- Posted by: Kurt Kragh Sørensen
- Category: Chatbots, DEX2018
Chris spoke at DEX2018 via Skype and started by telling us, that one of Gartner’s Top Strategic Predictions for 2018 and Beyond is that by 2021 more than 50% of enterprises will spend more per annum on bots and chatbot creation than traditional mobile app development. They also predict that Chatbots will become the interface to IA.
He also reminded us that Bill Gates had said that we always overestimate the change the next two years and underestimate the change that will occur in the next ten.
When and why to use a chatbot
1. You have a question and want an answer
2. Your messaging app is always open
3. Your internet connection is slow and unstable
4. You don’t know which system has the information you need
5. You need push notifications
6. Less than half installs apps every month
7. Your hands aren’t available (voice input like Siri)
8. You think intranets are hard to use.
9. We overestimate people’s computer skills. (The average user cannot use Office 365)
10. Everybody or more understands conversational UI
When to use the intranet?
1. You’re not sure what you’re looking for
2. You need long-form content (But couldn’t that be expressed shorter?)
3. You want to “see what’s happening”
4. You need a full graphical user interface
It is not either/or.
Start with a narrow domain i.e.:
• News bot (What is the latest company news?)
• Event bot (Who’s speaking today?)
• Contacts bot (What’s the number for Pat Davis?)
• Subject matter expter (When are we going to move to the new HQ?)
Have a killer feature that will keep the employees coming back. (Guest wifi pw)
Design failure carefully and steer the user towards supported tasks.
Chris introduced The Confusion Matrix with True-False and Positive-Negative answers. The True-Positive answer to the wifi password could be “It’s 123Password”. The False-Positive could be “We have 3,000 employees”. The False-Negative could be “Sorry I don’t know” if the chatbot actually has the answer. The True-Negative would be a correct rejection if the answer was not known by the chatbot.
Grow your pilot gradually as it is impossible to anticipate what users will say.
Introduce the chatbot to a very limited number and learn from that.
You grow the usage of your chatbot by reducing the incorrect answers.
Most of your time will be spent on exceptions, as you’ll have to review transcripts constantly.
But one of the best things is that a chatbots is automatically user tested.
Full presentation on Youtube
Slides on Slideshare