If there’s one lesson that has been validated in every conversational UI product I’ve worked on, it’s that building the interaction is just the beginning. Testing and iterating are the most time-consuming and headache-inducing (but worth the effort!) parts.
Here is what I wish I’d known about the design process:
1. Build with a clear objective.
This is a perfect example of defining a simple objective. The three steps to the objective can be determined later. (Credit: User Testing Blog)
Remember: a bot excellent in one thing is infinitely better than one that is mediocre in three things, so be picky with the objectives. Unlike mobile apps, users stay on one screen, so there isn’t as much leeway to note a change in context by switching states.
The goal of your chatbot should be apparent throughout the decision process — from the copy to technical integration options. These objectives will come from discovering who the users are, their lifestyle, and pain points.
2. Guide users towards this goal, through copy more explicit and straightforward than you may be comfortable with.
This user task flow for web contains the decision points applicable to chatbot products as well. (Credit: Nielsen Norman Group)
Define the different steps in a user journey, and how the bot should be programmed to respond at each of these points.
Pro-tip from a UX Writer: Be minimal with your copy length and simple with your language choice.
"Make the time required to comprehend each sentence ridiculously short, and the time and effort required to take action even shorter."
Try different phrasing to see which one is the clearest for users.
For example, instead of “I don’t understand,” how about “Get help” so it’s actionable?
You may find that the first option is preferable because the second option requires the user to guess what they should do with the bot. Instead of making guesses yourself, try different options and ask your users to think out loud.
3. Experiment with different questions about the users.
In my experience, a bot persona that truly fulfill your user’s needs requires both what I call “flat data” (age, gender, occupation, etc.) and dynamic data that are situation-specific. The latter type of data could be: their work routine, the thing they hate doing the most at work, etc.
Knowing the user’s typical schedule allows you to design a product that complements this schedule. (Credit: bananaworks)
Knowing users’ schedules and habits in a specific context will help you design a conversation that they could enjoy having, not just endure for productivity’s sake.
For example, if you know closing out tickets is the IT technicians’ least favorite part of their job, you could have some humor when they’ve gone through all their tickets:
???? “Hooray, it’s only downhill from here!”
When they start with their first ticket, you probably don’t want to have this kind of straightforward humor.
I was inspired to incorporate questions about the various situations users get placed in by the various “day in the life of” videos from YouTube creators. As you follow along their day, you get a sense of things like what they look forward to doing, what they put off until the very end, and how they wrap up their day. In an ideal world, you would be able to “shadow” your user to understand their day-to-day life.
With this kind of information, you can design a bot that complements your users’ lifestyle, not just a sliver of their day.
"A functional interaction only requires efficiency of task completion. A pleasant interaction requires efficiency and having desirable emotions along the way."
4. Gather KPI’s that will actually allow you to make changes.
Like mobile app products, data points such as the percentage of returning users don’t do much good without knowing which task had the highest rate of completion and which had the lowest.
However, because conversational UI relies on technology such as natural language processing, it requires that you also have data such as:
- When a bot’s answers don’t match the trained answers
- When the bot keeps the user inside of an endless loop of error messages no matter what they type, because it’s not properly matching the user’s answer to the question
By knowing when these unintended actions happen within a workflow, you will be able to re-design to always provide users with a choice of action.
- The kind of questions a user asks that the bot was not designed to answer
This will help you determine efforts to expand the different steps a user goes through to achieve established objective, or determine a new objective if users’ interests shine light on an objective you hadn’t previously considered.
Knowing what users are asking for is essential in meeting that demand through design. (Credit: Klüg analytics dashboard)
5. Know the strengths and weaknesses of each medium.
Take advantage of visual aids when you can. They take shorter comprehension time and are more likely to be remembered:
Edgar Dale’s cone of experience can be used as a guideline for how to display information to favor our retention abilities.
When choosing the type of information display, think about what will enable fast comprehension and action.
This article was originally published on Diana's Medium page.
