This is a follow-up to my first post, where I discussed how I built and launched a banking chatbot. I’ve seen a lot of great discussion lately on sites like Medium and LinkedIn about applying conversational interfaces to various industries and workflows. I hope this article advances that discussion, especially in the field of retail banking.
While I can’t disclose the specific data and metrics of my pilot, I wanted to share my general learnings. Some may seem obvious, but it is better to confirm your assumptions than rely on them blindly. One distinguishing feature of my pilot was the target market: an economically and culturally diverse set of credit union members in Brooklyn. This group of users is probably very different from the tech-savvy users of most other chatbots.
Here are my takeaways:
1. There is a lot of curiosity
Despite probably not knowing too much about chatbots beforehand, people were genuinely interested in learning more about a new “virtual banking assistant” (perhaps I can thank Siri and Alexa for that). My initial outreach message ended with: “You can say ‘More’ for some examples of what you can say.” The volume of people who responded with “More” exceeded my exceptions (although it created another problem, see #2). This also gave me some hope that, at the very least, people were receptive to trying new channels of communication.
This insight is also reflected in some recent studies done on chatbot and AI usage. A recent Accenture survey of 33,000 consumers across 18 countries indicated that more than 70 percent would be willing to receive computer-generated banking advice. Additionally, Gartner estimates that by 2020, customers will manage 85 percent of their relationship with a business without interacting with a human. It is up to the developers to make sure that we build the best possible user experience to complement the traditional support channels.
2. Limited exploration
While the first insight positively surprised me, this next one brought me back to earth. Although many people understood the concept of a “banking assistant,” almost no one tried commands that weren’t suggested. Only a handful of people tried asking personal, open-ended questions. I don’t believe this is exclusive to just the Teller platform either. I have an Amazon Echo at home, and I pretty much use the same few commands over and over again. Occasionally, I try new things, but I don’t think anyone has mastered teaching people how to explore the breadth of possible commands.
This may not necessarily be a bad thing — after all, if people are getting value from the suggested commands, than a banking chatbot is still very useful. However, I believe that the real power of conversational interfaces will be when people feel free to communicate any broad need.
Rather than treating this as a technical problem, I believe the solution is 50 percent onboarding and 50 percent marketing. I already have a few ideas for how I might encourage new commands, such as marketing the “AI” component more heavily, sending weekly suggestions emails (like Amazon does for the Echo), or even removing the suggested examples altogether. Solving this issue will definitely require some experimentation.
3. Chatbots still need a human backup
I won’t spend too much time on this because it is likely obvious to anyone who has been trapped in an automated phone system. (Side note: It drives me crazy that sometimes I have to resort to repeatedly shouting “AGENT” at my phone. And sometimes, the smug voice on the other end says, “It looks like you want to speak to an agent, but answering just 15 more questions will help us better route your call.”)
I really believe that having empathy for users is crucial. That means routing the chat to a human when the conversation feels stuck. Moving forward, I want to build a seamless process for integrating with support or CRM software that a bank already uses. While I don’t want to create extra work for the bank, making sure users are helped quickly is crucial to long-term success.
4. Proactive reengagement
The biggest insight to come from my pilot was the need for reminders and reengagement. As with most product launches, the real challenge is user retention and growing periodic usage. This “launch curve” also applied to our pilot: Initial reception was strong but then slowly decreased. This is where smarter technology and marketing can help reverse the course. On the technology side, I am building a smarter engagement engine and focusing on the customer journey (more details for this will come in a future post.) On the marketing side, it is important to nudge people toward using the messaging platform for their simple, repetitive questions.
The one thing I want to be careful about is not overdoing it by spamming a user. If a service sends too many messages, it risks users not just abandoning the service but holding negative feelings toward it. This balance is where a smart algorithm, potentially using machine learning, has an opportunity to create a personalized experience for each user. Based on knowing a few things about each user (account age, usage activity, engagement, basic demographics), this algorithm would determine the right message and the right time to send it and would also learn based on the response/feedback from the user.
5. Expect some strange responses
I was chatting with another founder recently, and he was telling me that his online support staff sometimes gets inappropriate and moderately offensive communication. I said, “I can see that — after all people will say anything behind anonymity.” His response was, “I wish that was the case. Sadly, these are logged-in users writing in with their real names, phone numbers, addresses, etc. ”
My experience with Teller matched this founder’s comments. On the silly side, I got a few expected messages from people asking: “Are you human?” On the more nefarious side, I received a handful of vulgar, emoji-laden messages (some I didn’t even know existed, but I had to applaud their creativity). Perhaps I should add emoji-translation to the next version of Teller.
These are some of the broad insights I gained from releasing my chatbot. Overall, I’m extremely optimistic about the applications of chatbots in banking and their ability to provide a convenient and personalized banking experience. I’m continually trying to improve Teller, including building features like bank account integration, adding new messaging channels, and operationalizing deployments. I’ll continue to post about my experiences.