In this podcast, Garry Bachmann talks to Sophie about the use of Conversational User Interface at CSS Insurance. In this podcast episode, you can find out what difference he sees compared to chatbots, what a Love Brand is and what is special about the technology used.
Garry Bachmann
Garry Bachmann is employed by CSS Insurance in the e-business division, where he is primarily responsible for the digital product experience. For five years now, he and his team have been designing interfaces to customers and potential new customers and then implementing them.
Conversational User Interface
It is important to emphasize to Garry that CSS Insurance does not use a chatbot, but relies on a conversational user interface. The decisive factor for this decision was the experience of using chatbots from other companies, where he and his colleagues were often disillusioned. Companies often try to make dialogs appear as human as possible, but for Garry, the human being remains unsurpassed in communication. In the insurance industry in particular, customer questions often quickly turn into advice, which is usually provided by customer advisors, which is why CSS has refrained from using a chatbot.
Love-Brand insurance, or not?
The aim is not to replace customer advisors, but simply to get users to their destination as quickly as possible via the user interface. Customers usually come to us because they have a specific request or want something specific done. Health insurance is not a love brand where you just pop in to store, says Garry. The Conversational User Interface is intended to create a kind of single point of entry, which in turn directs you to the right place. Of course, you want to enter into a dialog with the customer, but you don’t want to engage in long small talk, as chatbots often do, but rather get the customer to their destination as quickly as possible.
The first steps – try and error
It all started with a prototype that was based on an NLU and designed for a main customer journey. To this end, we looked at which inquiries customers have particularly frequently, such as a change of address, which can then have an impact on the monthly premium due to a change of canton. What was previously solved with a form has now been mapped in a complete dialog in the Conversational User Interface. Even though the change of address naturally remains an important journey for the customer, we at the company imagined the customer experience to be different and were not satisfied in the end, reports Garry.
The solution using a kind of triage was more satisfactory. The focus was on recognizing the user’s intention in the conversational user interface and then forwarding them to the system specialized for their request. CSS Insurance already has specialized target systems for most inquiries and concerns. Whether premium calculator or other forms – the user interface reliably directs the user to the right system. Step by step, we then looked at which journeys in the user interface could also offer the customer added value. Subsequently, the definition of the corresponding intents was started, so that today a large number of journeys are already mapped in the interface.
A technical refinement for more flexibility
CSS Insurance relies on Dialogflow from Google for the NLU. However, CSS relies on an intermediary between the NLU and the company’s core systems, which contains its own logic. In this way, the NLU can be replaced in the future as required and solutions from IBM or Microsoft can be used, for example, without losing the entire business logic. These in turn provide certain applications. The solution also offers more flexibility in terms of data protection.
There is no such thing as no answer
No matter how well you train your own conversational user interface and analyze user input, there will always be user queries for which the user interface cannot find a suitable intent. But this is precisely where CSS Insurance has come up with something clever. As there is an answer to almost every conceivable user question somewhere on the company’s website, as soon as no suitable intent is found, the search entry is forwarded to the website search. For example, it can happen that the appropriate answer from a previous blog post is played out to the user.
Learning where the customer’s needs lie
In addition to the learnings that arise during the conception and implementation of a conversational user interface, there are also the daily learnings. Although it was always in the back of our minds that such a possibility could arise, it has become one of the most important points over time. Every day, Garry sees the user entries in the admin interface and removes so-called no matches, i.e. entries that could not be assigned to an intent by the NLU. This enables him to see where the customers’ needs lie and, above all, what is not yet covered by the company’s current range of products and services.
Learnings
1. just try – just give it a go, advises Garry Bachmann. It’s not bad to make mistakes. The worst thing is not to try at all.
2. actively engages in adequate expectation management with users. Communicate clearly what the digital assistant can and cannot do.
3. think about what the user interface will be used for. Should entire processes be mapped in it or should the user “only” be guided to another target system, for example.
4. do you want to be flexible in your NLU solution or does data protection play an important role for your business? Then an intermediary between the NLU and your core systems may be the solution for you.
It’s best to listen to the podcast episode with Sophie Hundertmark and Garry Bachmann for yourself. Have fun!
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.
