Conversational interfaces. Can we talk?

When designing conversational interfaces, give your brand and users what they deserve.

tl;dr

Conversational interfaces are definitely going to be a design trend of 2017. They are going to make you look cool, win awards and garner admiration from design colleagues and clients. Additionally, some will be really, really great.

But before you grab a slice of this, it’s wise to think carefully about your brand and your users and make sure you’re tasking yourself with first building what is best for them.

Be approachable, of course, but be honest and accessible. Your brand will benefit when your users get their tasks completed simply and effectively.

What do we mean by conversational interfaces?

There are three basic design patterns that are inferred when using the term “conversational interfaces”. Without any real common terminology, I’ll “make shit up”:

Pattern 1: The statement filter

This is the easiest to implement as it simply re-phrases a form into a statement paragraph. It takes some of the — ahem — formality out of forms and hopes to make a boring task feel like it’s a common goal of the user and the site, and not just questions asked from above in a stiff input form.

What is problematic with this pattern is that it really is still the same boring form, it’s just less readable. Everything is a little harder, from telling at a glance how many fields need to be acknowledged, to what the current values are, to how inputs are differently labelled. These are all things that we understand immediately with normal row-label-input form structures. There’s no system one thinking to ease you into understanding the task — it’s all transferred to system two thinking by requiring you to read the whole statement. That might be cute the first time you see it, but it ends up feeling pretty old once you return a few times and miss that an option has changed.

Pattern 2: The chat script form

To my mind, this pattern actually has the most to offer. It’s a simple questionnaire that asks one question at a time. Each question could have multiple outcomes or it might just be a linear process, but the effect is the same — the user focus is put on the current question to be answered, future questions are hidden and past answers are only visible so that you can choose to revisit them.

But what makes this conversational? Haven’t we had interfaces focused on “one question at a time” for years? Well, it seems it’s partly down to the design. This pattern is often used to replicate a chat script, with questions on the left and answers on the right. Add the historical question/answers and a speech bubble or two and suddenly you’ve created a conversational interface that looks a bit like a chat or messenger transcript.

There’s nothing essentially wrong with this pattern (despite it’s re-clothing of old ideas), unless you mix in some of the fantastical elements that we cover later in “Unreal people”.

Pattern 3: The open question

This is a real client-pleaser, but it’s pretty rare as it requires some balls to pull-off technically once you move from a demo to real-user input. A single open question input to kick off a conversation is the most perfect interface — the Google of interfaces in fact — and it’s the essence of the chatbot style interaction. It forces no restrictions and there’s no slicing up users into defined types of audience, the visitor just starts the conversation with their opening question.

Sometimes it’s only used as an opening gambit and listens for basic keywords before quickly transitioning to pattern 2 (the chat script form), but if you’re super ambitious, you can keep asking for open text questions to every response as if it was a “real” chatbot.

If we’re honest, the main part of this is just new paint on an old friend, the search field. What ever process happens in the background when you start typing, it’s probably just looking for keywords (and if you’re shit-hot, intention and language) before using a look-up table to deliver a result. It only becomes “conversational” when you have a question/answer process as part of the results. At which point you might need to ask yourself if your process needs so many steps after the user had an open text field to describe their needs, is it so efficient? Or are follow up questions just filters in disguise?

This website is not your friend

You may well have a brand that is young, social and trustworthy. You may well have a target audience that you understand very well. If you don’t have these things, please be considerate of the level of familiarity in your tone of voice. This is true of all marketing, applications and websites, but does become a big venn-diagram-type intersection with conversational interfaces.

Large demographics of many audiences do not appreciate overly friendly language, especially involving tasks that may be important or dull. Apple Siri drives me mad when it fails to set a timer three times in a row, then it cracks a joke on the forth. It just reminds me that someone signed-off on more dev time for the wrong priority.

I do not need a utility, telecoms or charter-holiday company as a friend, I am not looking for a “brand relationship” before the service has proved itself to be simple, expert and execute tasks reliably. Appealing micro-copy is the first sign-post to that building of brand trust, not the foundation or proof of it.

On a more positive note, professional, concise, courteous and clear copy is obviously a great start to letting people discover your services are just as great as you tell them they are.

Unreal people

The tonality of a conversational interface is not just in the language, but in the messages inferred in the design too. If you have a stock-photo avatar of a customer services person at the top of your interface and an opening gambit of “Hi I’m Mandy, how can I help?”, some of your audience — even if for just a few seconds — will believe they are a communicating with a real person. Don’t kid yourself that all people are post-modern enough to instantly understand it’s just an image acting as a comfort blanket.

Another part of the audience will understand that they are just filling out a form and will wonder why all the theatrics with the cute picture and random names. Does it create a better relationship? Or a sense of uncanny-valley?

Real people can interpret and go off-script. Salespeople and customer service operators can tailor their information and offerings based on what the customer is saying to them and how they’re saying it . That’s the actual benefit of good customer service help-lines. Your app or web interface will have limited options in this regard, so try not to mislead the customer into thinking they’re getting anything more than first-line (even if it’s first-rate) support.

The art of conversation

Conversational interfaces suggest that users are fed up with characterless online interfaces and yearn for websites to have the personal touch you get with one-to-one contact.

The first part is certainly true, but the second is quite a curious assumption from a group that is supposedly as introverted as website developers. I consider myself relatively well adjusted on the social scale, but I’m normally (depending on task) less likely to call a help-line for most tasks for several reasons:

• Online solutions — such as forms or faqs — reveal what I need before I start the task, so I’m less likely to be surprised with a requirement during the task (see “Discoverability” below).
• I’m English, so I’m much too likely to defer to someone else’s opinion, after which I feel I didn’t get served correctly (anxiety about “experts” making me feel dumb or a sucker).
• Also because I’m English, I’ve had way too many experiences of human error, i.e. confident sounding operators failing to follow-up or screwing up their own interactions with CRM systems (See “Unreal people” above).
• I live in Sweden, so my no-so-great Swedish is a barrier to successful task completion (and another cause of anxiety).

In short, I don’t like being forced into a conversation — even a fake one — that’s not of my choosing, for logical and not-so-logical reasons. I understand that conversational interfaces often position brands as more approachable and helpful, partly in effort to reduce my anxiety. It’s just that most time as a user I don’t give a damn about the brand’s projection that it’s a good listener when I have a task to carry out. It’s about the successfully helping the user, not feeding the brand personality.

The last point about language leads also us on to…

Accessibility

Accessibility in websites can take many forms, and one that isn’t mentioned often is fluency in a language, whether by proficiency, dialect or unfamiliarity with informal language.

When I fill out a Swedish online form, I need to parse technical words, often outside of my personal dictionary. But when I parse a Swedish sentence that’s written in an informal “chatty” style, I have to parse the individual words, the grammar/structure, and finally the intention. All before I can determine how to respond. Give me a break.

The world is full of immigrants and people doing business across borders and barriers. Ask yourself if you’re making an interface for you and your peers, or if you can make one that is truly inclusive.

In addition, accessibility for screen-reader software and users with differing abilities are also still just as important as with other types of interface. But you might need to allow a little more development time to make conversational interface conventions truly accessible, for example hidden questions or revisiting answers in a chat script form pattern.

Discoverability

The funny thing about conversations is that you don’t know how they’ll progress before you start. You’re not sure about the depth or breadth of the available knowledge or abilities, or how long it will take until you come to a conclusion.

If a conversational interface isn’t considerate of this, it will fail to offer discoverability to the user. In most interfaces everything from headlines to link text to labels expose the needs, options, requirements and quite often the size of the task. Without a reasonable amount of these indicators, it will be up to the users to poke around and judge the capabilities and limits, before they need to back up and try again. Conversational UI needs to help users set expectations at the right time, rather than keeping the scope hidden until the user starts to give “right” answers.

This is especially true of user progress. A user will generally give you grace to answer a couple of questions without exiting, but they quickly get to the point of wondering how many questions are coming, and it becomes an overriding worry. Make sure your users don’t hit that point without re-assuring them.

Conclusions

Conversational interfaces — just like any other design trend — work best when they’re done with consideration for the right reasons. What can a conversational interface offer your brand as a differentiator after you have fully considered the user’s needs, not the other way around?

To promote the user’s ability to complete a task with confidence is inherently a brand building exercise, and additional personality and conversational interactions only enhance this if they’re honest and helpful in assisting the user in a delightful way. But considering the delightful aspects of conversational interfaces first is not in anyone’s long-term interest but the design team.

• Speech bubble icon Creative Copyright James Fenton — The Noun Project
• Other images from unsplash.com