Zendesk Z Bot

The challenge

The best chatbot experiences feel consistent, no matter how, where, or why a customer starts the conversation.

But for Z Bot — the first port of call for Zendesk users looking for help — delivering that consistency was no easy feat. The initial launch included over 100 conversation scenarios (or intents), serving 24 customer time zones in 12 languages — and too many entry points into the bot experience to count.

Despite the complexities, the goal was clear: deliver a modern, best-in-class support experience for Zendesk customers.

My role

Developed conversational principles and bot persona
Created voice, tone and style guide
Scripted high-value intents identified for launch
Reviewed script and design of remaining intents drafted by Sales and Advocacy teams

The process

Launching a bot with over 100 intents required a multi-pronged approach, including:

Making sure our bot design and language worked together
Using tools and techniques to help keep our teams, tech, and tone of voice aligned
Simplifying our bot messages so they worked for all languages we support

The solution

Design and language, working together

We quickly learned that making decisions about what language we wanted to use without the context of our structural design just wasn’t possible.

The welcome message is a great example. From a customer experience and content design perspective, it should be one unified greeting. But on the back end, we had to break it up into smaller chunks to make room for all the things that happen in between those chunks: data calls, branching based on entry point or persona, and more.

I came up with three phrases that, when put together, still felt like one unified greeting — “Hi there. I’m Z Bot. How can I help?” — but still gave us room for data calls in between each phrase, and were flexible enough to work for all branching scenarios.

An illustration of how the greeting message works for different user personas

And that’s just the welcome message. This kind of complexity existed for every part of the conversation, from asking customers what issue they want to solve, to gathering feedback on whether they got what they needed.

Considering design and language hand in hand was crucial in conquering this complexity, along with other approaches like a consistent structural design, modularity, and reusable templates.

A conversation flow I mocked up to identify where we could reuse messages

Tools and techniques to keep teams, tech, and tone of voice aligned

We’ve all had a bad chatbot experience before, whether it’s one that promises you the world but doesn’t deliver, one that doesn’t know how to “human” and expects you to speak its language, or one that’s just plain inappropriate.

As I began developing a persona for Z Bot, it was important for me to think about the values and ideals it should uphold. The resulting list took into consideration Zendesk’s existing brand principles—along with conversation design best practices—to create a set of guiding principles that provided a solid foundation for how Z Bot should behave.

With multiple authors, it was important to create guidelines to ensure Z Bot always sounded the same. And these guidelines needed to be fit for folks who weren’t necessarily word nerds like me.

We already had product content standards in Zendesk’s design system, Garden. On the whole, they’re pretty accessible to a layperson. But after trying to shoehorn our tone framework directly into the bot writing guidelines, it became pretty clear that I needed a different approach if I wanted to empower authors to get the job done without getting bogged down in the mechanics of language.

I reinterpreted our tone framework into a more lightweight form so authors could feel confident they were using the most suitable tone for their scenario.

The Content Design team’s tone framework reinterpreted for Z Bot

If we did have to refer to grammatical concepts, I tried to spell them out in plain language and show exactly how they might look in an intent.

Defining verbs and nouns for an everyday audience

It was also clear that with so many authors across multiple teams, I needed more than just static guidelines for how the bot should sound.

Examples were going to be super important to help convey the language recommendations and make sure we were putting a single cohesive bot voice out into the world.

I not only provided dos and dont’s for each of our guidelines, but also a word list that outlined how to use (and not use) words that cropped up frequently in intents.

From principles and to tone of voice to guidelines for specific elements like messages and buttons, the style guide covered a lot. So it was useful to condense the most important stuff into a simple checklist. When reviewers were assessing an intent, they could use the checklist to make that assessment quickly and easily.

Bot messages that work for all languages

The strings we wrote for Z Bot in English were to be translated into a number of the other languages spoken by our users. That meant we needed to think globally.

In the bot message pictured, we had to convey that an agent would be with the customer shortly. In a nod to Zendesk’s Danish heritage and our occasionally charming tone of voice, we added a throwaway joke about what the bot would do if it had that amount of time to fill.

“If I had 10 mins, I’d learn Danish” became “Within this 10 mins, learning some Danish could be a good idea”

All well and good, but our wonderful Globalization team informed us that when it was translated into Simplified Chinese, the meaning changed from a joke about what a bot would do if it had time to kill, to a directive telling customers to spend their wait time learning Danish.

It was an important reminder that Z Bot needed to speak in a way that was appropriate for everyone. (We nixed the joke.)

What’s next

Since launch, we’ve had a wealth of customer interactions to review and learn from. And we’ve made — and are still making — lots of improvements to Z Bot based on what we’ve seen.

A bot might not be a living, breathing person, but it still needs regular maintenance to perform at its best. Watch this space.

Thanks

To the other humans behind the bot:

Elissa Tikalsky, Maddie Hoffman, Austin Lacey, Terrence Wesolek, and the Sales and Advocacy teams
Han Li and the Globalization team
Chelsea Larsson and the Content Design team