With the explosion of messaging apps in recent years — Facebook Messenger currently boasts 1.2bn users worldwide — it’s no surprise that brands have been seeking out clever opportunities to connect with customers on these platforms, most notably through chatbots. Housed within messenger apps and on brand websites, these mini interfaces are opening up the potential for engaging with consumers, whether to entertain, to broadcast information, or to provide customer service. To explore what makes a good bot, and some of the innovative ways they’re being used by brands, we caught up with Isabel Perry — Head of Technology at marketing tech agency Byte London, which has developed smart chatbot solutions for the likes of JustEat, adidas and HTC.
What are the core elements that any good, functional chatbot should have?
A central decision tree structure with natural language supporting the flow of conversation.
A strong personality with consistent quirks and a sense of humour.
The suggestion that the chatbot is ‘listening’ as well as talking — it’s a conversation, it’s two-way. We see dwell time increase by 500% when the chatbot is less of a monologue.
A compelling use case. Getting people to message a brand is a big ask, so don’t underestimate it.
A reason to send useful regular content keeps people coming back to the bot over time. Clever fall-backs when the chatbot can’t ‘understand’ something that’s been said — this stops the jarring ‘sorry I didn’t understand’ fail.
And, as with everything else, surprise and delight go a very long way. The organic reach of an ASOS chatbot we developed increased by over 10% with the introduction of vouchers.
How does the chatbot you created for Just Eat combine fun and engagement with functionality?
Just Eat is the world’s largest online food ordering platform and has over 28,000 restaurant partners in the UK, connecting customers with the best restaurants in their local area. The chatbot lets people search restaurants near them with emojis, keywords (over 1,500) and a lot of natural language. But we also recognise there’s a need for food inspiration, and an opportunity to entertain customers. The chatbot has been delivering impressive results — 29% of people who interact with it will then go on to place an order within seven days, and they’re spending an average of two minutes and 14 seconds speaking to the brand.
How can chatbots be used as a broadcast tool, to engage in a personal way with a mass audience, for instance with the one created for adidas’s Studio LDN?
Having a broadcasting strategy is key to the longevity of a chatbot. In January 2017, adidas launched an east London gym offering free sessions to women. The ticketing system is run through a chatbot. Every week we broadcast a link to the updated gym sessions, reminding people to sign up. We also send reminders about upcoming sessions.
The concept is simple, but the chatbot has delivered amazing results. It has allowed for deeper engagement through regular one-to-one conversations, delivering year-on-year retention that outperforms most apps.
How are chatbots helping brands to learn more about their customers?
The obvious one here is the conversations. We analyse anonymous transcripts of conversations, and patterns emerge. People might say, ‘Me want potato’, ‘Need food’, or ‘what does my girlfriend want to eat’ to begin a conversation around restaurant discovery. These interests, values, phrases and slang words can inform our customer communication.
Different use cases obviously reveal different customer insights. One chatbot we developed asked people to set a budget for a gift. The chatbot was live for six weeks. We could see spending patterns change in the run up to Christmas, and the behaviour was the inverse of what was happening on our client’s website. There were also marked differences between the generosity of men and women.
What’s most exciting to you in the future of chatbots?
There’s a lot of misunderstanding around AI in this space. The vast majority of chatbots on Messenger can’t train themselves with historic conversations. You have to analyse conversations, train a natural language processing platform to recognise intents, and write responses. I’m most excited about a tool we’re building to begin automating the analysis and categorisations of free text. I think it’s the single most important thing we can do to improve our conversations over time.
For the industry more broadly, I’m excited about delivering on in-depth, personalised conversations at scale. It’s going to take time to improve the general level of NLP, but there’s a lot that can be done with scripts in the meantime. And voice!