Listening to bots
After spending much of 2016 knee deep in the world of bots and conversational interfaces I thought it would be interesting to post what I have learnt and what I’m thinking about in 2017.
By John Borthwick
Jan 20 2017
We backed or built thirteen companies at betaworks in the bot, or conversational tools space last year. This was an area of focus for us, which will continue in 2017 as we further develop these thirteen companies, look at new companies, and extend our expertise into voicebots and verbal computing. This post is based on our experience, most of the companies mentioned are ones we are involved with.
A few things I learnt listening to bots in 2016
A new medium requires new technology and new design. Each time we encounter a new medium, we do what humans do, we retrofit our assumptions and needs from prior experiences. Bots are frontier technology — starting to deliver on a promise that computing has been making for decades — that we will be able to speak to computers, in our language, be it text or spoken word. There are technical and design challenges involved in this promise. The technology; AI, NLP, NLU, the language training data sets, the design nuance of bots, how you write for a bots, how you engender personality in bots — are all new areas of expertise. While it’s trivial to spin up a bot today, making a good one isn’t simple. Bots, when well executed, are a new mode of interaction, natively mobile, different to apps, they are a new medium.
Bots aren’t apps. Just as apps are different to web pages, and web pages are different to pages, bots are different to apps. Bots can complement an app, they are both native mobile software, but they aren’t a direct substitute. Bots — like other new native software experiences — are taking time to emerge. It took 18–24 months for apps to start to emerge as distinct experiences from web sites. Looking at the app store, six months after launch and pretty much every app was an iteration of a web page. One year into its life and you can start to see native experiences emerge, mostly games. It was only after that, in the coming years that you started to see native app experiences. We are seeing a similar path and time scale of development with bots.
Channels for distribution in the bot world have yet to be defined. In 2008, the app store became the channel of distribution and discovery for apps. The app store was then, and is today, equivalent to the directory structure we had in the early days of the web. Over 2016 a set of bot stores opened as platforms assumed bot distribution should be via a store directory. This approach to distribution and discovery — these bot stores or directories — have not worked. The approach didn’t scale for the web and it hasn’t worked for bots. I realized in 2016 that so much of the power and the potential of bots lies in the specificity of the content or service and mapping that to social context seems to be an emerging solution to distribution.
In 2016 Slack became one of the most successful environments for organic distribution of bots. If someone on a team — in a company — installs a bot and finds it useful, they tell others in the team, via Slack. Statsbot leads the list of “brilliant bots” in Slack’s store, however, most users have found out about the service through connections on Slack. As a result, Statsbot has become one of the most popular bots for analytics today with over 20,000 installs. The real driver of installs is that people who are using it are doing so because someone in their Slack group installed it. Moving forward, I want to know the bots you like, then install and try them fast.
Discovery of bots should be in context (ie: part of the messaging flow), relevant (filtered by relevancy in time, conversation, location or social graph), social (easy to share), fast and simple to trial. There is evidence social distribution approach works — from Slack in the US and now with the launch of Mini-programs, WeChat in China is following the social path. One of the reasons the Xiaoice bot (see below) was so popular was that it initially launched as a ‘real person’ that anyone could invite into a conversation. Platforms need to build around this approach — the interface might be search or it might be message specific, that’s unclear, but I think it needs to happen inside of the messaging experience. I’m hoping that in 2017 we will see one or more of the messaging platforms in this US figure this out.