Today, Wolfgang Hildesheim and Lars Mallien talk to Sophie about the use of chatbots as digital assistants in the healthcare sector. In this podcast episode, you can find out what Project 116 117 is, what it has to do with elves and why the use of agile project teams can make sense.

On Wolfgang Hildesheim and Lars Mallien

Wolfgang Hildesheim is Head of Watson (Data Science & Artificial Intelligence DACH) at IBM and works with many young data scientists from all over the world on a daily basis. Lars Mallien is a Data and AI Specialist at IBM and is responsible for Watson Services, which offer AI capabilities locally from Switzerland.

Project 116 117

The 116 117 project is an information bot of the national patient service, whose telephone number is 116 117. People in Germany can normally use this number to make appointments with doctors and be put through directly to an emergency doctor in an emergency. During the first wave of the coronavirus pandemic, however, it became apparent that the hotline was overrun with panicked calls and was therefore paralyzed. In addition, many calls are always about the same questions, such as characteristics and procedures in suspected cases of infection with the virus. These findings were the starting point for a cooperation between the National Association of Statutory Health Insurance Physicians and IBM.

Agile approach with mixed teams

But how do you actually go about a project like this when you are under enormous time pressure? For Wolfgang Hildesheim, the key lies in an agile approach. Teams from KBV and IBM worked closely together and there were daily coordination calls. First of all, it was important to compile the relevant questions. At the top of the list was the difference between a normal flu and a corona infection, what types of protective masks there are and what the advantages and disadvantages are, as well as information on adhering to the AHA formula to protect against infection with the virus. The bot can also provide information about the nearest test station to the user. In this way, a first version of the bot could go online after 3 weeks.

Choosing the right technology

There are two main types of technology to choose from. On the one hand, there are rule-based chatbots that guide a user through a dialog with the help of buttons and, on the other hand, there are intent-based chatbots. Intent-based bots enable free text input from users by trying to find out exactly what the user’s intention is. Since there is a multitude of ways to ask a particular question and in order to offer the greatest possible flexibility for users, the chatbot is fed with hundreds of intents from an external database. In this way, for example, it is also possible to receive answers to requests that contain a spelling mistake in the question.

If the bot still doesn’t know what to do, it automatically queries the user and asks them to rephrase their request. At the same time, the bot learns with every user request and gets a little better every day by training the linguistic patterns in the background. In technical jargon, the type of intents that become finer and more specific over time is called disambiguation. IBM Watson also has access to a set of pre-trained intents that vary depending on the chatbot’s area of application and help it get started a little faster. Nevertheless, the bot must of course be trained for special intents such as Covid, test or breathing mask.

Success must be measurable

It is important for the optimization of the bot to make success measurable. In addition to user numbers, this can also be direct feedback from users, for example. In the info bot 116 117, the user can rate the conversation with a thumbs up or down at the end of each dialog. It is also a good idea to track whether users actually click the bot to the end at certain points or whether it is terminated prematurely by the user because, for example, the process is perceived as too cumbersome. As an example, Wolfgang says that if users want to arrange a test via the bot, they actually stay within the bot until the appointment is made.

Overall, both draw a positive conclusion about the use of the bot so far, which is used by several thousand users every day. Users no longer have to wait on the telephone hotline and receive the best possible support for their concerns around the clock. This is certainly due in no small part to the very high intent recognition rate of over 90 percent, meaning that users will rarely find a chatbot at a loss.

The sound makes the music – even in healthcare

Of course, IBM has also thought about the design of the personality. The bot is a mixture of friendly and humorous, but at the same time also very trustworthy and fact-based. To this end, all answers were checked in advance by both doctors and lawyers to ensure that they were correct. However, as this is a medical topic, it is important to make it clear right at the beginning that the bot does not replace doctors, but is merely a digital assistant. This is to ensure that no false needs are aroused among users that cannot be satisfied in the end.

A lot of effort has also been put into the visual design of the chatbot. As the number 11 appears twice in the hotline, two elves are recognizable in the bot’s logo. Of course, the elves also have wings, a phone hanging around their necks and read from a pink book – real multitasking. This design is intended to make users aware of the bot and arouse their curiosity. At the same time, Hildesheim emphasizes once again that the answers are scientific and highly precise and in no way frivolous.

Learnings

1. agile project teams enable you to achieve good initial results quickly, even under great time pressure.

2 Depending on the use case, consider whether the bot should be intent-based or whether a rule-based bot will also work. The first are associated with significantly higher costs.

3. success must be measurable, otherwise there is no basis for optimizing the chatbot.

4. be bold in the design of the chatbot, this arouses the curiosity of the users and makes them want to try out the chatbot.

It’s best to listen to the podcast episode with Sophie Hundertmark, Wolfgang Hildesheim and Lars Mallien for yourself pure. 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.