Javice’s Wharton Pal Created Fake Data for JPMorgan Deal

(Bloomberg) — A data scientist Charlie Javice met while studying at the University of Pennsylvania’s Wharton School testified that she asked him to create data showing her company had more than 4 million users a month before its acquisition by JPMorgan Chase & Co.

Adam Kapelner took the stand Tuesday at Javice’s fraud trial in Manhattan federal court. He is one of the main prosecution witnesses against Javice, 32, who is charged with greatly inflating the customer numbers for her student-finance startup, Frank, to get JPMorgan to pay $175 million for it in September 2021.

Kapelner said that Javice never told him why she wanted the “synthetic data” but stressed that she needed it fast and was willing to pay a premium for it.

“I’m in an urgent pinch and wondering if u still do consulting work and have some time,” she wrote Kapelner in an August 2, 2021, text message that was shown at trial. When he said his rate was $300 an hour, she responded by offering him $600.

The government alleged in its 2023 indictment that Javice hired a data scientist to create fake user information. Kapelner, a Stanford University graduate who was earning a statistics Ph.D. at Wharton when he met undergraduate Javice, is currently a math professor at Queens College. According to his website, he built models for hedge fund group Coatue Management in the summer of 2019.

The jury heard last week that, earlier on the same day Javice texted Kapelner, Frank’s chief engineer refused to create fake customer data in a Zoom call with her and Frank’s former chief growth officer, Olivier Amar, who is on trial with her. Patrick Vovor testified that he thought doing so might be illegal.

Both Javice and Amar have pleaded not guilty, contending that they didn’t intend to defraud JPMorgan. They claim the bank was not focused on the number of users and rushed its due diligence because it was worried a rival would buy Frank. 

Kapelner said he agreed to help Javice and spoke to her on the phone the day after her email. She sent him a computer file that showed Frank had fewer than 300,000 users. Over the next couple of days, Kapelner testified that he worked from specifications provided by Javice that called for 4,265,085 lines of data, each representing a fabricated user. 

Sometime after their first call, “I asked the purpose of the project, and she said she couldn’t talk about it,” Kapelner said.

In his testimony, Kapelner walked the judge and jury through his process of using Javice’s data as “the seed” to simulate “lookalike data” showing it had more than 10 times as many users than it actually did. He said the names were recombined from first and last names in the Frank database, but with no full names that were identical. He said the set was even programmed to approximate the percentage of first names in the original, telling jurors that the larger set would have the same proportion of people named Michael.

He was asked about a “Katherine Gordy,” who had an address, phone number, email and other personal and financial information.

“Ms. Gordy does not exist,” Kapelner told jurors.

On cross-examination by one of Javice’s lawyer Wednesday, Kapelner said “some basic diligence on the rows,” such as calling a few of the phone numbers, would have revealed that they were not real people.

Kapelner said he thought that whoever was using the data knew it was synthetic, noting that such sets can be used to present a picture of a company’s customer demographics without exposing any personal information.

He said he spent around 22 hours on the project, including an all-nighter, and sent Javice an invoice for $13,300, detailing the categories of tasks completed. She instructed him to replace it with a new bill with just a single line stating the payment was for “data analysis,” bumping up the total to $18,000, he testified.

Kapelner also testified that, after the deal closed, Javice, who was then working at JPMorgan, hired him to intregrate customer data she’d procured from a marketing firm into Frank’s database. He said he was paid $20,000 for the project. Prosecutors claim Javice was trying to cover her tracks with the new data. 

The case is US v Javice, 23-cr-00251, US District Court, Southern District of New York (Manhattan).

(Updates with detail of cross-examination.)

More stories like this are available on bloomberg.com