In January, I wrote about how 2016 had been declared the year of bots by many. At year’s end, the bots narrative has been very loud but mostly disappointing. Most of the major technology companies announced big initiatives. But the platforms, enabling technologies and new products have largely failed to materialize.
The demand from startups for a platform that opens up new channels for distribution, making new consumer experiences easier to deliver, is as strong today as it was a year ago. Consumer interest in moving beyond apps seems equally as keen. But in general, this was a year of over-promises and under-delivery on actual experiences.
What comes next will have a lot to do with whether or not developer ecosystems are a priority for the big platforms in 2017, and when the enabling technology gets good enough to provide quality experiences at scale.
No New Avenues of Distribution
As anticipated, the large technology players made several big announcements of new opportunities for distribution beyond the traditional app stores over the last year. In practice, however, no new real distribution materialized.
Facebook made its new Messenger bot platform a big theme of its F8 developer conference. The company announced several early applications that day, as well as a host of APIs for allowing companies to hook into Messenger. The primary initial excitement, harkening back to the original F8 in 2007, was that allowing apps into the large “white space” that Messenger represented would open up a flood of new app distribution and kick off a new startup arms race.
But, unlike those earlier cowboy days, the Messenger platform ended up being too locked down and the distribution avenues too weak to generate the hoped-for opportunities for developers. This failure can either be seen as a realistic and necessary casualty of scale, or a major misfire in Facebook’s quest to monetize the Messenger platform.
Amazon clearly leaned a lot further into the Echo and the bot platform it hoped to develop on top, to make it a more valuable and defensible ecosystem. You know Amazon is pushing something hard when every box you get from them is wrapped in ads for its bot platform on the tape. The product seems to be very well-liked and selling well. (I have five at home.) But having spent time developing some things on the platform, I can tell you that Amazon to date has made it very difficult for developers to distribute software on the platform and has failed to provide the discovery mechanisms that would allow new developers to flourish.
So the Echo, while continuing to be promising, in practice remains a very good alarm clock that can tell some jokes.
Google was relatively late to the consumer-facing bot party this year, but with the announcement of its Assistant product, its “Home” Echo competitor, and Allo, it has meaningfully tossed its hat into the ring. Google has not, however, been pushing the developer or ecosystem story as much as just creating great services—at least not directly. That might turn out to be the right strategy.
Even Apple inched toward enabling some bot-esque experiences, opening up some hooks in Siri for developers to use, albeit in in a locked-down Apple way that is focused on experience and not new products and new distribution.
This is all in addition to younger companies like Slack, which banged the bot drum hard through the year, as a way to kickstart a new platform on top of its messaging and communications suite.
Enabling Technologies
The other big reveal on the year from big technology companies was supposed to be a host of new enabling technologies, frameworks and APIs that made building bots far more accessible. Developments from this battlefront were mixed—some good and some not so good.
No one delivered a platform or framework that made building “bots” centered around natural language interactions easy for developers to actually build. Among many others, WIT.ai from Facebook, Amazon’s NLP for Echo, and even Google’s speech APIs all are far from being developed enough to be useful for developers. They are all 90% solutions at best. They are capable of dealing with specific problems, such as a timer for people to use while cooking. This level of accuracy makes it easy to plug Spotify into your Echo and play the right song a lot of the time, but isn’t close to what you need to do anything sophisticated.
Without good layers on which to build, it is exceedingly hard for people to make pure software bots that are good enough to warrant wide distribution. So it is important to note that you can’t separate the lack of real distribution in 2016 from the lack of good and useful software.
At the same time, as a developer it is worth acknowledging how amazing the 90% solutions are from a technology perspective.
It is interesting to watch Google use advanced machine-learned APIs as a lever to get people to adopt its cloud solutions more broadly. The same goes for IBM with Watson. The bot-focused, machine-learned APIs that are being released might not be good enough to plug and play with consumers directly. But they are clearly valuable and good business levers on the back end for cloud platforms.
In this respect, there is an analogy between the bot ecosystem development this year and self-driving cars. Increasingly, platforms are delivering to developers 90% solutions to very hard problems.
The bad news about 90% solutions is that they aren’t even close to deployment in a world where you need many nines of precision to offer a viable service. This is the same problem faced by self-driving cars.
The good news about a 90% solution is that it makes a great demo and can cause a lot of excitement and momentum into a space.
What remains to be seen is where the asymptote of difficulty is. I believe that each extra nine of accuracy in this problem space will take an order of magnitude more data and effort. That leads me to believe that without a step function improvement, truly viable bots might be years off, and self-driving cars a decade or more.
What Comes Next
The optimistic view is that in 2016 we saw the “fart app” era of bots that will soon be superseded by great experiences. The usual suspect companies that always rush into developing experiences on new platforms—usually unsuccessfully (the desperate newspapers, 1-800-Flowers, etc.)—have had their crack. Now more patient and serious developers will have their turn at building real things.
The pessimistic view is that the enabling technology for building good, pure software bots might be at 90%, but that each incremental nine of reliability will be insanely hard. At the same time, with big platforms focusing on video as a next battleground, and having been burned by unsuccessful initial launches, quality apps and distribution are still a long time off.
My personal belief is somewhere in the middle. I think that some vertical applications that focus on specific use cases narrowly will thrive. I also think that some services which focus on a narrow audience and can afford to mix human intelligence with computer intelligence will thrive (a bet I have been making with a talented team working on Fin for the last 18 months).
Long term, I cannot believe that we will not be naturally interacting with “bots.” But, I am doubtful that on a mass audience level we will see a major shift in the ecosystem in 2017 as 2016 at least initially promised.
This means that for Messenger platforms, there still is no clear monetization pattern that will work in the U.S., which is scary to the extent that more and more attention shifts into that format. It also means that startups will continue to face a very hard and increasingly antiquated app gauntlet to build good experiences and get distribution.
The upshot is that this upcoming year is going to just require some good old hard work for those focused on the space.