Introducing the Bots Landscape: 170+ companies, $4 billion in funding, thousands of bots

I Google search, therefore I bot.

Since Facebook announced a bot developer framework and distribution platform in April, the media has been hyperventilating over its impact. I know we’re a big part of this, and I don’t apologize. Bots, as a new (or revisited) paradigm for human-computer interaction, are here, and we’re observing hundreds of companies, billions in funding, and thousands of bots flying in your browsers and messaging apps. Bots are big. You can download the full landscape here, and more rich data is coming soon.

[aditude-amp id="flyingcarpet" targeting='{"env":"staging","page_type":"article","post_id":2027437,"post_type":"analysis","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,mobile,","session":"A"}']

This article is part of the Bots Landscape. You can download a high-resolution version of the landscape here.

Bot definition and scope

Bots, or conversational UIs, are (sometimes) autonomous services humans can interact with using (somewhat) natural language in existing messaging apps — and they’re finally starting to take shape. Hundreds of companies are sprouting up across each of these categories:

• Messaging apps (current means for bot distribution)
• Bot developer frameworks and tools (including foundational bot platforms like Microsoft’s or Facebook’s, or more specialized expert builders on those platforms like Assist)
• A.I. tools with specific domain expertise like natural language or voice
• Analytics (bot instrumentation and tracking)
• Discovery (early bot stores)
• Shared services like payments or security
• Hundreds of bots garnering real traction across a few key categories like productivity or customer service

I’m partial to the definition of the space from Amir Shevat, developer relations lead at Slack. He says, “Bots are digital users within a messaging product. Unlike most users, they are powered by software rather than by a human, and they bring a product or service into a given messaging product via conversation.”

But bots that matter — and therefore, are relevant and have a chance of sticking — are only as useful as the service they connect a user to. Messaging is literally the most common digital behavior on earth right now, and the land grab for bot attention means launching a bot as quickly as you can makes sense. But those crapbots won’t stick, and the tools popping up to service these junkbots won’t either. My apologies to many of the logos in the Bots Landscape.

There will be ongoing, healthy debates around designing conversational products that are useful and relevant. Most good bots have yet to be written, and unlike a relatively common trope supported by junkbot-enabling tools, take more than a weekend hack to put together. It’s hard as hell to build a good bot — mostly because it must invoke a relevant and useful service to be any good. That takes a good idea, a good team, and most likely capital resources to accelerate it.

Thus, the industry desperately needs a set of standards and practices, and thankfully good bot makers are up to task. But much like the “meaning-of-life-42” conundrum, what are the important questions to even ask and understand before we can start scripting this out together?

Should bots sound ever more human? When? What does the spectrum of designing conversational UIs look like, since there is no singular spoken or written English language (upwards of 15 percent of Google searches are brand new every day)? How do we address internationalization and memory for text-based products? What is the right balance between GUI (graphical user interface) and CUI (conversational user interface), and how do we most effectively integrate conversation? Perhaps most importantly, when should a conversational app be the first and only right way to build something, rather than a clunky buy-me-some-stuff-on-Amazon bot mistake?

[aditude-amp id="medium1" targeting='{"env":"staging","page_type":"article","post_id":2027437,"post_type":"analysis","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,mobile,","session":"A"}']

Beats me. But we’re actively doing research and will soon be asking for your help to suss it out with us.

Thousands of bots, few memorable experiences

Facebook, Telegram, Kik, and a handful of bot authoring tools like Chatfuel like to report the tens of thousands, if not hundreds of thousands, of bots built on their platform. But what they don’t tell us is which bots are any good, which bots people are obsessed with using, and which bots solve a real need. The Silicon Valley snark factor is high.

But this isn’t a fair assessment of the real work being done within the context of this new paradigm. MZ, creators of the mobile mega-game Game of War, built a bot platform to run New Zealand’s entire transportation ecosystem. No, really.

“You won’t just be running transportation from that,” MZ CEO Gabe Leydon said at MobileBeat. “Pizza bots are great tools as personal assistants, interacting with just my data. The bots of the future will interact with all of the data and make a decision based on what everything that is going on.”

[aditude-amp id="medium2" targeting='{"env":"staging","page_type":"article","post_id":2027437,"post_type":"analysis","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,mobile,","session":"A"}']

Or there’s Michael Perry, who founded Kit, a marketing bot that builds and optimizes Facebook ads and sends emails to customers on small businesses’ behalf. Kit was recently acquired by Shopify. Perry told me, “When we built Kit, bots weren’t cool. It wasn’t an industry. We actually built it because we figured that was actually the solution to a problem we discovered. For Kit, the problem we’re solving is time. We’d challenge any human to build or distribute an ad on Facebook faster or better than Kit can.”

And herein lies the challenge. Just where do we start, from a tech-enablement perspective, to make a dent? Slack stood up an $80 million fund to help figure it out. Now bots help you book travel, manage expenses, collate data, and save customers millions of dollars. And as of yesterday, Obama has a bot. Obama!

Bots allow software to work for us, not the other way around

After sitting through more bot pitches than I can count, it’s clear to me that building a system that can demonstrate truly human-like understanding of open-ended language is beyond the reach of today’s technologies. And that’s OK. But we’ve been presented a strong starting point to reshape the way we think about using software, especially at work.

Nir Eyal, a writer and entrepreneur, wrote an excellent post that articulates this concept in steps: “Die Dashboards Die! Why Conversations Will Reinvent Software.” You should read the whole thing. Essentially, when a human interacts with software, there’s a certain cognitive load placed upon that user to understand both what they’re looking for and how to use the tool to get it. We all play this internal question-and-answer game all the time without necessarily dictating those questions aloud: What’s important? What do I do next? How do I do it?

[aditude-amp id="medium3" targeting='{"env":"staging","page_type":"article","post_id":2027437,"post_type":"analysis","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,mobile,","session":"A"}']

In fact, around 20 to 30 percent of our day is spent looking up information. The standard industry response to this problem has been drop-downs and dashboards. Some good bots are emerging to help solve it. As Eyal sketches out, wouldn’t this be an infinitely more joyful way to use Google Analytics, the most common dashboard tool on earth (next to Excel)?



Natural language will remain a massive challenge

“Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo.” is a technically correct sentence. Building software that can recognize and account for the near-infinite intricacies of language and understanding will remain a huge challenge. Language is a constantly evolving interface on its own. Wut? I kno, srsly.

But it’s not uncrackable. Ross Goodwin, a student and technologist in NYU’s A.I. lab, built a bot (a recurrent neural network — a type of A.I. used for text recognition) by teaching it dozens of sci-fi screenplays he found online. The bot named itself Benjamin and spit out a futuristic sci-fi thriller starring Thomas Middleditch of the television series Silicon Valley. And it’s awesome. Sure, some of it kind of feels like Mad Libs — but it’s more a reflexive experiment on language and creativity than an actual movie you’ll want to saddle up with. The AP is already generating “automated insights” — bot-created stories for sports and earnings reports.

[aditude-amp id="medium4" targeting='{"env":"staging","page_type":"article","post_id":2027437,"post_type":"analysis","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"bots,mobile,","session":"A"}']

How big is the gap, really, between “automated insights” and “machine-generated creativity“?

We need your help

If you have thoughts on what we should be doing with our Bot Landscape and the types of research we’re pursuing, we’d love to hear them, both in terms of what categories we should add and what companies they should contain. There’s a simple three-question form here.