Killing Your Babies… and other Voice Chatbot Thoughts from Credit Card Helper Co-founder Owen Brown
Owen Brown is the CTO of Starbutter AI Helpers, a company that makes chat and voice agents to help people pick financial products like credit cards and mortgages. He previously worked at two adtech companies, Convertro (sold to AOL), and OpenMail (acquired InfoSpace, implementing Bandit Algorithms, and also as a private banker at UBS. Owen is a machine learning and AWS expert.
Austin Macdonald — How did you get involved in the AI agent and chatbot industry?
Owen Brown — Our team met about five years ago at UCLA. We’d trade notes on finance and tech, and by 2016 found ourselves constantly astonished by improvements in voice recognition technology. Between that and the democratization of machine learning tools, it became pretty clear that something big was going to happen with voice and AI, so we made a commitment to the space.
We immersed ourselves in the community, organized a Google Actions meetup in San Francisco, and reached out to experts in the field. We chatted up all as many great minds in the space as possible — Arun Rao, our CEO, just published an interview with chatbot legend Bruce Wilcox — to find out what design elements make or break a bot. On the tech side, the NLP and machine learning tools are evolving so rapidly, that the only way to learn is to roll up your sleeves and start building, so we did.
Helper Factory’s chatbots help people pick financial products, like credit cards and mortgages. We have tens of thousands of users and they’re a marvel. People love them and we’ve validated one study’s result that half of all Americans are willing to use a chatbot to apply for a credit card.
AM — Are bots just a fad? Have any chatbots made money or added value to people’s lives?
OB — I think the best analogy to chatbots is the laser. When the laser was first created, no one knew what to do with it. Initially, everyone thought it would be used as a weapon, but it’s actually a horrible weapon because it immediately cauterizes any wound it creates. It took a little over 20 years for first successful consumer application of lasers — the compact disc — to reach markets. But initially, the technology was the solution in search of a problem. I think chatbots are in a similar spot, though people are coming to solutions much faster. For example, SnapTravel is using interactive chatbots to make booking hotel rooms really easy.
AM — What are your thoughts on chatbot design and user experience?
OB — I think form should follow function. Decide what you want your chatbot to do, and then trust your user to be smart but clueless, so give them clues along the way.
Use all of the UI elements available to you. People hate big blocks of text. People love lists and images, and they feel more engaged by visual elements.
Get in front of your users. Your chatbot needs to do one thing, really, really well. So you need to cue your users to what that one thing is, and polish the hell out of it. You need to be disciplined and kill your babies.
AM — Kill my babies!?!
OB — You learn pretty quickly that, if you don’t cue your user to specific chatbot functionality, they’ll never use it. So, removing a suggestion chip or prompt essentially means that a feature you spent weeks on will never be used. You need to understand that, and do it anyway, ruthlessly.
AM — What is the focus of your current product, Credit Card Helper?
OB — Our company uses extensive research and data to help users understand and select financial products. We’re getting data from tons of sources. For example, we’re using machine learning to read through hundreds of thousands of complaints to the Consumer Financial Protection Bureau in order to assess which companies are shady and which are reputable. Users then have a short conversation with the bot, and we select cards for the user based on their preferences.
AM — What’s unique about this chatbot?
OB — We gave our chatbot a catchy name and some attitude, so sometimes he’ll make a snarky comment about a credit card he doesn’t like. Caz is trying to emulate that one smart witty friend who is an expert on credit cards and plays the credit card game, so we wanted to give him some personality quirks.
AM — How does Caz the Credit Card Helper make me smarter and faster?
OB — Well, he’s not the mushroom in Mario Kart, so I’m not sure he makes you faster. We can ask you five or six questions, and then pick out a better credit card for you then what’s probably in your wallet now. I think most people feel smarter when they know they’re getting a good deal, you can find a better credit card within 5 minutes of talking to Caz. We’re still in pre-beta mode, so Caz is definitely on the come-up!
AM — Can Caz solve all my life’s problems? Issues with my parents or my boss too?
OB — Hmm, sounds like your therapist or Eliza would be better at helping you with those problems.
AM — What chatbot or voice chat platforms are worth looking at? Will Google, Amazon, or Apple win the platform war?
OB — We’re making a big bet on Google Actions. When you look at the number of phones that will come pre-installed with Google Assistant in 2017, it’s pretty clear that the install base for Google Actions will be larger than Alexa. Next we think Facebook Messenger is going to be big for chatbots — they’ve invested a lot in their platform.
I’m in love with Apple, but they’ve been mysteriously reluctant to allow Siri to integrate with iOS apps. Also, empirically, it’s pretty clear Google is winning the data war. Seriously, pull out your iPhone, install Google assistant, play with it for 5 minutes. It’s significantly better than Siri in every way.
AM — What were the biggest challenges of creating this product, both technically and in the playing field of other fintech bots? I’m in school, and I’d Iove to knock out a chatbot over a weekend.
OB — If you want to have fun, there are some great off-the-shelf products for building a low-end chatbot. You could make something amusing in a weekend. But, if you’re interested in doing anything commercial that has to work and reach millions, I think you’re better off rolling-your-own, with the help of an intent recognition system and AWS tools. It’s a big investment, but it’s worth it for the security and flexibility.
AM — Where do you see the AI/chatbot industry 3 years from now?
OB — I don’t think in terms of chatbots. I think in terms of services. A chatbot is just a means of accessing a service, through chat. If our chatbot can be charming, and make a person smile occasionally, great. But the point is solving user problems.
I do buy into the ambient intelligence hypothesis — that in three years, we’ll be talking to the objects around us about as frequently as we check our mobile phones today.
Also, for better or worse, I think the way it plays out it that we’ll have a close relationship with a single AI agent — probably Google in the United States — who acts as our chief of staff, filtering our communication with all other apps. So, as a chatbot designer, your success or failure is dependent on doing a few single tasks extremely well, so that the chief of staff respects you.
AM — Do you think a super intelligence chatbot that is as smart a human is possible? In the next 5 years, or 30 years?
OB — Smart is a pretty vague term. Watson crushed humans at Jeopardy in 2011. Google’s knowledge graph probably crushes Watson. So, depending on how you define it, the super intelligence actually surpassed us almost a decade ago. Kevin Kelly has pointed out that there are already a variety of artificial minds superior to human intelligence.
You could define intelligence differently — as decision-making that allows a being to survive and grow. With this definition, you could ascribe intelligence to corporate persons and industry groups, such as Big Tobacco, that contributes nothing to society but continue to survive and grow. Big Tobacco in particular uses misdirection, camouflage, lures, and other process we typically associate with biological entities, to thrive.
Would I say that the tobacco industry is alive? I’d say it’s a grey area — sort of like viruses. But is it more intelligent than any one person? Absolutely. If it wasn’t, we’d have legislated it out of existence within a generation of finding out that smoking causes cancer. Clearly it’s communicating with lawmakers, but through lobbyists instead of chat.
How to talk to Credit Card Helper:
Credit Card Helper is available on any new Android phone, all Alexa devices, any Facebook Messenger account, and on iPhones through the Google Assistant app. Over 2bn devices and accounts!
1) For Facebook Messenger, click here.
2) For any Android phone (with the latest updates to the OS), click on the home button to pull up Assistant and say “Talk to Credit Card Helper”. More details here on how to get the Assistant app for any iPhone.
3) For any Google Home or Amazon Alexa device, just say “Talk to Credit Card Helper.”
