By giving away free credit scores to more than 60 million people, Credit Karma upended the paid credit report market. And by offering free tax returns this year — at the same time H&R Block and IBM Watson were bragging about AI-powered tax filings, the company stormed the $8.9 billion tax preparation industry. Bold disruptions like this matter, but Credit Karma’s effort to quietly assemble the massive trove of information underpinning each of these moves may be even bolder.
Credit Karma works like this: members supply personal information (name, address, phone, social security number) to receive credit scores and if desired, to be matched with the best credit card, personal loan, and auto loan offers, and to file their federal income taxes — all for free. The company gets paid a commission by the credit card and loan issuers, and vows it will never charge consumers. “We will always be free, that’s our core promise,” said chief technical officer Ryan Graciano, in a conversation with VentureBeat’s Blaise Zerega at Collision 2017. (See video above.)
The resulting data itself is invaluable — imagine what insights into one-fifth of America’s household debt might yield, but the tools and models Graciano and his team use to extract these insights are its magic glue. According to Graciano, they’re able to serve consumers up to the minute recommendations for their financial decisions within 20 milliseconds. And to do this, they rely not on general artificial intelligence (AI), but on artificial narrow intelligence ANI (ANI).
Now that Credit Karma has both tax and credit data, it can add more value to its services. For instance, according to Graciano, “We can do things like notice you have a mortgage and point out that your forgot to deduct it on your tax returns.”
“AI is definitely over-hyped. The promise of AI is generalizable intelligence, the ability to not only learn but to reason, to pattern match, to interact, and I think what we have today is best described as artificial narrow intelligence. ANI is actually amazing and deserving of the hype,” Graciano said. ANI is “very good at one very specific thing in a probabilistic fashion.” This means things like identifying a cat, instances where there is a big set of data coming in, and one specific output coming out. “ANI is transformational to do your business if you use it correctly.” Gracianio said. “For us, I think it’s actually a key part of our secret sauce.”
Graciano explains the process as taking say 200 million data points, crunching them into 2,000 factors per member, and then putting all of it together into a binary “approved” or “not approved” outcome. “We have to use data in production, for ad hoc analysis, for data science and modeling, and for online prediction, across all of our products and platform,” Graciano said. He adds that a single data set is 210 TB, and that their data is growing at more than 1 TB per day. (The company is adding 1 to 2 million members each month.) Further, Graciano and his team create 1,000 new models each month, which consider 120 billion observations, to predict everything from loan approval odds to the likelihood of interaction by members with offers to the value of particular services for members.
The 10-year old company claims to be profitable and reportedly generated $350 million in revenue in 2016, and it may be worth as much as $3.5 billion. An obvious next step for Credit Karma would be home mortgages and international expansion. The company recently entered the Canadian market. Another opportunity could be to develop a credit rating service like TransUnion or Equifax. Graciano discounts this and says no. “Being a credit bureau comes with its own challenges,” Graciano explained. “There’s a lot of regulatory stuff to worry about there.”
Graciano believes that his company’s competitive advantage isn’t big data and ANI per se, but the consumer trust it helps earn. Companies that jump on the AI hype wagon are going to have a bumpy ride. “I think AI will have a tricky time,” Graciano said. “Businesses that don’t master the ANI space are really going to struggle. If you work in volume, with the consumer especially, then you have to master ANI. You have to get really good at predicting what’s best for your consumer and what maximizes your business outcome.”