Chatbots are the future, but they almost certainly aren't the present.
Slack CEO Stewart Butterfield went on record that bot technology has been oversold, which is a pretty bold statement considering Slack is currently the most high-profile playground for chatbots. VentureBeat has been more blunt, saying that the current crop of chatbots suck, largely because the technology isn't ready to make them useful just yet.
The purpose of a good bot is to observe what questions you post in a messaging app like Slack, Skype, HipChat, or even Facebook Messenger, and then respond with appropriate answers. This is much harder than it seems—for two reasons.
Programming Bots Requires Too Much Data
First, natural language processing requires a lot more data sampling than most bot-users or bot-builders are prepared to provide.
The gold standard for bots is IBM's Watson, which responded to questions asked by Jeopardy! host Alex Trebek and provided enough correct answers to out-play the game-show's most accomplished champions. All it took was rigorously analyzing decades of past Jeopardy! questions, and then cross-referencing those Q&As with the entire text of the Wikipedia. Your organization likely doesn't have decades of Slack posts nor hundreds of millions of pages of user guides to fuel as impressive an index of questions and answers. Thus, hoping for Watson-like success with a company-focused chatbot is unrealistic.
The typical solution to this problem is to aggregate data from several companies and look for commonalities. With enough sample data you can build a general-purpose chatbot that each organization can modestly tweak to their unique needs. This brings us to the second reason that chatbots aren't ready from prime time: no two companies are really all that alike.
Companies are Too Unique for "Generic" Chatbots
Consider a basic chatbot function like making it easier to onboard a new employee (something Talla is looking to solve). A software development company will likely have an entirely different set of common employee questions than a residential real estate firm. The former will have a number of questions about file access, version control, and (to indulge a stereotype) ping pong tournaments. The latter will likely ask about referral policies, shared commissions, and (to indulge another stereotype) where to get a good headshot for their business cards and brochures.
Now imagine that one of the above examples is a small business, and the other a major corporate enterprise. Imagine further that one of them has all its employees in one central office, and the other comprises a group of dispersed remote workers who rarely, if ever, gather in any one place. Imagine further still that one is composed entirely of full-time employees, and the other entirely of contract workers.
No out-of-the-box chatbot can cover the gamut of questions and answers this diverse universe of companies would need to address. There isn't a default set of keywords or subjects that will be remotely adequate.
Messaging Apps Need Campaign Tools, Not Bots
The better answer is to apply well known technology to this new world of messaging-based employee interaction. Campaign-building tools have a long history of improving information delivery. We may hate robocalls come election season, but we grudgingly appreciate them when they remind us of doctor's appointments. We may not like email spam, but we really appreciate those automated delivery messages from Amazon that alert us when our order has been received, processed, shipped, and delivered, respectively.
The campaign-management principles behind these notifications can and should be applied to Slack, Hipchat, Microsoft Teams and the like. If the future of work is mediated by chat applications, then the campaigns we use to inform and engage employees have to be directed at chat platforms. Slack campaigns and HipChat message blasts may not be as sexy as self-organizing chatbots, but they are a lot more practical, at least in the near term.
Sophisticated campaign tools have branching conditional logic—did the recipient respond: if yes, send second request; if no, do send thank-you note—that can work just as well in messaging apps as with email, postal mail, or phone calls. And from this relatively straightforward tool set, each company can build the specific, customized message campaigns that meet the needs of their particular organization.
Until we've all accumulated the amount of data Watson-type tools needs to succeed, we'll have to keep it elementary with some good, old-fashioned content management.