The Real Future Isn’t Bots — Its Distributed Apps
We’ve spent the last year at Talla building enterprise bots, and we have learned a lot. Over the past year, Talla launched a natural language to-do list bot, a business Q&A bot, and a Trends bot to analyze words in messaging applications like Slack. We’ve learned a lot from these endeavors, and we’ve also tried out dozens of other bots to figure out what is working and what isn’t. Here is the summary of where I think bots are going:
- Bots aren’t apps
- Natural language isn’t useful for most startups.
- Messaging platforms are going more visual.
- The future is in distributed apps.
Bots Aren’t Apps
The biggest mistake I see bot companies making is the bots-as-apps analogy. The general thought is that when new platforms arise, everything you used before moves over to that platform. Messaging is the new mobile so, the way web apps moved to mobile is the model for how web/mobile apps will move to messaging. Except that it’s wrong.
Messaging platforms are just a new UI. Think of it this way — when email came out, no one built a purely email driven CRM. That would be dumb. In a similar way, building a purely chat driven CRM is also dumb. The way to think about bots is as a new user interface, not a new type of application.
Natural Language Isn’t Useful For Most Startups
A startup that is building a bot that does 3 key things would be better served focusing on the base functionality of those 3 key things than they would focusing on great NLP. What I’ve seen at Talla is that while NLP is incredibly sexy, it doesn’t translate into any different usage patterns, and it is very expensive to build. If you are building great NLP into your product, you risk focusing more effort on your NLP input funnel than you do on the core use cases you solve, and that’s the wrong capital allocation mix for an early stage company.
It is easy to get misled by this because, if you show prospects some slick NLP, they will salivate. But think about this — the value in broad conversational NLP as an input format occurs for tasks that you don’t perform very often, and thus can’t remember the exact syntax. Yet, tasks that you don’t perform very often are poor candidates for functionality an early stage startup should be building. There is an inherent tension between these two issues.
What I believe will happen is that, in the next 3–4 years, as natural language understanding moves forward rapidly, there will be an industry-wide step function in conversational NLP for all kinds of apps. It will be pretty evenly distributed, and input-funnel NLP won’t be an advantage for anyone.
Messaging Platforms Are Going More Visual
Slack now allows buttons in messages. Hipchat has a full sidebar that can take visual input and non-conversational notifications. I expect both of these key platforms, and their competitors, to allow more code, and more visual representation, within messaging flows. This makes much more sense as a way to work than being mostly conversational.
The Future Is In Distributed Apps
What this means for app builders is that you need to architect your new applications differently. Rather than having a centralized place where a user can go to interact with your app, your app must be architected to be dynamic and distributed. It needs to be able to deliver data, or receive input, from the user, wherever that user is, at whatever time and in whatever context the user needs. The future is in microflows, and microviews, and microinputs. The question to ask is, can a user get the piece of your app that is needed, at the right moment, in the right format? It’s a different way of thinking, but I feel strongly that this is the way the world is going.
Bots are cool but, their limitations are quickly becoming apparent. Bots are simply a small part of a new emerging ecosystem of smart, distributed, context driven work.
