EXCLUSIVE— How do credit recommendations really work?

As services such as Credit Karma, one of the larger credit monitors, increasingly turn to artificial intelligence and machine learning to manage customer recommendations, it’s worth considering what data those AI engines are using to pitch loans, credit cards, or other services to their customers (especially as consumers are growing more concerned with their data usage when it comes to companies like Google and Facebook).

Take a company like LendingTree, which just launched its own credit monitoring service, as Bank Innovation reported yesterday: the data used for this tool on its My LendingTree platform comes straight from a user’s TransUnion credit file, Justin White, vice president of product marketing for LendingTree, told Bank Innovation.

This data is then used to make “personalized recommendations in a more timely fashion,” White said, adding that LendingTree does not use additional data from third parties for this purpose. “It’s about the right product, that’s right for their financial goals.”

The company considers goals set by the user — like saving on student loan payments, or saving for a home, for instance — before pushing recommendations, White said.

These personalized recommendations are the result of running the data through “a machine-learning engine,” developed internally by LendingTree, White said, which compares a user’s data to others in the LendingTree marketplaces (there are about seven million users on the MyLendingTree platform at present) that “we believe are similar” to that customer, White said.

“We are not comparing income, or geolocation,” he noted, when asked which data factors the company is using for personalized recommendations, which typically steer users to loans, credit cards, or additional financial products, ostensibly based on that user’s individual financial position.

LendingTree’s recommendations completely rely on what’s in that user’s TransUnion file, White said: which includes, as all credit reports do, information on that user’s address and employment, although that information is not used to calculate one’s credit score. Rather, credit limits, loan balances, and payment histories on bank accounts are all typical factors involved in one’s credit score.

Essentially, this puts LendingTree users at the whim of the company’s machine learning algorithm, something younger financial customers are growing a bit wary of in finance — like roboadvisors, which have lost a bit of their shine, as Wired reported this week — for their financial health.

This comes at a time when user data is under some scrutiny due to technology companies like Facebook, whose ability to safeguard (and right to use) user data has been called into question this week.

White declined to state what specific factors LendingTree considers in its machine learning model, stating in a follow up email that it is “proprietary information which our competitors would be very interested to know. I can say it is the information in the TransUnion credit file along with our first party data, including user reported goals.”

Calling all startups: the deadline for admission to our sister accelerator, INV Fintech, is approaching. Apply here.