Chatbot for Slack
An internal experiment to build a bot that will be integrated into the internal Slack channels to reply to any queries related to benefits.
Business goals
To reduce the load on HR services that tend to reply to the same questions.
User needs
"As an FT employee I would like to learn more about pension plan, gym membership and other perks but it is very tricky to find relevant information on Google Drives or internal sites. I am not sure where to find information I need and even if I find a relevant documentation, it is quite difficult to find an answer to the question."
Outcomes
Chatbot @Ask was integrated in a slack channel of one of the teams. It is being used on a regular basis.
Team
It was a side project so we had a team of volunteers:
- PO
- Senior engineer
- Myself
- Another UX designer
Process
We started off with defining what technologies we could use.
After considering several platforms that could integrate into Google chat and Slack,
the decision was made to start with Slack. Having a global team,
we also narrowed the first iteration to the UK offices as it was also easier to run usability testing sessions.
The second step was to analyse all the question that HR team received over the years and breaking them down into user scenarios.
User research
The goals of user interviews were to:
- understand how people resolve their queries (talk to HR, talk to other colleagues)
- identify the personality of the chatbot and the tone of voice
- identify different variations of the same questions (i.e. in the same scenario how different people ask bot for help)
- understand how people express their satisfaction or dissatisfaction with the raised query
- understand how people follow up with the questions and what’s the flow of the conversation.
I’ve created an instance of the slack channel and pretended to be a bot on the other end when the person was typing in the questions on the laptop of the interviewer.
There were 4 people involved: interviewer, notetaker, participant and a person who pretended to be a chat bot. We’ve interviewed 5 people and we’ve learnt the following:
- What the most popular benefits were
- Every participant had very different ways and used different channels to find out the information they needed
- Participants liked personalised answers (i.e. Hi Olga, how can I help you?) and use of emojis but they didn’t expect the chatbot to be too informal
- There were different ways to ask the same question, i.e. "Hey @Ask, how do I join the gym", "How do I sign up for the gym?", "Hi @Ask, how do I sign up for the subsidised gym pass?", "How can I sign up for a gym membership through the FT?"
- For certain queries such as personal training the participants wouldn’t approach a bot to learn the answers and would approach the gym directly
- There was a need to have a fall back answer option if the bot didn’t know the answer to the question with the reference to the email address or a contact that could help
LinkedIn
Medium
