The Dangers of Mining Consumers' Transaction Data

Personal financial management technology is evolving to include cash flow estimates and deeper transaction insights to guide consumers during their shopping decisions. Banks and alternative financial services providers are offering these emerging services to help consumers become quickly informed about their financial picture before they buy something they may not need, and thus stay on track to meet financial goals. But skeptics point out flaws inherent in such financial modeling.

"The overwhelming thing to keep in mind is that [modeling] is a shot in the dark," says Cathy O'Neil, who earned a Ph.D. in mathematics and pens the blog Mathbabe, which explores quantitative issues. O'Neil is co-authoring a book called "Doing Data Science" with Rachel Schutt, senior statistician at Google Research and an adjunct assistant professor in Columbia's Statistics Department.

One issue of modeling is what is known as a feedback loop, which occurs when a model gives a person the illusion of control; thus potentially making the data a self-fulfilling prophecy. "It doesn't mean the models are right, but they become more right if they've engendered trust," says O'Neil, adding that whether that's good or bad is somewhat subjective.

Picture a person who has been dealing with his finances for 40 years and one day decides to sleuth out short-term cash flow concerns through a model. Perhaps the concerns highlighted by an algorithm never occurred to him previously, she explains. The possibility of a worst-case scenario might inspire him to buy insurance to avoid the risks, which then adds new concerns and increases the person's dependence on a model, O'Neil argues. In other words: Be wary of a model that also sells insurance as it has a vested interest to scare its customer, she says. This could lead to a consumer, or worse, regulator backlash against the financial services provider.

Another red flag could be raised by account aggregation services designed to help people manage their money. When consumers provide their transaction data to a third party, they might risk predatory marketing. Say someone has an addiction to gambling and offers their transaction data up to a company. A bank marketing partner might take advantage of that consumer's vulnerability with unscrupulous deals that reflect badly on the bank.

Her examples underscore how companies' reasons for offering spending insights will vary.

"Predictive spending patterns have different usage," says Mohamed Khalil, director of product, data and partnerships at Movenbank, a financial services startup that has yet to launch. "It might not be financial wellness as much as to make sure a [customer] has enough funds to pay on time. Everyone has their own reason to do it. ...It's an emerging area with a lot of experimenting."

At Movenbank, "We want to be sophisticated like a Netflix," says Khalil. "We don't want to be Big Brother but want to provide context. ...Why use a product that gives no insight?"

This idea, already used by online retailers, is in its early stages for financial services players looking to provide trending information about consumers' total spend, says Khalil. To that end, Movenbank plans to initially offer experiences that get people to pause while spending. How? By showing them data nuggets at the point of sale like how their general spending on Tuesday compares to previous Tuesdays and how much more they can safely spend, for example.

"We're focused on financial wellness," Khalil says. The market is ripe for such innovation, according to Khalil, for these reasons: better aggregation capabilities, maturing mobile technology, transactions being posted more quickly, and recession-weary consumers and unemployed recent college graduates craving better spend management guidance.