EN speaks to Adam Greenwood, CEO of Greenwood Campbell, about the company’s use of chatbots to help Glastonbury communicate with its audience.
Tell us a bit about the background of Greenwood Campbell.
We’re a digital agency. We started in 2009 as a purely web tech agency and then about 18 months ago I started doing a lot of research and travel learning about the AI space and where communication engagement was going. I spent some time in Silicon Valley learning about everything from AI machine learning to autonomous vehicles, drones, automated supply chain – all sorts, as well as chatbots and voice.
Then we had a very young workforce of which a lot of them go to festivals, so off the back of this, it must have been march or april time last year, we were trying to find a suitable innovation project we could work on using a chatbot AI that would solve a problem that we could identity. We brainstormed that and people started talking about how difficult it was to know who was playing at Glastonbury, because of a few of the guys had been. You can’t really get twitter updates, if you’ve forgotten to download the app there isn’t really enough signal to download it at the festival. Sometimes people maybe didn’t have room on their phone to download it anyway.
We started looking into facebook messenger as a potential platform, mainly because it’s already on the vast majority of people’s phones and also it doesn’t take very much data to operate.
We thought that was a great way to supply information around acts and stages. Then we started connecting to web services which provides information about all the acts and that stages and where they were, but we also connected to twitter as well and were looking for certain hashtags around secret acts, so if anyone had spoken to the chatbot during the festival and highlighted their desire to learn about secret acts then we could broadcast back to them, telling them who the secret act was and what stage they were playing on.
Then we launched it at Glastonbury and over the weekend we had around 50,000 messages. It was very simple; you could either search for the name of an act or you could search for the name of a stage or search for secret acts, it would ask you a couple of follow-up questions and then it would give you the information that you wanted.
Was it learning as it went?
At this stage it wasn’t because at the time we started writing it, because Glastonbury was coming up very quickly, these things need a bit of time and lots of questions. We tested it as much as we could amongst ourselves and colleagues, friends and family, but then when we actually launched it there were lots of things it couldn’t understand, so we had three or four of us manning the bot over the weekend so that if the bot was asked a question it couldn’t understand then we were able to help the bot learn for the future or jump in and answer ourselves.
No matter how much you test, there’s no accounting for what festivalgoers will ask a chatbot. There were some weird, strange requests…
What were the benefits of the chatbot?
It’s all about the experience of the festivalgoer; there’s not someone at Glastonbury answering questions via phone or messinger, so it wasn’t saving time, what it was doing was creating a better experience.
You’re either given a paper programme that gets ruined or lost, and then it’s trying to remember, and being such a huge site you need to know which direction you’re headed in. Certainly around the secret acts, which is something people get excited about, a lot of people can’t get twitter working.
How did you market the chatbot?
It had its own facebook page, and it very quickly happened through word of mouth. As soon as a few people started using it we saw massive spikes and it spread. We found people were searching for Glastonbury Festival on facebook messenger anyway and finding it that way.
The bot would say ‘Hi, I’m the Glastonbury bot, these are the things you can ask me’ – and away you go.
I think from what we’ve learnt in terms of the way people ask questions and the way that the bot learns, the bot is currently offline, because Glastonbury isn’t on this year, but we still have the code. All the different ways that people would ask a question – ‘when are Foo Fighters playing’, ‘what time are Foo Fighters playing?’ – typos etc and all the different ways people speak to it, it’s learnt all that, In terms of rolling that out to other events…there’s so much use there. Anything that’s at the O2, or any of those big venues, again mobile phone signal can be such a pain. It’s a question of a data feed with all the information on it and how easy it is to get that fed into the bot.
Can you see a future where we would message a venue bot and ask questions?
Absolutely, in terms of the efficiency around customer service, the number that most people talk about in the industry is when it comes to customer service and bots is about 30% saving. You can either use that efficiency to make cuts, if you wanted to, in your customer service team, or what we would certainly encourage is to use that efficiency to empower people to answer more complex questions.
Any FAQ type question that would normally require a person to answer would be much better served by a messenger bot.
Off the back off that we started talking to AFC Bournemouth about building them a fan engagement chatbot.
We started to play around with some AI and the thing we started to look at the most was something called Microsoft Emotion API, and what that does is that it’s able to recognise emotion in someone’s face from a photograph, it can recognise nine different emotions based on the face that you’re pulling at the time. We started thinking how much fun it would be to be able to track the emotions of fans during a football game, and create a timeline where you could see – based on the selfies that people upload – how the crowd were feeling based on whether there was a goal, free kick, penalty etc.