Tech Time: Chatbots in the Credit Union

One of the key takeaways for new board volunteers at credit unions is the emphasis on the human touch. Credit unions like to promote the superior quality of their member interactions versus those of commercial banks. While commercial banks usually have a Net Promoter Score in the negative or low positive number range (-20 to +30), credit unions have often had Net Promoter Scores in the high 60s with some achieving scores in the 90s. So why would a credit union use something as impersonal as a chatbot to communicate with members? It’s all in the use case.

Recently, I was applying for a new position on a company’s HR site, and instead of presenting a form to fill out, I was greeted by a chat bubble: “Hi, I understand you want to apply for the position of X, is that right?”

I was quite skeptical at first. How was this going to work? “Yes, I do,” I replied. And the conversation proceeded to go something like this:

Chatbot: Tell me a bit about yourself. What’s your name?
Me: Brian Garr.
CB: Thanks, Brian. Tell me, what makes you special for this position?
Me: Well, I have 25 years of experience building conversational bots.
CB: That’s great. Who is your current or last employer, and what did you do for them?

So it went on for about 15 minutes, and then I was done. It was so much less painful than copying and pasting parts of my CV into boxes on a web page. It also made me feel good about the prospective employer who took the time to create a much less sterile environment for applying for a job. In a form environment, the applicant knows that, regardless of qualifications, if they fill something out wrong or use the wrong terminology, they will never make it past the algorithms used to turn a thousand applicants into 10 applicants.

What areas of member interaction might have a less then hospitable environment today that could benefit from the gentle hand of well-written bot? What about loan applications or opening a new account? Members are usually nervous about filling out a loan app. They often don’t want to do it on the phone or in person because it is a lot less embarrassing to be rejected by a faceless webpage than by a human. Let’s imagine the following interaction:

Chatbot: Hi. I understand you want to look at a loan. What’s it for?
Me: I want to buy a new car, and I want to have a loan approval before I go to the dealer.
CB: That’s a smart idea, and we can help you with that. How much will you spend on a car?
Me: I’m looking at electric cars in the $45,000 range.
CB: And how much cash do you want to put down?
Me: 10%
CB: OK, I can see based on your history with the credit union that this should not be a problem. If you would like, we can send you a PDF file that you can print and take with you to the dealership. I show your email address as brian@domain.com. Shall I send the commitment letter there?

Isn’t that a lot less stressful than filling out boxes for the member? And with chatbot logic, we can fork the chatbot’s prompts and responses based on the member’s answers, so the experience is different for everyone based on what they want to do. We can also send the information to the credit union’s underwriter and bring back the answer to the chatbot, which can then deliver the news to the member. Of course, if the answer is not positive, the chatbot can then hand off the call to a customer service rep, if that is what the member would like to do.

Technically, none of this is difficult given the existing capabilities of high-end conversational artificial intelligence frameworks such as IBM Watson Assistant, Google’s Dialogflow or AWS (Amazon Web Services) Alexa. It’s not the technology that will make or break a chatbot—it’s the team building the chatbot that is charged with the dialogue, persona and flow. It’s a conversation, not a form-filling dialogue. Ideally, well-written chatbots take pressure off customer service reps so the humans can work on higher-yield products such as home loans and commercial loans, while members benefit from a clean, non-invasive discussion that leads to a car loan.

“We already have one” is the answer I got from several credit union executives when asked about their interest in chatbots. Very basic bots have become available inside of some credit union software packages, but they may be causing more frustration than providing quick service. Here are some sample questions asked to typical credit union chatbots:

Me: What’s your best rate on a CD?
CB: It looks like you were looking for our loan rates. Here is what I found.
    Can the interest rate change?
    Home Loans & Mortgages
    Loans

Me: What’s the current rate on your money market account?
CB: <Identical response to my question about CD rates>

This is hardly AI. The bot in question is simply word-spotting; it saw the word “rate” and assumed I was talking about a loan. What about something as easy as the following:

Me: I’ve lost my credit card.
CB: It looks like you’re searching for information on a new card.

OK, so what would a real AI-driven chatbot do in this case?

Me: I’ve lost my credit card.
CB: That’s awful. Would you like for me to freeze that card while you look for it, or would you like to cancel it ASAP?
ME: Just freeze it for now, and I’ll see if I can find it.
CB: OK, we will freeze the card once we have validated your credentials and verified which of your cards we should freeze.

AI-based chatbots are not just menu-branching dialogues based on word-spotting. Real AI-based chatbots work on understanding the entire request and intent of the member and try to react in a way the member expects.

For credit unions to keep those high Net Promoter Scores, they need to be a bit wary of pre-packaged, free chatbots that are an afterthought to most credit union technology vendors. The skillsets required to build and deploy a successful, highly effective and member-pleasing chatbot often do not exist in large software shops. For now, those skills tend to exist primarily in small chatbot-focused service groups. If your credit union isn’t working with a specialized vendor, be careful to avoid member-facing technology that can cause more frustration than satisfaction!

Brian Garr is a past chair of the audit committee for a $1 billion credit union and AI business solutions innovator for Dayhuff Group. He has spent over 20 years in the artificial intelligence industry.