Why bots were D.O.A but messaging is still thriving

Spela upp

In early 2016 Facebook released an API for Messenger and introduced the age of bots. Quickly it became obvious to everyone that bots were the next huge trend, startups got funded, Medium posts were written and the space was officially hot. I’d been doing messaging for about 8 years at the time and the it wasn’t clear to me that bots would work. Other people that had experience with messaging didn’t jump on the bot bandwagon either.

It’s safe to say, 3 years later, that the bot era hasn’t happened. I don’t think it’s coming soon. This podcast digs into why bots didn’t make it between 2016–2019, and why messaging is still thriving. The reasoning starts with the boring basics, but gets more advanced. So please stick through obvious stuff.

The first big problem with bots was the hype. You could see it from a mile away. Every tech publication featured articles titled, The Bots are Coming with images showing 1980’s robot toys explaining how every interaction that we were having on the web would now take place with a bot.

Oh you want to buy shoes? Soon you’ll just message into Zappos with your shoe size and their AI will automatically pick out the best shoes for you. Then you can just press a single button and complete the purchase. They hype was outrageous. These interactions would play out everywhere, from doctors offices to ordering pizza.

It was clear that tech was thirsty for the next big thing and the public oversold in the process. Of course Facebook and the promotion around the Messenger Platform launch played the biggest part. But hey, it was early 2016 and Facebook was on a winning streak, riding high. It was still before the other (Russian) bots came.

Which brings me to the second reason that the bot-craze crashed and burned — the word “bots”. I couldn’t think of a name with more baggage hanging on it. Russian bots ruined the American election. Twitter bots are spreading fake news. And both social networks are now cracking down on bots on their platform.

There are also the bots coming for your job (a lot of headlines when I was googling for this post). The robots are coming for your job and if they don’t get you, AI bots surely will. Until Facebook coopted the word, “bots” were just spam accounts on social media.

Did we mention AI? A few paragraphs back I talked about sending a message in to a webstore and that store knows exactly the right product for you. How will that happen? AI. How about when I message in to reschedule my delivery. How will the system understand my request? AI.

The tech press projected bots as the coupling of messaging with AI. The problem is that AI doesn’t exist. For some reason when the interface changed to messaging, AI came closer to possible. But this doesn’t make sense. You’ll know when it’s possible to produce the 1 correct answer to any query because google search results will just include one link.

Positioning bots as AI, spreading this through the hype cycle and at the same time conflating messaging interactions with Russian troll farms attacking American Democracy caused a little cynicism and backlash. But none of these reasons are why bots were DOA and messaging is still a strong and growing channel.

From the descriptions above, a simple explanation of a bot is that a user can text in, there will be some logic to understand what user is saying and the bot will respond back with the correct answer. This type of interaction lends itself to customer service or the actual usage of a product (what I call customer operations). Specifically this interaction is started by the user, it’s not meant to be driven by the organization. Because these bot conversations are directed by the user they aren’t marketing.

I want to quickly distinguish another approach that I call messaging as a marketing channel. The idea of using messaging for marketing means building an opt in list and then sending outgoing messages to this list. This is exactly how email marketing works. From all my experience, this is where messaging makes the most sense. This list building and activation approach is not what people are thinking when they use the word bot.

A simple way to distinguish the two approaches, when a user directs the conversation it’s a bot and when the organization is driving the conversation, it’s a marketing use case.

With this context for bots, here are the specific problems that kept bots from taking off:

• It’s extremely complicated to let the user drive a conversation, understand understand that user and respond appropriately. For the user to find value the scope of conversation topics must be wide enough to address anything the user might ask. It’s just an incredibly hard problem to be able to parse and respond to so many potential inquiries.
• Compounding this core problem described above is that the bot must be correct. Even a 5% error rate will be noticed. On messaging channels, it’s not particularly easy to help the user when there’s a problem. Sending the user a menu/list is a horrible UX (and contradicts the reason to use a bot) and starting the conversation over would super frustrating.
• Focusing on customer service or customer operations (like calling an Uber) use cases are not easy places to build & test MVPs. These use cases can require deep integrations, and the stakes are high in these interactions. A frustrated customer failing to get an answer from the customer service bot is a much worse situation than a marketing lead not converting. Marketing departments are constantly testing and failing ideas — customer service, not so much
• Which leads to the final point, customer service (and potentially customer operations) aren’t revenue generating divisions A hit strategy in marketing can drive growth. So marketing departments are looking to innovate, have test budget and will grow the channel quickly when something works. An amazing customer service bot might reduce cost, but it won’t significantly increase growth.

In summary, bots were a new interaction that tackled incredibly hard problems in high stakes company functions with little room for errors. Bot makers were pitching to teams that aren’t built for innovation and even if things work amazingly well, the upside is limited. That’s why bots were DOA.

This was a hard article to write. If you’ve made it this far, listening to the podcast might explain some of the ideas from a different angle. I hope you’ll subscribe on itunes.

Finally, I think that AI and customer service automation will start working soon. I just don’t think it will happen on messaging first. It’s more likely that call centre audio + AI is the place where humans will first start to interact with bots, if we aren’t doing that already.