In January I wrote a post on bots, conversational apps and Fin. In the post I called out that it felt like 2016 was the year of a broad shift in the developer ecosystem away from apps and toward conversational interfaces and bots.
A lot of cards have flipped over in the past six months. Facebook, Google, Amazon and Apple made various announcements around bots. At least a dozen bot-focused incubators have sprung up, as have hundreds of fledgling companies of various levels of seriousness. Yet a lot hasn’t happened. None of the major platforms have yet delivered solid ground on which to build. The apps that have launched in broad public view have been disappointing, and there is no distribution to speak of.
In light of great hype and quick disappointment in most places, the hype-cycle backlash has already started.
There are a few things which I didn’t internalize deeply enough six months ago about how things were going to play out in 2016. Three big issues stand out.
Underestimated
First, building compelling bots is extremely difficult with the tools that currently exist for developers. At the beginning of the year, I noted how freeing it was for developers to be able to move away from a world of supporting multiple different client-side software stacks and toward more server-dominated bot services. This is true, but only part of the story. The rest of the story is just how insanely hard it is to get good experiences through natural language interfaces.
For those startups that are pursuing technology-driven bots with no human support, the challenge of natural language interfaces has become apparent. None of the bot frameworks that big companies have released or endorsed, like wit.ai, are close to being ready for prime time. This issue is compounded by the fact that round trips to servers versus software interactions on the phone are slower, making bots feel lethargic compared to what people are used to. I expected more developed services from the big platforms on which to build.
Some companies, like my company Fin, are pursuing a hybrid approach with human and computer intelligences working together. This can make the bot services feel a lot more intelligent, but getting it right takes a lot of time and effort to refine. After a year of work I think we, at Fin, are close, but what we do and how we got there is not obvious. And in many ways we still have a long road to travel to get to where we want to be.
Second, there is the problem that most of the bots released to date are just crappier versions of apps that already exist, pushing news and weather, ordering cars and flowers, etc. Developers have been busy at the completely imagination-less job of converting the ecosystem of known and understood apps directly into conversations.
It feels like the early days of movies where no one could think of anything to do with the medium but copy plays. Or the early days of TV where people just copied radio shows.
My belief is that the great services that will be conversational in nature will be new experiences that don’t work well as apps but can work as conversations. It turns out, as some people have wisely noted, ordering cars is a great experience via apps. But converted into a messenger thread it’s a worse version of the same experience.
Third, there is a serious interaction problem that has become apparent. There is a spectrum between software and humans providing services. People expect software to be limited in what it can do, but mostly perfect at what it does. But they know humans can do a broad spectrum of things but make mistakes.
The hybridized feel of conversational apps, where users rightly ask “am I speaking to software or to a person,” creates a strange set of expectations for service providers. If you can’t do what a human does, people consider the service dumb. If you, as a hybrid service, perform complicated tasks that a computer couldn’t but you make mistakes that software wouldn’t make, people consider the service a failure. This consumer component around expectation setting and how to interact with a service poses a big challenge for conversational interface services.
The upshot is that I still believe 2016 is pretty clearly the year with a big push around conversational applications on all fronts, but building good experiences is harder than it looks.
Still True
If those are the big areas I missed, I think in general my post from the beginning of the year got a few of the biggest issues and themes right.
First, I noted six months ago that the biggest liability in terms of ecosystem development was a business model question. Developers need to know how they’ll get paid to provide bots. Developers also need a clear sense of how the bot ecosystems will benefit the platforms on which the bots sit. Otherwise no one is going to trust the bot ecosystems enough to develop on them.
The business model question has not been clarified. As a result, there is hesitation to engage deeply. And, understandably, without a business model that makes sense, the platforms are providing zero meaningful or trustworthy distribution to bots, which just further turns off development.
The launch in 2007 of Facebook’s original platform was poorly thought out from a long-term business perspective in many ways. But it offered enough distribution to developers that it prompted a gold rush nonetheless. This time around, no carrot or stick has been offered.
Second, as I pointed out previously, the incentives to make bots happen continue to be very strong. Six months later, from basically zero, at least four major western bot platforms—Facebook, Google, Apple and Amazon—are at war. Interestingly, this doesn’t feel like a winner-take-all market, but rather yet another situation where developers will have to be present on all ecosystems at the same time. That means there will be a war of protocols, discussions of standardization and coalitions, etc. I believe that these will end up looking a lot more like web-standards discussions than like mobile operating system discussions. The platforms and developers continue to be aligned in favor of this shift.
Third, as I suggested it would be, consistent engagement is an issue. From what I’ve heard, one of the biggest issues bot developers face is the split between a tiny minority of hyper-engaged customers, and the long tail of users who try their service but who they can’t retain. This is what happens when you don’t get on the home screen as an app. Because most of these bot-ish ecosystems lack any sort of permanent home which apps can vie for—and at best must compete generally with friends and family, who are far more engaging—it is extra hard to build retentive services.
Conclusion
Lest there be any confusion, I am still very bullish on this shift. The move from clicking on icons to simply talking to great services is part of the technological imperative. It is a consistent trope in science fiction of how the world will work, and what people want. To believe that this transition will not happen is fool-hearted, akin to believing that VR will not “eventually” be a space.
That said, it is interesting to see the hyper-fast boom-bust pattern on bots generally. Just like the first major tech crash of the early 2000s, we have already seen the excitement around business-to-consumer bots drop and the rise of a business-to-business narrative, where increasingly people are pitching business-oriented bots. We have also seen some of the platforms quickly moving toward more structured Q&A type interactions which are easier to manage for developers. There is also a clear pullback in funding.
Just like that tech crash, however, it is pretty clear that the fundamentals of the experience and technology will emerge. Some companies will pull through the boom-bust cycle to get the experience right. If in the past six months we witnessed a bot era equivalent to the web 1996-2001 era, I believe the next year for bots will be like the web between 2001-2007.