A few years back, chatbots were one of the biggest tech stories, especially for the enterprise and businesses. As a prime example of user-facing artificial intelligence (A.I.), the hope was that companies would integrate the artificial chatterers into everything from deep search within messaging apps to conversational weather bots inside Facebook Messenger. But a funny thing happened on the way to chatbots forever changing customer service and other fields: Nobody seemed to like them. Late last year, data from the Pew Research Center showed that some 80 percent of those who have heard about chatbots had a negative impression of them; contrast that with the 17 percent who believe that bots are mostly used for “good purposes.” Compound that with numerous stories of chatbots falling down on the job (The Information described testing in which Facebook Messenger bots only fulfilled “about 30 percent of requests with human agents”), and it's worth asking whether the technology is primed for history's big dustbin of dead technology. It wasn't supposed to be this way. Modern chatbots try to bring a friendly version of A.I. to the everyday user. Historically, chatbots have been linked to language processing engines such as MIT’s ELIZA, which back in the 1960s used pattern matching to create a digital conversation. It’s much the same model we find in today’s chatbots. Google’s API.AI is proof we’re still following that model. While it’s pretty easy to get a bot up and running with API.AI, developers have to program the interactions manually in order to add depth and complexity. Ask the bot a question it wasn’t designed to answer, and it might just ignore you. [caption id="attachment_139656" align="aligncenter" width="2125"] Poncho Chatbot[/caption] Facebook Messenger bots likewise follow the same model. The earliest example of a purportedly robust chatbot on Messenger was Poncho, which returned info about the weather via the persona of a cat. When they first rolled out, bots were meant to be the heroes of Facebook’s annual developer’s conference. Only they weren’t. Most reviewers panned Messenger chatbots as clumsy and useless (myself included), and they were right. The problem with chatbots-as-a-service is that they compete with native apps and user muscle-memory; we know to open a weather app because it’s been de rigueur since the iPhone launched in 2007, but texting some cat to find out when the sun will shine again is not quite as intuitive.

All is Lost for the Chatbot?

The problems with consumer-facing chatbots abound. In Facebook’s case, you have to use Messenger, find the bots, decide that holding a conversation with a bot is more useful than opening an app, then learn how to communicate with it. In their purest form, apps are simpler than bots; when it comes to business services, however, that's often not the case. Robert Johnson, CEO of the business-to-business (B2B) customer support software company Team Support, thinks chatbots might be better suited for the enterprise, and that the demands of B2B could improve bots:

While even the best chatbot technology has an 85% accuracy rate for its responses, this percentage is still too low for many businesses. In this industry, being right is often more important than getting an answer quickly. A chatbot mistake in the B2C industry could result in a single account cancellation worth a few hundred dollars, but in B2B it could lead to an error in a six-figure order.

Customer service is collectively making a big push for textual interaction with bots, which can easily handle many boilerplate issues (freeing up humans to focus on anything else). Gartner thinks that up to 85 percent of human interactions with business will be handled through bots by 2020. There’s pushback to that idea, though. As Joey Anderson of Ampersand and Ampersand wrote: “The purpose of a chatbot is not solely to automate responses to users. It is also meant to reduce the amount of more expensive interactions that take place through other, less effective methods of engagement. For example, from a consumer perspective, would you rather order an Uber through multiple interactions on your mobile device or through one simple voice command? For most people, the answer is obvious.” Anderson strikes at a newer issue for chatbots: the interface. Amazon’s Alexa digital assistant is already leading the way in this realm, powering the Echo devices found in many homes around the world. It’s always listening, so a simple "Alexa" followed by a vocal command typically gets you what you need (it may be music selection, or, as Anderson suggested, hailing a ride). Recently, Apple made Siri available for many types of communication, leaving us with the same ability to request a ride with your voice on your phone or watch.

Speed, Chat and Bots

As Johnson notes, chatbots as a B2B service should be focused on accuracy ahead of speed, as business interactions often carry the weight of monetization. B2B often arrives late to the tech game, too, which is why Johnson says “some businesses in certain industries (i.e. healthcare) are still using paper records.” Meanwhile, consumers want the right results, and fast – and can’t understand why bots are so slow. Upon its launch, users found Facebook’s own ‘M’ bot really slow to respond. (It hasn’t improved much in a year, either.) Part of the frustration may come from how we interact with bots. Humans typically talk faster than they type, so taking the time to ask a bot what the weather will be like today via text, only to be met with a response like "and where are you today?" can be maddening, especially since it demands your whole attention. But barking the same thing repeatedly at a small plastic tower in your kitchen isn’t quite as bad for spoken queries. You can move freely through the room, multitask, and even ask the device to repeat the response. For many lighter use-cases, the chat is better than the bot. [caption id="attachment_138540" align="aligncenter" width="1000"] Amazon Echo[/caption] Bots are tricky in form and factor. At their core, they are just interfaces for web hooks and APIs, which don’t necessarily need a dedicated interface. You can hail an Uber or Lyft from Apple or Google Maps, so why do you need to have a text conversation via a messaging service in addition to choosing a destination? Services like Slack might justify Johnson’s position that chatbots are better as B2B solutions. While it might be overrun with bots, Slack’s bot categories (analytics, marketing, office management, etc.) suggest that interacting via keyboard is better suited to the office environment. Slack is also making a big move into the enterprise space. So 2017 likely won’t be the "year of the chatbot" any more than 2016. We’re still mired in trying to figure out where to use bots and chat services; is it easier to have Alexa read you the news, or would you rather just poke through an app when you have downtime? There’s no right answer, which gives both voice and text interactions room to blossom. There probably won’t be an a-ha moment for bots, either. It’ll take time for them to improve their response time and accuracy but also (most important of all) for user behavior to change. We’re still in the era of chat and bots, but we should probably start considering them separately. More robust actions are better for bots with a keyboard and a screen; for example, enterprise solutions with heavy or intensive data might be better served by visualizations summoned with a simple typed command. But something like the weather needs accuracy and speed, not a clever cat; for cases like that, chat might be better. That's why it might not end up being chatbots that run the commercial world (and customer-service interactions), but voice-activated assistants along the lines of Google Duplex.