Thinking of implementing conversational AI? Read this first.

Sandi Boga, M.Ed.
ATB alphaBeta
Published in
4 min readJun 27, 2018

--

If banking is a thing people do — not a place that they go — how can we ensure customers can bank where, when and how they want to without asking them to learn a new software interface?

I believe that an AI-powered virtual assistant will enable exactly that. We can embed financial advice and services where people are already spending time so they will be closer to their money than ever before. Having a relationship with their finances may just mean people make better financial decisions, save more money and achieve financial health. This ease of access may also increase financial confidence — the mystery of money is removed when you can see it, access it and control it anytime you want to.

Imagine talking about your money as easily as you talk about the weather. Maybe the conversation is with an AI to start, but if some of that ease is carried over into a conversation with your financial advisor, everyone will benefit.

In other words, I believe banking can change people’s lives for the better.

Proving the conversational banking hypothesis

At ATB, I have been leading a proof of concept that allows customers to conduct banking by voice, using any IoT device that uses the Google Assistant AI. We started with a demo, conducting internal team member interviews to understand if this tech could solve a particular problem. I then partnered with a vendor, completed the relevant security assessments in order to connect the system to real banking data, and recruited internal team members to pilot the service on the Google Home. We did diary studies, surveys and a few more interviews.

After an intense development cycle to iron out the kinks, we conducted a successful customer study and are working on the next phase of development.

A-ha moments:

  1. Voice interfaces are different than text interfaces. Messier, less predictable. Choose an omni-channel vendor and test multiple channel types before a large-scale implementation. Test first with your own team members, including a diverse group of users (especially non-native English speakers). This not only gives you a chance to work out the bugs, but it gets people excited about the project and invites them on the journey with you.
  2. Know that which channels your customers use will depend on where they are and what they feel comfortable with. This underscores the importance of providing channel options at launch if you expect to see fast adoption and growth.
  3. Discover which features your customers deem mandatory, which they would find delightful, and which they don’t care about. Plan around their wants, not yours.
  4. Go beyond the transactional and create a personality for your AI. People want to connect. Help them do it by using humor and adding answers to questions like “what’s your favorite color?” (why, ATB blue of course!)
  5. Keep utterances short. People have difficulties retaining long strings of auditory information.
  6. What about security? People expect voice authentication, but home speaker devices are not yet there. Voice-based pin can work if you are willing to limit functionality and spending to mitigate risks. For example: don’t allow bill or e-transfer payees to be added in a voice channel; limit the amount that can be transferred, introduce two-step authentication for anything riskier. Partner with your your Risk and Fraud teams on this one (well, on everything really).
  7. People want to control what type of notifications they receive. Don’t assume they want spending/saving advice alerts of all kinds.
  8. Educate your customers on best practices to help them achieve success. For example: how to create good account nicknames, categorize spending, and get help if they need it.

Get your own house in order

Vendors aside, you need to seriously evaluate your own system’s capabilities and limitations, especially in regards to data structure, automation, available APIs, and the timeout window of the channel you are using (which may be shorter than what your systems needs to call and retrieve data). Some of the key dependencies I have identified include:

The bottom line is: start small, test, learn and then keep iterating until you know the experience will blow your customers away… or, better yet, will steal customers away from your competition.

And about competition: it may seem that some of them are too far ahead to catch up with, but look beyond the press release and the splash page. What’s truly on offer here? Probably just a glorified google search showing branch locations and product information. Now look at your customers. What do your customers want?

Forget the competition, and focus on activities that truly add value to your customer’s lives and makes banking work for them. I guarantee you’ll be ahead of the curve.

To learn more about our work around intelligent machines, visit www.atbalphabeta.com. Follow Sandi on LinkedIn or on Twitter @sandiboga.

--

--

Sandi Boga, M.Ed.
ATB alphaBeta

Passionate about defining, exploring and solving complex problems related to the future of money and banking. Tweet @sandiboga or www.linkedin.com/in/sandiboga