Do you prefer to write or talk? How chatbot and voicebot automated NEXERA’s customer service

Challenge

During the preparatory work, we came to the joint conclusion that it is not worth creating several bots for two problems handled in two communication channels. We decided to utilize the capabilities of the KODA platform and prepare a single, universal solution that will be able to communicate both by phone and through a dialogue window on the website. This solution means that most of the work – including integrations with external systems – will only need to be done once, and each subsequent integration and expansion of the bot’s knowledge will be more efficient and possible to do in one place.

This is possible because every voicebot is, in fact, a chatbot that bases its understanding on a transcription of the command it heard and gives a response by reading what has been marked as the answer to the question it understood using a voice simulator. However, designing a textual and voice conversation presents different challenges – in chatbots, difficulties may arise in understanding the speaker’s intention resulting from logical, grammatical, or spelling errors, while in voice communication, problems with call quality, understanding a person who speaks unclearly, or is in a noisy environment may occur. However, a large volume of conversations can minimize these difficulties.

Finally, in order for the conversation to take place, bridges needed to be built between the systems involved in the entire process, which are:

We also decided to look for ways to enrich our interlocutor’s experience and minimize the risk of errors in understanding what they want to convey, wherever possible. Therefore, for phone conversations, we decided to speed up the reading of the informational obligation regarding personal data, and when the bot is waiting for feedback from connected systems, sound effects appear in the conversation. It was also important to check how the speech recognition mechanism (ASR) and voice simulator (TTS) would cope with the names of places that are already or will soon be appearing in conversations. And these are not simple names, just mention, for example, Gietrzwałd, Gierzwałd, Rąbity, Wąbrzeźno, Szwarcenowo or Zawidz Kościelny, and many others. We tested the components available on the market, and decided to go with the Google Speech solution.

In the text version of the chatbot, we also decided to add an interactive Google Maps map, on which the user can place a pin, which reduces the probability of incorrectly interpreting the entered address to zero.

As you can see, the chatbot handles considerably more conversations, and 70% of them are about network availability. In turn, voicebots are more often used to report irregularities. This shows that voicebots are used more interventionally.

We also noticed that over 80% of people who click on the chatbot icon on the Nexera.pl website engage in conversation with it – almost three times more than we observe in other chatbots. The percentage of people who go through the entire conversation path with the voicebot is even higher – as much as 99%.

And to top it all off – the solution we implemented was recognized by the organizers of the Titans of Hyperautomation (Mocarze Hiperautomatyzacji) competition, who acknowledged the world-class implementation of the project and awarded it with the 2nd prize in the “Intelligent Assistants” category. The Titans of Hyperautomation competition is the first initiative of this kind in Poland. Its main goal is to identify and reward the most effective hyperautomation implementations, whose effectiveness is measured by comparing the business goals set for the implementation with the results achieved after its completion. The event is organized by the Centre for Studies on Digital Government Foundation.