Six success factors for workplace chatbots
Consumer and workplace chatbots share the same technology, but there are some fundamental differences between the two. Consumer chatbots tend to be narrower in focus and deal with high volumes of shallow requests. Workplace chatbots, by contrast, have to deal with a diverse range of tasks, and these individual tasks may be more complex too. Chatbots are not yet ready for casual implementation, but with the right groundwork they can add value.
Many of my clients have been asking how chatbots can best be deployed, so I’ve distilled my advice into six things you need to have in place for a workplace chatbot to succeed.
1. The chatbot is available where you need it
The chatbot should be accessible in an interface that’s already open, typically a messaging tool such as Skype, Workplace by Facebook or Microsoft Teams. A major appeal of chatbots is they allow users to perform a ‘side’ task without having to open up another application. For example, during a chat with team members on Slack, you may decide you need to book a meeting room for later in the day. Doing this within Slack rather than a separate room booking system helps maintain focus.
The anti-pattern for this is requiring people to use a dedicated chatbot app. If you push people to do that, they may as well go directly to the room booking app.
2. Maintainable by anyone
Content will die if the person responsible for it can’t make direct changes. We saw this in the early days of intranets and websites, when a required change meant a call to the agency that built it. When the budget ran out, everything got stale.
Meeting this criteria is a challenge for current chatbots: the ones that are simple to maintain tend to be rule-based and don’t perform very well. But content owners in HR and Facilities, for example, aren’t going to be able to structure complex dialogs, so selecting a friendly management environment is essential.
3. Bots that perform transactions, not just provide information
Bots that just give you information add very little over a search box. Indeed, given the current challenges in delivering effective enterprise search (see many of the excellent posts by Martin White), they will probably make it worse due to the ambiguity of a natural language query. The real values comes when the bot can say “Boardroom 2 is free at 3 pm, would you like me to book it?” and all you have to do is respond “Yes”, leaving the bot to do the rest.
4. Universal capability
A workplace chatbot should handle a wide range of requests rather than having multiple chatbots for each type of service. If someone tips coffee on a projector in a meeting room, people won’t know if they should talk to IT-Bot, Facilities-Bot or Mop-Bot. Right now, we’re seeing a lot of bots that are pilots by one corporate function, such as HR. That’s fine as a pilot, but the long-term goal should be to extend its capabilities, not force complexity onto the user.
5. Presents answers, not links
If a primary goal of the chatbot is to accomplish tasks within the current interface, then links to other places should be avoided wherever possible. Of course there will be times when going to a dedicated system is the right thing for the user to do: if the interface is too complex for chat-based execution, for example. In general though, this principle means content should be re-factored to work well in a chat interface. It should be short and action-oriented, and perhaps presented as a card. If your chatbot expects the user to open a PDF in a mobile chat app then you need a long hard look at your strategy (or a session with Eliza).
6. Allows for true conversations
Not all chatbots can sustain conversations, but they should at least handle anaphora — words like ‘it’ and ‘that’ used to refer back to previous context. This is where many rule-based bots fall down. For example, if the bot tells you “The rooms available at 10 am are: Boardroom seating 12, Creativity2 seating 8 or Closet seating 2” you should be able to reply “Book the first one”.
Bots should also allow the user to backtrack if they want to, in the same way you might jump back to an earlier field in a form. For example, if you are partway through a booking and say “make the start time 11 am instead”, the bot should handle that without being too locked into a linear workflow.
Should chatbots be proactive?
I toyed with adding ‘Proactive’ as a seventh factor, but decided it was merely a nice to have. Some bots do a good job of making a suggestion without you initiating the conversation. For example, if you ask a question in a group discussion, the bot might interject and say “here’s an answer from two weeks ago”. But it’s hard to pull off without that being disruptive much of the time. Whenever in doubt, ask yourself “what would Clippy do?” — then do the opposite.
This article was previously published at CMSWire.
