Mar 14 2017
Any over-hyped technology comes with suitably hyperbolic statements of how it will impact and change everything. Chatbots are no different. At Deeson, while we are very excited about what chatbots can do, we do not believe that chatbots will replace websites, replace apps, or become the defining future of web development.
Chatbots, or better still, conversational interfaces in general will become a powerful and important additional tool in our toolbox. A way to better engage and solve a user’s problem - provided the right context.
The right question around chatbots is not whether 2017 is the year of chatbots, or whether it is all a fad. The right question is simply: what is the appropriate context for a chatbot solution as opposed to a website or an app? If your problem has the following characteristics then chatbots are most likely a very good way to solve it.
Chatbots provide a level of immediacy that is simply not possible on websites or apps.
The friction of finding an app in an app store, downloading it and then starting it up is often too much. It is no surprise that as app designers we have to deal with users’ app fatigue.
If you are visiting a museum and want to know what time it closes you are not going to install an app. Similarly you might feel that the website probably has the information but you’ve been burned so many times trying to find a simple piece of info that you simply don’t want to try again. You know you will end up having scan around some crazily designed page or fish around some strange menu to figure out where closing times are.
Instead with a chatbot the interaction is resolved to:
“What time do you close today”
“Today, Tuesday, March 14th - the museum closes at 6pm”
The museum bot will then slowly go down our contact list in the messenger app and we can call it back up whenever we need it again. Straight to the point, minimum friction, ephemeral.
Tightly related to ephemeral interactions it is important that we are trying to resolve a specific problem within a tightly defined context. This allows the chatbot to zero in on the exact interactions required and provide a resolution quickly.
For example, if we are about to go for a jog or a walk outside we can ask a weather chatbot:
“What will the weather be like in the next hour or so”
“It doesn’t look like it’s going to rain, but some clouds will hang around”
The chatbot can make a safe assumption that if no location has been specified we are looking for a local weather update and solve that problem immediately. It also allows us to zero in on specific words to guess the user’s intent (“weather”, “next hour”).
A specific scenario we are exploring at Deeson is the combination of beacon technology with chatbots within the context of cultural tours or museum visits - this allows us to determine the user’s location.
Given the well defined context of, say, a museum visit, we can provide a chatbot that answers questions relevant to an exhibit the user is looking at. It avoids providing a dry and often long summary of the object in question, and can instead add some interesting historical context or connect it to other exhibits.
Chatbots rely on natural language processing and the bot developer’s own training of the bot to understand the intent of phrases a user types in.
In completely generalised open-ended scenarios we will inevitably run into situations that we can’t handle. If, however, the domain is well defined, then we stand a good chance of adequately covering the most likely ways a user is going to ask a question.
Similarly, from a user perspective, they are more likely to stick to the usual terminology and, crucially, the expected scripts. Our tendency to stick to specific behavioural scripts, developed through social norms, is a well-known human psychology characteristic. As humans, we actually find deviations from such script strange and disorientating.
A typical example is the restaurant scenario. The minute we walk into a restaurant we become actors in a play. We know we are supposed to say how big our group is, and then wait politely while we are seated, we will receive the menus and the waiter may offer the day’s specials before leaving us to choose, etc.
As bot designers we can take advantage of this very human characteristic and avoid upsetting scripts. Our users have already been trained in the scenario we are interested in through years of human interaction. Let’s stick to it.
We are not going to stop using websites or stop building apps. What we need to start doing, however, is view our problems through the additional lense of chatbots. Is it a situation that requires short-lived interactions? Is the context well-delimited? Do we have a clear script or set of terms that we can safely rely on the user to use? Then chatbots are very likely a great fit to our problem.
About the author