Building a bot? Beware system-speak

Does this sound familiar?

Welcome to XCo. To hear your balance, press 1. To change your account details, press 2. For our business accounts, press 3. To talk to one of our advisors, press 4.

These were the dark old days of ‘voice user interfaces’. Sitting on a phone to your bank or energy company for hours, being forced through phone trees until you were eventually allowed to talk to someone (or lost the will to live).

But lots of modern bots still run like these old phone trees

You start talking to a virtual assistant and you’re greeted with options: I’m XBot. Would you like to check your bill, change your account details or switch supplier?

As a user, you’re not having a conversation. You’re being corralled.

It’s an example of what I’ve started thinking of as ‘system speak’

All the little tells that remind the user they’re talking to a computer with rules, parameters and limits.

It’s the difference between choose any of the following options… and so what would you like to do today?. Or to try that again, type ‘try again’ instead of shall we try that again?

It can even come through in your sentence construction. Using the passive voice (you’ll be connected to an advisor) screams robot. Most of us would normally say something like let’s put you through to someone.

So why is it an issue?

Now, this isn’t the time to try and pass the Turing Test. I’m not saying the bot should try and trick the user into thinking it’s human.

But conversational design experts say it should get as close as possible: we all judge conversations by how co-operative they feel. And trying to train the user to follow your rules with phrases like for X, type Y? That doesn’t feel co-operative at all.

In the words of James Giangola, a conversation designer at Google, the subtext of this sort of system speak is: ‘this bot is not intuitive. Proceed with caution. Follow my instructions… or else.’

And it’s really just a symptom of some bigger problems

System speak tells users two things: first, that your technology isn’t smart enough to understand me if I try to talk to it in the way I want to – your natural language processing isn’t up to scratch.

Second, it tells me that you need to train the user to use the bot. In other words, you haven’t really designed a conversation at all. You’ve designed a flowchart. And you’ve missed the whole point of a ‘conversational’ experience: it should be intuitive.

So start by speaking each task through

Your start point shouldn’t be processes or journey maps. It should be conversation.

In writing workshops, we often tell people to speak their writing aloud to hear how it sounds. With bots, you need to flip the order: speak it through before you write a word of dialogue. One person be the customer, one person be the bot. If you were face-to-face trying to do this – book tickets or change your account details or whatever – how would it flow?

Record the chat and then try to design the bot to follow it as closely as possible. It’ll feel much more like a genuine, co-operative conversation. And much less like a dalek.

If you need a hand, you know where we are.

Our ideas and freebies, straight to your inbox

Got it. You’re on the list.