You might not have realized this yet, but there is actually a very good example of a case study you can test right now as you are looking for good examples of conversational user interfaces that actually solve practical problems.
Case study: Google Analytics
Have you noticed that you can ask queries in natural language at the top of your Google Analytics dashboard?
So how well does it work?
Well, I decided to ask the same question that Google suggests - well, almost the same question 🙂
But, that is actually irrelevant in my view. Almost certainly, people are going to ask queries which are not going to be handled by your bot, and that's completely fine.
Here are the more useful takeaways in my view.
Make the UI sensible
I think you will see that more and more SaaS tools will adopt a user interface similar to what Google Analytics is providing here.
1 Prominent question box at the top
2 Suggestions pop up as user types in a question
3 Results load in a sidebar without re-loading your page and taking you away into a new view
I just wished they also added some kind of "context" to the question, that is the question suggested changes based on what page you are viewing.
Add answer boxes where relevant, and make it prominent
Here is a more successful query:
As you can see, it is actually doing a pretty good job of understanding the question (well, its Google after all).
But also notice the following:
There is an answer box which just gives you an answer, and it is very prominent. Well, obviously I have blurred out the actual number, but the important thing is that you can see it is shown in large, bold font and there is plenty of whitespace around the answer so you are not looking around to find the answer.
Verify the actual question the bot has answered with suitable information
This is a problem with conversational user interfaces, because the bot has to reinforce what it understood without creating too much clutter.
For example, in the above image, there is a prominent title "Number of new users" and right below the corresponding date range is shown.
With these two pieces of information, you can actually be certain what question is being answered.
You should try and emulate that for every question you answer via a conversational UI.
Add suggestions to follow up
This is quite common these days, but you should also recommend more information as follow up where it makes sense. In the image above, for example, there is a list of potential followup questions you can ask which are (somewhat) relevant.
Get feedback on the answer
Given the richness of the web interface, why not also get feedback on the answer provided by the bot?
Prediction: The best chatbots will not live in messaging apps or VUI devices
Because of these reasons, I think the best chatbots (as in useful beyond one-and-done tasks like "setting a timer") will not live inside messaging channels like Facebook Messenger and Slack, or fully voice-controlled devices like Alexa or Google Home, which are inherently limited when it comes to presenting information.
My view is that they will be a hybrid of old-school and cutting edge tech, with the conversational abilities adding more and more automation into what is already a profitable and lucrative business (e.g. web apps).
Its just a prediction, so I could certainly be wrong and won't worry about it too much if I am. 😉
Which type of SaaS should use it?
I think if your SaaS has some of the following attributes, adding a conversational user interface can help.
1 You have a really large menu
If you have, say, more than 20 different options in your Menu + First level sub menu, I think that is already a lot of time users are spending searching for stuff in your menu bars.
2 There are many different kinds of reports you already display
For example, a typical email service provider (ESP) will show a bunch of reports. These are good candidates to turn into conversational questions, as people otherwise need to spend a lot of time just looking for these reports.
3 People have to click on a bunch of dropdown boxes to do a task
Once again, this is very appropriate for report like user interfaces. You can cut down on the wait time by making these queryable.
Here is a little trick that I believe can help - imagine you are Amazon, and each help page is a product page. How can you surface the appropriate task/report/help pages, in the relevant order, when the user is looking for a particular "product" by typing in a query?