Designing Conversational UI with Information Architecture — Part 4
Onboarding without arrows
Welcome to part 4 of a series on designing for conversational UIs using information architecture. This part will be about the art of onboarding when you don’t have a static screen.
Onboarding is your first date with your user — you don’t want to talk their ear off and you want to be as charming as possible.-MK COOK
This part will be the least IA focused of the series, but I wanted to draw attention to it.
Why is onboarding so hard? Part of the reason is that designers forget some basic rules while creating onboarding flows. Users of your application will have only so much brainpower/attention span to go through the laborious process of clicking through 7 onboarding screens. Which is why I am a big proponent of empty states that nudge users in the right direction.
Making a good first impression is important, but true success in your first empty state means driving an action. Make your user want to fill that screen.-Dina Chaiffetz
Material design has dedicated a portion of their guidelines to empty states.
A good empty state is clear, concise, and prompts users to take action in a meaningful way.
Empty States for Screenless Interfaces? What?
Right. So this doesn’t make much literal sense. But I just wanted to start off with empty states, because they are not always seen as an onboarding technique.
For chatbots, let’s start with how we onboard users at the beginning, before that default menu prompt.
Slackbot uses both an empty state and a prompt, knowing that many users will ignore one or both. The first set of helpful tips is designed like an empty state. It tells you what type of question you can ask by giving you an example and then asks you to add a profile picture. The prompt that slackbot then uses reiterates the idea that you can ask questions while starting off the conversation.
But how would we do this in a non-visual way?
Is there anything I can help you find?
If you’re in the US, you’ve probably had someone greet you when you enter a large retail store at some point.
“Welcome to ABCD Super Store, can I help you look for anything?”
Those of us just browsing may find it a little annoying, but for anyone actually looking for a specific item, greeters can feel immensely helpful in an otherwise overwhelming big box store. Most interactions with your chatbots and voice interfaces will be from users who are specifically looking for something. That may change in the future when we look to AIs more and more for actual conversation, but until then, let’s give our users some help.
A word of caution…
This should not turn your voice UI into an endless chat with a telephone prompt!
Thank you for calling Fanny’s Repair Services. If you’d like to repair your car, press or say 1. If you’d like to repair your sink, press or say 2. If you’d like to repair your air conditioning, press or say 3. If you’d like to repair your other sink, press or say 4. If you’d like to speak to a human being, yell loudly.
That is an obviously fake example, but not so far off from the truth. The main difference between a phone voice prompt system and your conversational UI should be a human-understandable voice. Not a voice that sounds human. A voice that speaks like a human.
Hey Chuck, thanks for coming to me with your repair problem. Can I help you with anything specific? I can do estimates, give you times I’m available, or diagnose a problem.
Where IA plays a role
It’s time to take that hierarchy and do some usability tests on it.
I’m not going to detail out the steps of testing an IA (without a screen!) but I will say that this will help you figure out how you can and should onboard your users. The usability test will point out some obvious breakdowns in your work, but for some areas, you may not be able to change enough to vastly improve your content. So instead of scrapping your hierarchy and starting over, use that information for your onboarding prompt.
Let’s say you’re testing your hierarchy and content model from your movie recommendation bot…
According to your search history, we believe you’d enjoy The Social Network, also by David Fincher. Have you seen it?
…and you discover that people don’t have any idea how you came up with a David Fincher moview as a recommendation. Well, you could clarify in your onboarding by telling users how recommendations are surfaced.
Or you may discover that usability test participants don’t generally like that recommendations are based on something they can google search themselves (like: David Fincher Movies) and want something more robust. Well, instead of redesigning your entire IA, you may just want to prime users with this information.
Hey Chuck. I’m here to recommend movies based on some information you’ll provide to me through your search history. When you ask for a recommendation, I’ll give you some based on clues that you’ve left, like movie directors, genres, and actors.
Suggested Reading:
How to Test an Information Architecture
Test your navigation with a reverse card sort
This was the hardest part of the series to write and still feels slightly disjointed. Please send me a message, leave me some feedback if anything is super unclear!
Although this is a 4-part series, I was asked about content writing for bots in general, so we’ll have a bonus part at some point. Stay tuned!
