Banking startup LendUp shows why design is king as big data gets personal

So far, the rise of big data has largely been a passive affair. Many websites and companies focus on the big part — collecting as much as possible in order to determine what’s relevant and where it’s valuable. This probably works fine when they’re trying to uncover macro trends in customer behavior, latent causes for slowing business or even that proverbial needle-in-a-haystack insight. However, as consumers expect more-personalized experiences, companies might need to get smarter about what they collect, how they get it and how they use it to create a consumer experience.

Personalization, it seems, is really about gathering exactly the data that’s needed in order to perform a particular task. Think about how Amazon asks users whether purchases were for themselves or as gifts, or how streaming services like Netflix and Pandora ask users to rate content. Consider how Google Now asks very clearly whether users care about the new information it surfaces. That someone bought, watched or listened to something — even traveled somewhere — doesn’t mean they liked it or even are interested in it.

Arguably, the more important that information is to carry out the business, the more aggressive (or clever) companies should be in trying to get it. This is a topic numerous speakers will be addressing at our Structure Data conference in March, as they talk about building businesses and products that rely on data to improve, or actually provide, the consumer experience. The services above really give users the option to provide information, presumably because personalization isn’t that important to the business, or because their personalization algorithms don’t rely too heavily on that data.

One of many ways tries to get us users to rate content but doesn't force them to.

One of many ways tries to get us users to rate content but doesn’t force them to.

When the business relies on data …

For banking startup LendUp, however, really understanding its users makes all the difference in the world. The company is trying to be a low-friction source of relatively cheap loans for underbanked individuals and, CTO Jacob Rosenberg told me during a recent trip to the company’s San Francisco office, “We set it up for ourselves so we don’t win unless [our customers] win.”

Assuming the company walks its talk, Rosenberg isn’t kidding. In a nutshell, the company’s business model is based on offering quick loans with relatively low interest rates (compared with payday lenders). The more times someone borrows and pays back — and the more of LendUp credit-education courses they complete — the more money they can borrow for less interest. There are no late fees and, at a certain point, LendUp even reports positive information to FICO to boost customers’ credit scores. For the most part, everything is done online.

If a customer needs more time to pay back a loan, he or she can change the repayment date online. If they’re still late, LendUp will reach out and try to figure out a plan, but there are no harrassing phone calls and no accruing interest or late fees of any kind. According to Co-founder and CEO Sasha Orloff, that’s because it doesn’t help LendUp get paid back if its customers are now on the hook for more debt and possibly getting overdraft charges from their bank as they try to pay back LendUp.

“We don’t do any of that,” he said. “… If they don’t pay us back, we don’t make money.”

… you get the data

It’s a laudable (arguably humanitarian) approach to lending, but it puts LendUp between a rock and hard place from a data perspective. The company can’t possibly ask users for all the data it might want in order to process their applications and still keep the experience as painless it wants, but it also can’t rely on the relatively small number of data points that traditional banks use to assess credit risk. LendUp’s solution was pairing smart site design with smarter algorithms.


As soon as someone comes to its site, Rosenberg explained, the company is gathering data. Did you come from the site of a credit-building partner, or from a Google search for “fast money no credit check”? Did you immediately move the slider bars on the LendUp site to the maximum amount of money and maximum payback time, then hit “apply”? When it comes to the actual application, he said, LendUp asks for standard data from each applicant (including Social Security number so it can look at credit scores and other data), but it might also ask certain applicants to connect using Twitter and Facebook, if only to assure their email address is the same across accounts.

Obviously, the data LendUp generates about how people interact (by completing those credit-building lessons, for example) and repay once they’re in the system also helps the company determine future rates. The whole experience is based on Orloff’s experience at Grameen Bank (which focuses on lending to “the poorest of the poor” around the world) and Rosenberg’s experience as an architect at Yahoo and most recently Zynga, building gaming platforms that reward users, and generate more data, the more they engage with the system.

“We’re looking for data that has relevancy to repayment,” Orloff said, primarily around an applicant’s identity, ability to repay and willingness to repay.

Machine learning does the hard work

Most of the variables — thousands overall — are fairly insignificant on their own, but every little piece of data matters because the company’s goal is to build a case for approving applicants rather than to find a reason to decline them. Machine learning algorithms help LendUp fill in the gaps where certain variables might look bad, or where data is sparse for a particular applicant, by analyzing patterns across its user base.

Watch a 7-minute video, take a quiz, earn points.

Watch a 7-minute video, take a quiz, earn points.

LendUp’s models are nowhere near as complex as the models that some other lending startups claim to use, and that’s by design. For example, ZestFinance, a lending startup focused on licensing its underwriting model as opposed to issuing loans itself, boasts about its machine learning expertise and the 70,000 variables its models analyze to assess risk. Orloff said he hopes ZestFinance’s tech-focused approach to underwriting catches on — any progress in serving the underbanked is good — but focusing too much on the math might detract from LendUp’s user experience, around which the whole company really is premised.

Further, he added, LendUp follows state and federal banking laws (some short-term lenders are based on reservation land and operate under tribal law), which can make storing data for the sake of it kind of problematic. There are rules about what types of data financial institutions can collect and use to calculate the terms of loans, and Orloff said he doesn’t want to be left explaining tens of thousands of variables should a regulator come knocking.

Besides, LendUp should already be getting the data it needs because of how it has designed its lending experience to be easy, intuitive and optimized for engagement. When the business relies on finding the right borrowers, making the right recommendations or otherwise really knowing what customers need — and when there are plenty of other options to choose from — being smart about data collection seems like a very smart way of doing business.