Chatbots like Slack’s GrowthBot were a major theme of SXSW 2017. Dharmesh Shah, HubSpot Founder/CTO talked through the transformation of our communications styles and internet interactions to move away from:
Generation 1: The “click” interface (desktop computers)
Generation 2: The “touch” interface (smartphones)
to
Generation 3: The “conversation” interface (chatbots)
He made a really great point that “people don’t go to your website because they love your web design, they just want the answer.” And it’s true, in many cases: people have a specific question, they want to answer it, and they want to ask it in a natural way. We can see this also in the two styles of Google search queries:
Keywords
“Harrison Ford age”
(or, if you need to filter more, “site:imdb.com “Harrison Ford” +date +birth”)
vs.
Conversational
“How old is Harrison Ford?”
Similarly, Shah discussed how GrowthBot integrates with tools like Google Analytics or CRM systems and allows users to ask simple questions like “how many site visitors were there from social media in June” and “how many customers from Texas have more than 1,000 followers” and get quick and accurate answers pulled from your tools.
Simple and handy: a main line straight to the answer rather than digging around in the tool itself.
The issues that are still being wrestled with are discoverability and certainty.
- Discoverability – How do you know what a chatbot does? Let’s assume that the backend work is already done of hooking it into your databases. Let’s break that down more:
- How do I know how to make it understand me? Users know that their computers are getting a lot better at natural language processing, but there’s still a level of mistrust and uncertainty about whether the computer has actually correctly interpreted the request.
- What can it do? What questions are legitimate to ask and are possible to answer within the tool?
- Certainty – How do you know if you’ve asked the question the right way for the bot to process? You’re offloading the decision-making to the bot and asking it to understand your intent. When a user wants to pull year-over-year trends from Google Analytics, it’s important to frame the question the right way, or you’re going to get a useless answer – or worse, a misleading answer. Ask for the last year’s monthly results – is that 2016’s monthly results or monthly results starting at the end of the previous month? The bot’s making the choice. It’s not a showstopping problem – you can take a look and see the issue – but it does take away some of the simplicity.
Fitting chatbots into your communications plan
Today’s chatbots work best for people who know that they want. That suits customers at particular stages of the consideration process or of the customer service process. They want to know “What are your hours?” “How do I get a refund?” or they want to take a specific action “I want to get a cab.” For that purpose, they let people communicate in the same way they would with a salesperson or customer service representative.
They’re not so great when users are doing research. For a user interested in purchasing, say, vacuum cleaners, they may not know what features they value or what questions to ask. They’re more asking: “Tell me what you think I should find important about your product?” Or, similarly, if you’re looking through your website analytics to see what unexpected events and trends took place, there’s not – at present – a way to ask those abstract questions.
What’s Next?
Users are still in a trust-building stage with chatbots and other conversation-based systems. Once that trust is established, there will be room to push further on their capabilities – to create more predictive systems and “personal assistants” where the bots can begin to anticipate what will be interesting to their users and provide those insights proactively.
Until then, chatbots can help take the load off of existing systems and help users who do just need a particular answer or particular bit of help.
Image by barnimages.com, accessed via Flickr and used under a Creative Commons CC-BY-2.0 license.