Editor’s Intro: Various forms of AI are being used increasingly develop insights, improve business processes and drive overall performance. Dana gives us a good example of how an AI-powered chatbot dramatically helped an online dating service. I’m particularly intrigued by the findings that customers liked interacting with the chatbots more than with humans, while at the same time, employees felt that their time was being used more valuably.
The online dating world has a reputation for being both superficial and disappointing, yet it’s often the only way for singles to meet potential partners. The three founding sisters of the online dating app Coffee Meets Bagel (CMB) sensed this frustration and knew they could do better, and developed their app to create fewer but more authentic connections, leading to what they envisioned as real, lasting relationships.
The concept was a huge success for them, earning $16.7 million in total funding and expanding internationally. However, as the app’s popularity grew to millions of users, so did the number of questions that came flooding into customer support trying to understand this new model: What is a bagel? How do I use beans? How can I edit my profile?
The support team of four was spending eight hours a day—and still never getting to the bottom of their support inbox. And that’s when everything was running smoothly. “We were just getting a mountain of emails constantly,” says Melissa Rosen, the current Head of Customer Experience (CX) at CMB. Taking a closer look, the CX team realized that 90% or more of their inquiries were from confused customers, asking questions that were already answered on their FAQ. For whatever reason, customers weren’t going to their FAQ or weren’t understanding it. Having to answer repetitive questions meant that the majority of emails sent by her team were saved replies, explaining what the app does and how it works. It was draining their resources and detracting their focus from improving in other areas of the customer experience.
As the app worked to solve the problems of the online dating world, the CX team needed to equip customers with more knowledge and tools to better optimize the user experience. They decided to integrate a chatbot powered by Ada, a Toronto-based provider of AI-powered customer support, to help filter the flow of inquiries.
At first, the CX team was skeptical. Melissa recalls, “We have a lot of jargon in the app. People are called bagels, there’s an in-house currency called beans, which get called points or all types of different things. I thought, ‘There’s no way a bot is going to understand all the vague or incorrect terms our customers use for the same thing. You need a human to translate what people are saying.’”
Beyond whether a bot could understand the customer, Melissa worried that it would sound too impersonal or robotic. But as the team started building their bot, they became drawn to the new system. Through the process of inputting saved replies into the Ada dashboard, the team was able to edit and organize their standard responses, streamlining how customers received information. “Slowly but surely, I started to see,” says Melissa. “Plus, we had control over the training, so I could see if a customer was saying, ‘I didn’t get my points,’ I knew they were talking about beans and could train it appropriately.” Good word embedding is the key to AI that works, explained Gordon Gibson, Ada’s machine learning lead at a Toronto Women’s Data Group Meetup.
Ada support launched the CMB bot in less than a month, initially to their Android customers and then several months later to their iOS clientele in a second roll-out. “We went from eight hours in the inbox to three,” says Melissa, who was pleasantly surprised at how well customers responded to the bot. The chatbot now answers general questions, captures customer information and performs smooth hand-offs to human agents. CMB’s ticket volume has been cut in half, the average resolution time reduced by four hours, all while increasing overall satisfaction rates during a time of company-wide growth. The bot has become a critical filter for the CX team, since only 13% of customers still need human attention after chatting with it. The majority of emails now in their inbox are about issues that need more complex human attention such as bug reports or glitches that require manual changes.
And, not only does the bot understand what customers are asking, it is able to reply in a personalized way that’s on-brand for CMB. “When we first started inputting answers, I realized this could be our brand’s voice. This isn’t a robot. This is us talking,” Melissa says. While customers seem to appreciate responses that seem human, Melissa acknowledges the post-launch rise in conversations, revealing that many are more comfortable talking to a bot than a human—since dating can be a highly personal and sensitive matter.
“We’ve seen so many more people talk to Ada than have ever spoken to us,” she says, describing how more customers are engaging with their brand through the bot, with fewer escalated inquiries that need human intervention. “To have a bot that can mimic the immediate conversation you can have with someone, without the pressure of actually talking to another human, it’s providing a much more positive experience for our customers.” CMB saw this increase in positive customer engagement without additional costs to the company or work for their CX team.
The speed, accuracy, personalization and privacy offered by the CMB chatbot has not only improved users’ customer experience, it has also allowed Melissa’s CX team to be more proactive and strategic about their role in the company. Having Ada be the first line of defense and answer more general questions has allowed the team more time to focus on deeper tasks like channeling the customer insight they’re collecting rather than just responding to queries.
CMB’s CX team are able to collect data trends through the Ada dashboard and harness Week on Week (WoW) insights, analyzing metrics like handoff rate (percentage of conversations that generate an e-mail ticket to the support team), containment rate (percentage of conversations that do not generate an e-mail ticket to the support team), review rate, satisfaction rate and recognition rate, while collaborating with engineers and other departments to improve their product. Customer satisfaction and chatbot efficiency metrics are analyzed from CSAT surveys, Net Promoter Score (NPS), first response time, time to resolution, touches per ticket, support costs, usage, recognition and overall ticket deflection. Ada allows CMB to get more volume behind these customer requests since more people are engaging with the bot resulting in improved customer experiences in the app.
The CMB team have also used the bot to proactively communicate changes in services and system interruptions right in the bot’s initial greeting, reducing ticket volumes during times that would otherwise see huge spikes. Importantly, the chatbot has allowed Melissa and her staff to do their jobs better, instead of doing their job for them, which is a fear she thinks a lot of reps have.
Coffee Meets Bagel has doubled in size since Melissa started out as a sales rep with them two and a half years ago, and is poised to grow even more. Previously their goal was just to get fewer people talking to support, whereas now their goal is to engage with customers in more positive ways with their chatbot.
Photos and some of the information referenced are from Perri Maxwell’s slides (Customer Success) at the Toronto Women’s Data Group Meetup in April.
The Marketing Research and Intelligence Association’s Emerging Leaders Blog publishes this and more new and innovative work here.