The University of Southern Denmark (SDU) has implemented the IT chatbot Kitt to support their internal users
At the University of Southern Denmark (SDU), the IT Support Department has implemented the IT chatbot, Kitt – named after the Artificially Intelligent car “Kitt” from the 80’s adventure tv series Knight Rider. The chatbot is developed for their internal users, the employees and students.
We talked to Bilal Bahij, the IT Coordinator and Team Leader at their IT ServiceDesk, about his experience with developing and managing the bot. Read along to find out how a chatbot can improve daily life at university.
Key Takeaways
Building the chatbot
- From the acquisition of the BotXO platform to launch of the beta chatbot version: 4 days.
- The chatbot is built using NLU.
- The chatbot is integrated with Skype Teams.
Results after 2 mths
- Customer Satisfaction Score in beta version: 4.6 out of 5.
- The bot answers frequently asked questions 24/7.
- More departments at SDU are interested in getting a chatbot.
Goals for the future
- Decrease the number of support tickets in the IT-department, so Support can spend the time on more demanding tasks.
- That all inquiries with the IT-department start via the bot.
- That the chatbot evolves into a personal assistant for the students.
Bringing the IT-service forward at SDU
The University of Southern Denmark opened its doors in 1966 after the three institutions Odense University, the Southern Denmark School of Business and Engineering and the South Jutland University Centre were merged. Besides the main campus in Odense, SDU has expanded to other campuses in Esbjerg, Kolding, Copenhagen, Slagelse and Sønderborg, with many researchers and employees working in- and outside of Denmark.
Moving on from human-to-human chat
SDU used to have human-to-human chat, also called live chat, where a supporter was able to answer the inquiries from the customers, but as the technology at that time wasn’t good enough for their needs, they abandoned that line and chose only to have support via phone, email or at their service desk.
“Other companies have, for many years, used human-to-human chat. We tried it 4 years ago, but at that time we had a lot of problems with some browsers. For instance, if the user tried to contact us in a Safari browser, we never received the message. It was too early at that time.”
Now with the technology being miles away from the old chat-based system, SDU decided to take two steps forward and reintroduce a chat-based support system, this time featuring a chatbot, Kitt. The aim was to decrease the support tickets handled by human agents, so they could focus on more demanding tasks in the IT-Service. As Bilal explains it:
“There are many inquiries that we get again and again and those are the inquiries that we would like the chatbot to help us answer.”
Moving away from human support and introducing an automated, AI agent is often looked at with some concern by employees, as it could seem as a way to get replaced. But does the introduction of Kitt mean that the colleagues in the IT department will be replaced?
“My colleague’s first concern when we told that we were getting a chatbot was whether they would soon be unemployed. You can safely say that they will not! There is, of course, a lot of work in keeping such a bot updated, plus we get many inquiries that cannot be solved by a chatbot and would never be resolved by a chatbot.”
So, according to Bilal it is safe to say that the chatbot has not come to take over the SDU employee’s jobs, but to make them better. A chatbot is meant to help take the pressure off customer service agents and allow them to focus on more meaningful work.
The chatbot Kitt
“What we imagine the chatbot can give us, is that we in the Support have more time to focus on the tasks that are more demanding. When the chatbot can solve the cases we get repeatedly without the involvement of a supporter, it will, of course, release more time in Support.”
It took Bilal and his team 3-4 working days from the day the chatbot was bought till the first version was launched on their website. At that stage, Kitt could handle all the basics such as the inquiries that the IT service department receives most frequently; the opening hours, location, how to get in touch with the department and reporting of computer problems. The chatbot can be found on SDU’s web page via the IT-Service ‘Support’ or ‘Contact’ pages, and for now it is available in Danish.
Implementing a conversational bot in the IT service.
Ever since the implementation, Bilal has updated and trained the chatbot’s NLU almost daily and today it works 24/7 as SDU’s IT department chat support. By implementing the chatbot, SDU has made sure that calls and emails from all users, including those who are abroad and work in different time zones, are answered round the clock. But why did SDU choose to develop a bot for the IT department first?
“Here at SDU, we have chosen to develop a chatbot for the IT department first, because we have a lot of data that we can use for this chatbot.“
The department had in fact already been collecting data through their previous chat and the general support and thereby knew which questions the users usually asked and would be interested in getting answered through a bot. This meant that a bot could most easily be built and implemented for the IT department. Once SDU decided to go ahead and test it out, they saw the business value very quickly.
“We thought that it was very natural to test it out in the IT department, so that later we can help the other departments get their own bots. There are several departments that have already expressed that they would like to have a chatbot. Especially our Student Services, that is receiving admission applications, has been asking for it (…) I have no doubt that this is something they will be able to use and that it will also be used by the young people who are in contact with the Student Services.”
Solving problems using NLU
SDU has chosen to use the BotXO NLU solution in their chatbot because of the variety of their users and their way of expressing themselves. Professors, other employees, students – all have different ways to explain what their IT-related problem is. By training the NLU chatbot to understand many of the sentences users type when asking for assistance, Bilal has made the bot into a smart assistant.
“The biggest advantage of using NLU is that our users can write to us in the way they want, with the wording they would normally use. And then we train our bot to understand their formulations.”
NLU is built up on domains, intents and example sentences or keywords that are identified with certain sentences. When using NLU it is beneficial to make use of pre-existing data of user communication. Analysing the intents of the users and labelling the data will contribute to building the right communication flow and structure in the bot implementation. The intents are then linked with connections which can be either local or global, and determine where the user is led to during a conversation. Global Connections is the tool that is used when creating a communication flow where the user freely writes input to the bot. Bilal points out the combination of the NLU and the Global Connections feature of the chatbot platform:
“It was easy to build the bot and very easy to understand the way it works. The combination of Global Connections and NLU works like a dream. The way you can train the chatbot by asking the same question in different ways and using those sentences to prepare the bot for the way our users explain their IT problems, and what they actually mean when they say the things they say.”
Of course, the bot can only solve problems in the way that the problems are intended to be solved and in the way that it is trained to solve them. As Bilal puts it:
“The bot can’t be better than the information you give it or the training you build it on. In other words, yes, the bot has received many questions it couldn’t solve. But it is an ongoing process to make the bot better and better.”
So, what does Bilal do to train the bot, and how was his experience using Natural Language Understanding in the bot?
“I have configured the bot so that it sends me a notification via mail each time a conversation starts. This way, I can read the conversation between the customer and the chatbot in real-time, and if the chatbot received any message it didn’t understand, I can instantly train it to be able to answer it in the future. I once managed to train the NLU with a new sentence while the customer kept asking the chatbot for help. The customer was using a name of an application that we hadn’t mentioned yet in the NLU. So, when the customer was asking for the 5th time, the bot understood and guided him on how to solve the issue.”
The chatbot and Skype for Business integration
Besides being an internal IT chatbot for a university, the bot will be the first of its kind to be integrated with Skype for Business. The integration didn’t exist, but it turned out there was a need for it, because, what do you do when you are not dealing with support for external users and there’s, in fact, no entry channel?
“When we got the chatbot up and running, we were in a situation where we had no place to launch it. Not because there were problems with the bot, but because we did not have an entry channel, i.e. a website where people contacted us.”
The solution would be to integrate the bot with Skype for Business – however, the integration has not been activated yet for Kitt.
“We want our chatbot to act as a contact person in Skype For Business. Skype For Business is installed by all our employees and it would, therefore, be a natural place to find the chatbot.”
The idea behind the integration would be that users can chat with the bot via their own Skype for Business and if there is something the bot cannot answer, the chat will be handed over to a human support agent.
“We are very excited about the Skype for Business integration. (…) And of course, the integration, so our support agents can be invited into a conversation and take over the conversation from a bot, and then also send it back again.”
As a concept, the integration would work the same way as other BotXO integrations: for Microsoft Teams, Facebook Messenger, Slack, Zendesk: companies can automate chat support and leave the most common inquiries up to the chatbot to answer.
Building a chatbot is ongoing work
Working on the bot is ongoing work for Bilal. The success of a chatbot is dependent on data analysis and on taking action based on that data. Here’s some of the data that SDU collected for the months of January and February 2019.
During the period from January 1 to February 28, Kitt had 58 conversations.
- 7 conversations where Kitt assisted the user by creating a support ticket.
- 15 missing plugins in Outlook – solved
- 2 asked for handover to a human, not possible – no agents were available at that time, so an agent emailed the users later.
- 5 asked for help to connect to the wireless network at the campus – solved
- 14 asked for the opening hours – solved
- 1 needed help for filling the online form for new employees’ access to the IT facilities – solved
- 2 needed help creating SharePoint Team Site – solved
- 1 asked for a renewal of STATA student license – solved
- 11 unsuccessful conversations – the conversations were dropped by the customer.
All these conversations contributed to pinpointing which changes that should be made first and foremost.
“We make changes to our chatbot daily. It is an ongoing process that we spend a lot of time on. What we have done so far, is that we have used the communication that has been between the users and the chatbot to strengthen the chatbot. We have done this by reviewing the conversations the chatbot has had.”
Understanding the users and observing user experience and feedback is essential when building a bot and according to Bilal, the work they put in has paid off even in the chatbot beta version for the employees.
“We received feedback from the customers, and they were impressed. We have a Customer Satisfaction Score at 4.6 out of 5, and we will work hard to keep our users satisfied, with the chatbot too.”
The first movers’ next step
The primary goal for the implementation of the bot is to decrease the time the IT department spends on repetitive customer inquiries. The goal for the future is to have the chatbot as first line support.
“One wish could be that all future contact between the user and the IT department starts via the chatbot. For the single reason that the inquiries we already know and have a solution for are being handled. It will be a great help for us in the support but certainly also for the user who is having the problem.”
Another goal for SDU is to get other departments to use chatbots. This will be useful for not only the employees in the departments for gaining more time for other tasks but also their users.
“We believe very much in chatbots and many of the other departments at the university, both the Library and the Study Service have shown interest in chatbots. Especially the Study Service, gets a lot of inquiries from students and most often it is the same kind of inquiries they get.”
Furthermore, Bilal has a vision that the bot will turn into a personal assistant for the students.
“We have a vision that our chatbot someday can become a personal assistant for each student. We have many ideas that we already are brainstorming on. But we are not ready to share these thoughts, but I am sure that we will be the first university with this. We always want to be first movers!”
When Bilal got asked on his final remarks concerning chatbots for other companies too, not only universities, he replied:
“Any company that needs to provide its customers some kind of support will soon start working with chatbots. We believe that chatbots can be a great helping hand in customer service.”
If you and your company are interested in chatbots and want to hear more about how to get started, you can request a demo or sign up for a free trial here below and have a test run with our platform for 14 days.
The article is based on the interview carried by Beatrice Carraro with Bilal Bahij.
Article written by Kristina Egholm Rove.
