When I was a teenager, text messaging was mostly for updating my mother. “I’m home :)” I would painstakingly type out on the crunchy number pad of my second hand Nokia phone.
Today, messaging has evolved far beyond 160 characters. It’s the second language we speak across a growing number of platforms — from desktop to mobile, Slack to Snap. It mimics face-to-face connection, where a wink and a grin is replicated by sending just the right gif.
For businesses, messaging is naturally the place to have meaningful yet casual conversations with their customers.
But imagine being the customer service agent handling these interactions. You’re in hundreds of one-on-one conversations across multiple channels — Facebook Messenger, WhatsApp, WeChat, Twitter… As your company grows, so does this dizzying list of platforms. Sometimes, you’ll get a reply immediately. Other customers might take a week to respond — with a single, unfathomable poop emoji.
Our team was formed when we began partnering with messaging platforms in 2015. We were an experienced group from Zendesk Chat, with a goal to bring businesses and customers closer through messaging. Over the last two years we’ve learned numerous lessons, and we’re excited to share some of our key takeaways from the experience.
1. An ongoing relationship needs an ongoing conversation
Zendesk Chat was already an established live chat product, with hundreds of thousands of active users. It was clear that live chat and social messaging had similarities — both involved short, informal, back-and-forth conversations with rich media, such as attachments and emojis.
Thus, we began exploring social messaging in Zendesk Chat.
Once we started channeling messaging conversations into Zendesk Chat, we quickly realised the distinct differences between messaging and live chat.
Live chat was built on the premise of the conversation being live, where agents and customers were both present. Live chat on a website had a clear beginning and end, and the conversation could be captured in one neat session. The Chat dashboard was designed for an agent to have only a handful of live chats at a time, and it made sense for agents to close chats that weren’t receiving a response from the customer. Additionally, chats all came from a single channel source — your chat widget.
In contrast, messaging doesn’t have the paradigm of a ‘live’ conversation — it is asynchronous. Unlike live chat, where a customer starts and ends the chat over one session on a website, messaging takes place over an app. A customer can start a conversation and continue it at their convenience, at a later time.
Agents could not predict when customers would reply — they could take hours or days. Closing ‘quiet’ conversations meant that you might end up with 10 different conversations with the same customer, each containing one short reply, across the span of 3 days, for a single issue. We found that agents ended up confused with the lack of context, and unsure how to handle asynchronous solutions in the real-time world of live chat.
For instance, in one of our early customer interviews with Zulily, they wanted to return to the chats they had worked on previously to better follow-up with customers, but the Chat dashboard wasn’t optimised for this type of interaction.
We learned that messaging was more suited to the framework of a never-ending conversation thread. Through an iterative approach, we took the leap and refocused our efforts on a specialised dashboard — Zendesk Message.
Armed with research, interviews and paper prototypes, we began experimenting with an entirely new dashboard design, and a new workflow — one that was optimised for asynchronous and multi-channel messaging.
Our new design allowed agents to retain conversations that required more follow-up. By keeping multiple conversations with the same customer in a continuous thread, agents were able to understand the context of the conversation faster and no longer needed to ask customers to repeat information.
We learnt that businesses weren’t interested in racing through as many conversations as possible in a single shift, but rather they wanted their agents to build sustained relationships with their customers.
Ultimately, building a separate dashboard for Message allowed us to experiment and iterate quickly. We could then use the new learnings to improve the core chat experience.
2. Efficiency is great, accuracy is better
Bot integrations with Zendesk Message are able to trigger shopping receipts, shipping details, and many more types of automated chats. Naturally, grateful customers reply with a 👍emoji or a quick ‘Thank you’.
This meant agents had to deal with hundreds or even thousands of such replies. We observed that many agents were overwhelmed by the sheer number of bot-triggered active conversations. So, this was the question we posed to ourselves:
“What if assessing, triaging and resolving thousands of conversations quickly was the number one priority of Message agents?”
This was our hypothesis: Once an agent reads the last few messages from a customer, they can determine if a conversation should be archived or not.
We began testing this hypothesis with sketches and prototypes, and one experiment that stood out was a design we called the Triage Interface.
When our customers saw the prototype, they initially loved the speed and ease of working through so many conversations at once. But as we delved deeper into their feedback, we discovered that this workflow introduced new problems.
For instance, a customer might enquire about a shipping error from a previous order, and subsequently add ‘Thank you’ to a new receipt received via a bot. At a glance in the Triage Interface, it would appear as though the conversation could be closed, even though the customer still has an outstanding question.
This potential mistake in judgement would certainly be detrimental to the business’ relationship with their customer. Any risk of inaccurately resolving a conversation and unintentionally leaving a customer in the lurch was simply not a risk worth taking, no matter how small.
Thus, applying what we learned, we chose to go with a dashboard design that showcased the entire chat history instead. We also included custom filters that can be applied to each agent’s chat list. This would enable them to segment their active chats, discern which chats require attention, and get the full context of the conversation before resolving the chat.
As designers, we know that use-cases are infinite and we can never predict them all. Therefore, we found that we should strive to empower agents with the most accurate context and information to prepare them for any scenario. A fast workflow is great, but maintaining a well-informed relationship is paramount.
3. Automation multiplies the equation
Today, automated interactions are more prevalent than ever before. Traditional customer service has drastically shifted. Bots, third party apps and other innovations enable customers to interact with businesses in new and unprecedented ways.
We already know businesses are facing unique challenges with automations in customer service:
- Automated customer interactions happen with your brand name and your logo, but you aren’t involved in them.
- You must sift through a lot of noise created by automations, such as bot messages almost doubling your number of conversations in a single night.
One feature we designed to enhance an automated workflow is the bot handover. We wanted to avoid any collisions between bot and agent — imagine the chaos if both were replying a user at the same time. Thus, we designed clear separations to indicate when the conversation was owned by a bot, and when it was owned by an agent.
We’re still in the early days of conversational bots, and there are still many unanswered questions. What’s the best way to handle an automated conversation with a customer that has turned sour? How will businesses know and intercept it? What’s the optimal way to flag these conversations and let humans take over? Are we able to design systems that help agents cut through the noise and find the right customers who need personal help?
As the demand for automation increases, design must be flexible and robust enough to support and equip agents in this emerging space.
Building relationships that last
Designing Zendesk Message has been an incredible and exciting challenge. To me, it’s been a step forward into not only the future of messaging, but the future of customer service communication as a whole.
Currently, our team is still pushing forward and exploring new ideas. We’re thinking about how to organise conversations into custom, individualised lists for agents. We’re looking at how to encourage personal, human ownership over chats, even when you’re part of a large team, or when bots are leading the conversation. We want to make agent-to-automation interactions seamless and fast, without losing that precious human touch.
Trends inevitably change or fade. New technologies emerge. Therefore, we must keep asking the right questions, and focus on the bigger design vision of building relationships that last.
It’s a privilege that I was able to work with such a talented and humble team in building Zendesk Message, as we constantly learnt, fearlessly experimented and explored beyond the familiar. We know we’re only at the tip of the iceberg and we’re excited to improve and evolve our product in the months to come.
* A very special thanks to my co-writer, Jill Quek, for playing a huge role in putting this article together.