5 things we learned from The Future of Chatbots event
CognitionX held an event called Let’s Chat about Chatbots: What Really Works and What Doesn’t? on 21st February 2018. With over 100 attendees, 7 speakers including William Tunstall-Pedoe, the inventor of Amazon’s Alexa, the event was a huge success. Here are some of the things we learned…
1) There’s no guarantee that chatbots will be the right solution to your business problem.
You need to understand if your customers are ready for it. How? Test it on them. You can prototype basic chat interfaces in Google Forms or, like Sharmadean from Beautystack, using WhatsApp. She started taking bookings for beauty salon appointments that way and the system currently contributes 40% of the bookings to her London location.
“Our mission was to turn the salon experience into a very conversational one. We did 6 months of Whatsapp bookings. We had hundreds and hundreds of conversations. People were sending us images and emojis – having a full-on interaction.
Our customers loved how easy it was. But they were using the service 24/7. We couldn’t keeping replying to our company phone buzzing at 1am! At that point we decided to build a chatbot.”
2) It’s not as hard as you think.
Systems like IBM Watson allow straightforward proof-of-concepts to be built in a matter of weeks. Steve Gill’s experience at EY of introducing a HR reference information chatbot for onboarding was experimental but showed great results, even at an early stage.
His key piece of advice? “Don’t try and make the perfect chatbot to start with. We stood up a global chatbot system for 250,000 people in just 31 days. The trick was just to get started. If you have the right business problem it will show its value quickly.”
3) Done well, chatbots can unlock interactions you never expected.
Sagar Gupta built a chatbot to support young people dealing with homelessness. A bot built to answer basic questions, simplify complex processes and be available 24/7 had another effect – it got vulnerable people to engage in ways they weren’t with a real-life agent.
Sagar said in order to get started, “You have to spend a lot of time with your users to understand them and what they’re looking to do. This helps you build the personality and the functionality.”
4) In the future of chatbots, context is key.
Chatbots can be trained to learn about the user in front of them, either by scraping additional public information or by learning from the direct user interaction or interactions with similar users. In this way, Danica from Sentient Machines said chatbots can become highly context-aware and hence very engaging. “The goal for our clients is to build something that doesn’t sound robotic.”
5) Building a chatbot is less about the technology than you’d think.
Aldous Birchall, AI Lead at PwC, said that “normal machine learning, to paraphrase, is ‘data in, do a bit of feature engineering, train a model, get something out the other side’.”
However “the design space for chatbots is enormous”, he said. “Chatbots are about elegantly guiding people. There’s the Natural Language Processing, the ontologies, the dialogue flow, the conversation design and a whole range of different technology stacks to choose from. A lot of it is around the personality side and the psychology.”
Guy Gadney from Charisma shared some great stories about how his company had developed highly engaging characters including ‘perfect girlfriends’ and ‘serial killers’ using the latest chatbot technology. So far they have been commissioned by games developers and artists to tweak chatbot responses for different audiences.
In terms of how to build a chatbot team Aldous suggested, “I’m yet to see a unicorn ‘conversational designer’. You need to get a very diverse set of skills and get people that are really interested in the technology.”
A huge thanks to our speakers for taking part. To find out about future CognitionX events, head to our Events page.Published in