by Janus Boye
A screenshot from the member call with Aiste 2nd from the top.
With 3 chatbots in production and one more launched during the height of COVID19 in April, Danish insurance firm Tryg saw a substantial usage increase and also learned a few valuable lessons.
Aiste Hoffbeck is digital front runner at Tryg in Copenhagen and recently joined our member conference call series to share insights on their chatbots.
Below I’ve shared my notes from the call, including some additional perspectives from the Q&A. You can also browse the slides.
4 chatbots in production
At Tryg they’ve been working with chatbots for a while, with the first one dating back to June 2018. Aiste started the call by listing their 4 chatbots currently in usage:
Rosa - launched in 2018 as an internal chatbot in Denmark. Almost 62,000 conversation to date and offers limited escalation to human as a part of the flow.
Mia - also launched in 2018 as an external chatbot in Norway. Recently reached 265k conversations and offers human escalation as a part of the conversation design as well as Bank ID login for proper verification.
Herdis - launched in late 2018 as an internal chatbot to all Tryg employees in Denmark and Norway. Less than 4,000 conversations and offers no human escalation.
Ebbot - launched in April 2020 for Swedish customers and at around 4,000 conversations with human escalation enabled.
As a chabot platform, Tryg uses Norwegian vendor Boost.ai with integration to eGain and Vergic.
The chatbots are currently not connected to their existing knowledge base, but there’s a business representative role, which reviews the content to ensure changes are aligned with website.
In terms of governance, the chatbots are included in the website policy, it doesn’t ask personal questions and the solution has automatic filters which deletes sensitive things like name or address.
What went well with the chatbots during COVID19?
As COVID19 hit the Scandinavian home markets for Tryg in mid-March, Aiste and her team quickly added a ”Corona” button in welcoming messages of Mia and Rosa with relevant information on claims, coverages and guidelines.
With the continuous update of available information, usage of Mia in Norway increased in March – April by 50%, equalling about 20.000 additional chats. Most of it (more than 80%) were handled by Mia. To manage those additional chats by manual chat or phone would have required 18 employees. Meanwhile, claims-related phone traffic only increased by about 20%.
For Rosa, the internal chatbot, internal communication regarding corona virus and consequences for our customers quickly became about 30% of Rosa’s conversations. This amounts to about 10.000 questions, saving about 650 hours that would have otherwise been used to put customer on hold and ask a colleague or search in the knowledge base.
As an interesting lesson learned during the past months, 30% of secret shoppers didn’t even realize they were dealing with a chatbot when shopping for insurance on the Tryg website.
What did not go so well with the chatbots
Ebott was launched in April in Sweden with limited scope to handle travel and Corona questions. It was too late to target the corona questions and Tryg customers were not used to chat as a channel. Consequently, customers ignored welcoming message and kept asking all other questions for which the bot was not trained, like about car claims to a bot focused on travel claims.
According to Aiste, they didn’t anticipate user behaviour well enough and it was too little, too late.
They also worked on a travel bot for Denmark which never made it into production. As Aiste put it:
The business was simply too overwhelmed by a 400% increase in calls and cases to spare any internal resources to own the bot.
Scoping work started last week for a proper chatbot solution.
What’s next when it comes to chatbots at Tryg?
Voice is the next thing according to Aiste. They are working on two approaches at the moment:
Interactive voice response powered by a chatbot with a to-be scenario of speech-to-text processed by chatbot to trigger transfers to right department or give guidance to customers.
Agent co-listening. Specifically, the idea is to let Rosa, the internal Danish chatbot, listen actively to conversations and suggesting answers, issuing warnings and reminders based on context.
They are already running into limitations with the Scandinavian languages, as voice recognition is not yet as far down the road compared to English, Spanish or other major languages.
Learn more about chatbots
In our community, it was at the Boye 2017 Aarhus conference, where chatbots initially gained major traction. Read this summary - Artificial Intelligence: How To Capitalise On The Huge Potential - where learning #3 is on chatbots being here to stay.
Since 2017, chatbots has been covered in several peer groups and we’ve also had a member call on chatbots at Royal Caribbean Cruises. Without revealing too much, customer support was a major use case and content governance also came up in this call
Interested in learning more? Or do you have any questions about chatbots? Feel free to leave a comment below to continue the conversation.