sessions per month.

In addition, the voice bot was set up and went live in June 2020. This is implemented in the telephony environment. This not only automates the text channel, but also the voice channel.

Voice Bot

To the PostFinance voicebot

The voice bot is part of Post Finance’s overarching bot strategy. This is driven in particular by the contact center. This area represents a very expensive channel with hundreds of agents answering customer queries by phone, chat and email. The mission statement is to shift the focus from support to consulting and sales.

There are large volumes of repetitive requests for the text and voice channels that could be reduced. Mail and live chat reductions could be introduced using the text-based bot. However, one of the most popular contact channels is still the phone call. There is great potential for automation using a voice bot.

Usecase

One use case that has already been implemented is the ordering of interest and account documents. When the tax return is due and customers can no longer find their interest documents, the contact center receives many calls within a short period of time to order these documents. Using the voice bot, documents can now be ordered via a voice bot instead of a traditional agent. The advantage of this application is that customers can order documents around the clock, do not have to wait in queues and can reorder more efficiently than with an agent.

The application is currently being used and several orders are placed each day. However, Post Finance is still reluctant to go live, as the voice bot is still in a pilot phase. However, the bot is always available during official opening hours. Such phases help to gather experience and observe how customers use and cope with voice interfaces.

Development and approach

The development of the voicebot was based on experience with the textbot. Two hypotheses were put forward.

• The great effort invested in the text bot can be reused to apply it to the voice bot. The content can therefore also be reused.
• The same technology and dialog type can be used for both the chatbot and the voice bot.

The first hypothesis had to be rejected because the content that was developed turned out to be unsuitable for voice bots. Texts that work via graphical user interfaces are too cumbersome and long. Read aloud, they are not intuitively understandable. Furthermore, the trend data collected for the text-based bot could not be used, as customers express themselves differently in writing than verbally.

However, the second hypothesis was confirmed. However, various factors relating to the selected technology must be taken into account. In the end, it took five to six months to develop the voice bot, from the license negotiations to going live.

Learnings

1) A good business case, an iterative approach and the choice of technology are key points when implementing a chatbot project

2) A trial and error approach is worthwhile if there is not yet much knowledge and experience of the technology.

3) Not only the automation of text channels can create added value for customers and companies, but also the automation of voice channels.

4) Various text-based and voicebots can be used to reduce large volumes of repetitive requests and save costs.

5) When developing the voicebot, you can build on experience with text-based bots. However, there are some differences and factors to consider.

It’s best to listen to the podcast episode with Sophie Hundertmark and Dominic Spalinger for yourself. Have fun![vc_empty_space height=”40px”]

And if you don’t want to miss any more podcast episodes, then sign up for the mailing list at the bottom of this page.