Featured article

How AI Can Help Your Support Teams Be More Empathetic

Last week, our CEO and Head of Product — Etie Hertz and Tadzio Smith — had the opportunity to participate in a Zendesk Startup Central event where they discussed how AI can help support teams be more empathetic.

The session also covered topics including:

Why empathetic support has become increasingly important for retaining and satisfying customers over the last year;

Common mistakes agents make when communicating with escalated customers; and,

Communication techniques to speak more effectively.

Read on below for the complete transcript of the event.

Q: We know this past Holiday season was more challenging than most with shipping delays and supply chain issues on top of everything else. Can you elaborate on why you believe empathetic support has become increasingly important for retaining & satisfying customers over the last year?

Etie: The past year has been trying on so many dimensions. Between the global pandemic, presidential election, economic turmoil, and protests — stress has been high. We’ve all been dealing with a lot of uncertainty and anxiety about the future and this impacts consumers, brands, and frontline agents. Everyone is worn out and patience is running thin.

At the same time, companies have been dealing with an influx of support requests and have been turning to self-service and automation to more effectively manage their queues. As a result, we’ve seen the conversations that reach human customer service agents are becoming both more complicated and more emotional.

Put yourself in the shoes of a customer: first, they may try to self-serve on your website. If they can’t find an answer, they may reach out to a chatbot if you have one. If they still can’t find an answer, then at this point, they’re probably pretty fired up… so by the time they reach a human agent for help, the agent is managing a more emotional customer who also likely has a complex issue.

With all that is happening around the world, customers are demanding more empathetic service interactions. We believe these emotional and complex conversations actually provide opportunities — the companies that show up for their customers in these times will build loyalty long term.

Q: You got your start from Crisis Text Line, so you’ve had a lot of experience talking to people through SMS when they’re upset. Can you share with us the mistakes you’ve seen customer care teams make when putting together strategies on how to talk to escalated customers?

Tadzio: We’re fortunate that our data team has been able to learn lessons from the largest mental-health data corpus in history. Sometimes the learnings are intuitive and sometimes they’re surprising.

One of the mistakes we often see customer care teams make is instructing agents not to acknowledge the customer’s emotions because there is a belief that if you focus on what the customer is feeling, it will escalate the customer further. In this line of thinking, it’s more important to get right to solving the problem that the customer has. And I can understand why people think this may be the best tactic: it seems intuitive that if someone is upset about something which has gone wrong, just focus on solving the problem. However our data shows nothing could be further from the truth — it’s much more effective to start the conversation by acknowledging what the customer is feeling before working to solve their problem. In fact, we’ve seen an 18% increase in survey results when this approach is taken.

Q: Have you found that there are certain types of issues which agents struggle with the most when trying to solve?

Etie: We’ve seen that regardless of vertical, agents get tripped up a lot when they basically have to tell the customer ‘no’. For example, they can’t offer a refund because it was a ‘final sale’ or they can’t share the phone number of their delivery person for privacy reasons. This type of interaction often doesn’t end well.

Q: What are some strategies to deal with these types of interactions?

Tadzio: Telling a customer ‘no’ is never fun or easy. We’ve built a three-step formula into our software to help agents ‘say no’ more effectively, and we’ve seen it work really well:

Acknowledge your customer’s feelings. As mentioned above, this first step is critical and it’s often missed.

Apologize and say no. Apologies work best when they’re followed by an ‘action statement’, or telling a customer what you will do to help them.

Offer an Alternative. Focus on what you can offer, not what you can’t. This conveys that your customer has options, which opens, rather than closes, a conversation.

Check out a blog post we recently wrote which gives advice on how to tell a customer that you don’t offer phone support after they ask for it. We’ve seen this come up a lot as an issue as companies transition more to digital channels for support.

Q: You’ve touched on the fact that as the easy customer interactions get automated, agents are left with the more challenging interactions. And this added stress on agents can lead to higher turnover which is costly for companies. What tools have you seen to help companies address this?

Tadzio: There are a lot of great tools available for deflection (like the Zendesk Answer Bot) and there are awesome startups helping companies automate more of their customer interactions (many are here in the channel!).

In terms of helping agent morale, we’ve seen companies use tools which reward agents more for their good work. For example WorkProud or Espresa.

To lessen the cognitive burden placed on agents, we work on reducing the decisions they have to make in each interaction by giving them step-by-step clickable text suggestions which guide them through challenging conversations.

When building features to help agent job satisfaction, we follow The Progress Principle: of all the things that can boost emotions, motivation, and perceptions during a workday, the single most important is making progress in meaningful work. So even recognizing small wins can make all the difference in how agents feel and perform.

One way we employ this principle is by gamifying the customer conversation by showing agents a real-time CSAT predictor and emoji of how the customer is feeling. As agents chat with customers, their goal is to improve the predicted CSAT score. When they do, they receive ‘stars’ and ‘trophies.’ We find this continual live feedback helps agents feel more appreciated and see the positive impact they’re having. This is one of the ways AI can help the satisfaction of both agents and customers.

Q: What makes you most excited about the future of conversational AI?

Etie: The growth of the field! Conversational AI is advancing because the field of Natural Language Processing is rapidly evolving with more advanced language models being released. Current conversational AI has the ability to run multiple models in parallel and incorporate more of the context of the conversation.

We’re especially excited about augmented intelligence because the shift towards more computer-human interaction and automation has produced more conversation data to analyze and more opportunity to hone conversational AI techniques. The more data… the more opportunity.

Q: Any particular AI tools that you’re using which other support teams could utilize?

Tadzio: There are so many great new tools and libraries being developed for teams building their own AI. SpaCy is a transformative library for NLP with version 3 being released soon. HuggingFace’s transformers library has been a game changer — open sourcing 100s of cutting edge language models that can be tailored for conversational AI. The technology as well as the surrounding community have led to explosive growth in the field and provides exciting opportunities in the world of CX.

Special shout out and kudos to Greg Geibel at Zendesk Startup Central for giving us the opportunity to participate in this Q&A!

Additional questions on your mind? We’d love to hear about them. Join the conversation on Startup Central or shoot ’em our way at hello@loris.ai and we’ll tackle them in an upcoming Q+A.