I haven’t posted in awhile.
BUT.
I’ve been on pins and needles with this whole Frank-committed-fraud situation (the company, not the boy).
Now - before you come at me (Izzy why are you so obsessed with Charlie Javice), I need to level set the situation.
The year is 2019.
And Charlie starts Frank - the fintech company making it easier for students to apply for financial aid.
Okay.
FAFSA works for students with married parents and steady incomes.
But what about students with half siblings? Divorced parents? Who live in foster homes?
Charlie wants product market fit so she designs the solution around a group of high school students in the south of Bronx.
She asks herself - How do you build something that is multigenerational? That your grandma can complete and Gen Z loves.
There’s no holistic financial wellness platform for students (nothing like TurboTax that enables you to make the right decision around college).
And maybe that’s because colleges are large and in charge and government funded… IDK. But Charlie Javice DOES something about it.
According to The Bill and Melinda Gates Foundation, it takes an average of 13 hours to complete FAFSA. With Frank, it takes 4 minutes and 12 seconds.
Did you know that tuition is negotiable?
(Charlie went to Penn, and she decided to save her parents some money by graduating in three years. But that wasn’t the real cost saver. The real cost saver was when Charlie’s little brother also decided to go to Penn, and FAFSA made their family eligible to pay two tuitions for the price of one). I mean … WHO KNEW that’s how FAFSA works?
After all, there really isn’t any transparency behind the cost of college.
One in five Americans have student loan debt, and 78% of borrowers are behind on payments. So, yeah. It’s really hard to measure ROI if parents aren’t chipping in. Frank saves students an average of $7,000 on tuition by appealing financial aid.
Okay.
Frank is free for students. Schools pay for Frank because it helps connect students to empty seats. And Frank’s dashboard takes an evidence-based approach towards matching students with the “right” college (data points include desired major, tuition price, career pathways, etc…).
BTW. There are empty seats at the top 500 schools, and graduation rates hover around 50%. Creating a student dashboard with insights into which college they should choose and why is super smart.
Most Frank users come from unpaid channels. Instead of looking at content as an SEO Farm, Charlie has a bunch of landing pages with FAQs. (Like - how do I pay back my student loans 30 years later?)
So, yeah.
I totally get why JP Morgan bought them.
Build a trusted relationship with a demographic that is typically failed by financial institutions!!!
(AKA. Low to medium income households who are about to be college educated and therefore eligible for a line of credit or a loan. Which, BTW, is a huge TAM. 50% of Americans fall in this category).
Okay.
Let’s talk about where Frank f*cked up.
The year is 2021. And Charlie lies about her total user count. She makes a slide deck and says that Frank has 4.25 million active users.
In reality, she has 300,000.
I don’t understand how to reach scale at a company like hers.
But the person who grows a company from $0-5 million revenue likely isn’t the person who takes it from $5-10 million revenue.
And maybe JP Morgan could have helped. If she had just said - hey here are the skeletons in my business, let’s get to the bottom of due diligence and see if we can reach a deal, things could have turned out differently.
Instead, she does some bad, like really bad stuff.
And since JP Morgan now owns the severs which Charlie did that really bad stuff on, we have access to all of her juicy emails.
READ THEM NOW PLEASE!
As written by someone you should totally subscribe to, Matt Levine of Money Stuff:
Charlie sent Frank’s Director of Engineering an email with a link to an article entitled “Generating Tabular Synthetic Data Using GANs.” The article notes that “[t]he goal is to generate synthetic data that is similar to the actual data in terms of statistics and demographics.” The article suggests that “it[’]s fairly simpl[e] to use GANs to generate synthetic data where the actual data is sensitive in nature and can’t be shared publicly.”
Javice, Amar, and the Director of Engineering had a Zoom meeting during which Javice and Amar asked the Director of Engineering to help them create a synthetic list of customer data. She asked the Director of Engineering if he could help her take a known set of FAFSA application data and use it to artificially augment a much larger set of anonymous data that her systems had collected over time. The Director of Engineering questioned whether creating and using such a data set was legal, but Javice tried to assure the engineer by claiming that this was perfectly acceptable in an investment situation and she did not believe that anyone would end up in an “orange jumpsuit” over this project.
The Director of Engineering was not persuaded and told Javice and Amar that he would not perform the task and only would send them the file containing Frank’s actual users, which amounted to approximately 293,000 individuals at that time.
Aieee, this is not legal advice, but if your boss is prone to using the phrase “orange jumpsuit” you should quit! And probably go claim your whistle-blower reward while you’re at it. Anyway Javice then allegedly “sought help from external sources”:
Javice then turned to a data science professor at a New York City area college (the “Data Science Professor”) who advertised his “creative solutions” to data problems. Javice provided the Data Science Professor a list of 293,192 individuals who had started or submitted a FAFSA application through Frank. She then directed the Data Science Professor to use “synthetic data” techniques to create 4.265 million customer names, email addresses, birthdays, and other personal information based on the list Javice supplied.
The professor allegedly created the list, with lots of hilarious guidance along the way that JPMorgan now has:
Regarding creating physical addresses, the Data Science Professor wrote to Javice, “I can’t seem to find addresses in my raw files . . . . Should I attempt to fabricate them?” Javice responded “I just wouldn’t want the street to not exist in the state.” Later, the Data Science Professor determined that “‘real addresses’ may not be doable,” and Javice responded “If we can’t do real addresses what[’]s the best we can do for that?” …
Javice was particularly concerned with the email addresses, asking the Data Science Professor “will the fake emails look real with an eye check or better to use unique ID?” He responded “[t]hey will look fake,” at which point Javice agreed to use a “unique ID” instead.
Eventually the list was done and the professor sent Frank an invoice for $13,300, listing his hours devoted to tasks like “college major generation.” This was not the invoice Javice wanted, so:
In response to the initial invoice, Javice demanded that he remove all the details admitting to how they had created fake customers – and added a $4,700 bonus. In an email to the Data Science Professor at 12:39 p.m. on August 5, 2021, Javice wrote: “send the invoice back at $18k and just one line item for data analysis.” In total, Javice paid the Data Science Professor over $800 per hour for his work creating the Fake Customer List, which is 270% of his usual hourly rate.
The Data Science Professor provided Javice the revised invoice via email seven minutes later at 12:46 p.m., commenting “Wow. Thank you. Here is the new invoice.”
Yeah I mean a general lesson here is that you want to be in the sort of business that sends one-line invoices, not the sort of business that bills by the sixth of the hour. [2]
This basically worked: Frank sent its list of 4,265,085 customers to JPMorgan’s third-party due diligence consultant, [3] the consultant okayed it, JPMorgan was satisfied and they signed the merger agreement. But you can see how this would be a problem for Frank in the medium term: It had a list of 4 million customers with fake email addresses, and JPMorgan was basically buying it so it could send emails to that list. If the email addresses were all fake, JPMorgan would notice immediately. So Frank allegedly went out and bought an email list:
At the exact same time Javice was working with the Data Science Professor on the Fake Customer List, Amar separately reached out to ASL Marketing, Inc. (“ASL”), a marketing firm that purports to have “the most comprehensive, accurate and responsive data of high school students, college students and young adults available anywhere.” Amar caused Frank to purchase a list of 4.5 million students from ASL for a cost of $105,000 (the “ASL List”) on the same day Javice transferred the Fake Customer List to the third-party vendor for validation. ...
Amar and Javice would have had no reason to buy a list of 4.5 million college students if Frank actually had the number of customer accounts that Javice represented to JPMC and provided to Acxiom.
The ASL list had about 4.5 million names, but only about 2 million of them had email addresses, so Javice allegedly went to another company to try to buy the relevant email addresses:
Javice again engaged the Data Science Professor to work with another third-party vendor, Enformion, LLC (“Enformion”), to obtain additional email addresses to add-on to the individuals in the ASL List. …
After receiving the data, Enformion ran the ASL List through its databases in an attempt to find matching additional data, including email addresses. Enformion then provided those matches to the Data Science Professor.
Upon information and belief, Javice and the Data Science Professor’s first request to Enformion was to provide email addresses based on the existing full names and addresses, but that provided a limited number of matches. Upon information and belief, Javice decided to instruct Enformion to match against just last names and addresses, which could generate email addresses of siblings, parents, and other family members of the individual on the ASL List, even if the email affiliated with the exact individual on the ASL List could not be found.
Ultimately, Javice and the Data Science Professor obtained at least one email address for approximately 1.9 million of the 4.5 million individuals on the ASL List from Enformion.
In January 2022, after JPMC acquired Frank, JPMC began work on a marketing campaign to test the quality of Frank’s customer account list and the receptiveness of these customers to JPMC’s products and services. To facilitate this marketing campaign, JPMC asked Javice and [Frank Chief Growth Officer Olivier] Amar to provide Frank’s user list so that JPMC could send the marketing test by email. …
JPMC reached out via email to a random sample of the list Frank provided – approximately 400,000 purported customers of Frank – with offers to open Chase checking or savings accounts. Of those 400,000, only 103 even clicked through to Frank’s website…
OKAY.
(Me again).
It’s really easy to reduce Charlie Javice to Elizabeth Holmes, but I think we should also talk about the things Charlie did right.
She tackled something that is slow and bureucratic head on.
I don’t know how to bridge the gap between 300,000 and 4.25 million users. But using your server, the server which JP Morgan now owns, to commit fraud isn’t… smart?
So take one from Charlie Javice.
And accept the slow burn, PLEASE.
And that’s the skinny.
Wow! What a crazy story. Thanks for an in-depth, yet succinct overview of this.