Earning the Right to Win in Embedded Lending

Matt Harris at Bain Capital Ventures wrote the definitive piece on embedded finance back in 2019. In his two-part essay, he summarized the benefits of embedded finance this way:

When you look at the benefits of this embedded financial services model, the first point is obvious: as a technology founder, you’re already going through the hard work of acquiring customers, and as a result you have created the opportunity for a zero customer acquisition cost cross-sell. But the opportunity goes well beyond that basic business logic.

Having these financial functions integrated with software enables new functionality, leveraging the persistent connection to move beyond transactions to relationships. We’ve already been trained to conduct financial transactions inside of software applications (think payments inside of Uber), so if you’re utilizing software to run your business, using that same software to get paid and make payments is logical and more natural than going to your financial institution to do so. These relationships are data-rich, which leads to smarter cross-sell, prequalification and massive risk reduction. The monetization opportunities are not only large, but actually meaningfully larger than the original software opportunity.

Zero CAC. Proprietary data to inform marketing and risk management. The opportunity to cross-sell additional products and services.

To quote Dame Julie Andrews, these are a few of my favorite things.

And when we look at embedded lending specifically, we see these exact benefits paying off huge for banks and fintech companies:

Zero CAC? Take a look at BNPL companies, which have built the user bases for their shopping apps from the customers that their merchant partners sent them.
Proprietary data to inform marketing and risk management? Ask Goldman Sachs how nice it has been relying on Apple for ID verification for the Apple credit card.
The opportunity to cross-sell additional products and services? If you’ve ever bought a car at a dealership and gotten financing from the Finance & Insurance (F&I) guy, you’ll know how devastatingly effective this ‘cross-sell to a captive audience’ strategy can be.

All these embedded lending benefits are real and very compelling.

However, I think they miss something subtle and deeply important.

The Fight Against Adverse Selection

In economics, adverse selection occurs in a transaction between a buyer and a seller, when one party has more information than the other.

The reason that adverse selection is a problem is that the party with more information will be tempted to use that ‘hidden information’ to take advantage of the other party. The classic example is used cars. If I know that the 2001 Volkswagen Passat that you’re looking at is a lemon, but you don’t know that, I’ll be tempted to sell it to you for more than it’s actually worth.

In lending, the most common example of adverse selection is the discrepancy between what the borrower knows about themself (who they are, what they are planning to use the money for, how capable they will be of repaying, how serious they will be about repaying, etc.) and what the lender knows about the borrower.

This is, fundamentally, what makes lending money really difficult.

Lenders are just stores with big signs on them that say, “Money for Sale!”

And when you operate a ‘money store,’ you get all kinds of folks walking in your front door. They know who they are and what they want, but you don’t.

Thus, your job, if you are going to operate a profitable lending business, is to minimize this information asymmetry, and identify and accurately price the risk of the applicants that you want to work with.

This is the purpose of loan underwriting.

Join Fintech Takes, Your One-Stop-Shop for Navigating the Fintech Universe.

Over 36,000 professionals get free emails every Monday & Thursday with highly-informed, easy-to-read analysis & insights.

No spam. Unsubscribe any time.

Everything that a lender does to underwrite a loan – from pulling credit files and credit scores to measuring the speed and cadence with which an applicant types in their annual income – is designed to shrink the information asymmetry gap and reduce the odds of getting hurt by adverse selection.

Is this person a criminal applying with a synthetic identity? Is this person in some financial distress that they’re not telling me about? Does this person have a track record of managing their financial obligations well?

These are the types of questions lenders are trying to glean answers to during the loan underwriting process.

Now, obviously, there is a practical limit to the amount of effort that a lender will put into fighting adverse selection. They have to make a risk/reward calculation. The juice has to be worth the squeeze.

One consequence of this is that loan underwriting looks very different for different loan products. In mortgage lending, it’s not uncommon for an underwriter to call an applicant’s employer on the phone to confirm their income and employment status. In BNPL, that would be an insane thing to do. The margin wouldn’t come close to justifying it.

Another consequence is that lenders will usually design the sequencing of their underwriting process to be maximally efficient at sorting out the good applicants from the bad ones.

A simple example of this is asking for an applicant’s birthday on the very first page of the loan application. If the applicant is under 18 (and doesn’t have a cosigner), then the lender is legally unable to enter into a lending contract with them and they can stop the process right there. The hidden information in this case was the applicant’s age and once that information asymmetry is eliminated, the lender can reject the applicant and move on.

A more nuanced example is the credit score.

One of the nice things about credit scores is that they (and the data they are derived from) are broad-based indicators of risk. They can help lenders accurately predict if an applicant will pay back a loan based on their prior history of doing so. They’re not perfect (newsflash – no statistical model is!) nor are they completely comprehensive, but they are efficient.

No sophisticated lender uses just credit scores to make a lending decision, but most do use them as an early sorting mechanism – is this applicant sufficiently risky that I shouldn’t bother continuing to uncover the rest of the hidden information about them?

But, again, credit scores aren’t perfect.

FICO can tell you what the statistical likelihood of default on a lending obligation for a consumer with a 660 FICO score is, but it can’t tell you what that specific 660 FICO consumer will do. It’s an average. A useful average, but still an average. By definition, that means that there are 660 FICOs out there that perform like the average 650 FICO, and there are other 660 FICOs that perform like the average 670 FICO.

Credit scores tell lenders where to stop spending their time (the obvious declines and the obvious approvals) and, crucially, where to spend more of their time (the promising maybes).

A significant source of alpha for lenders is finding those positive outliers; the 660s that perform like 670s. So this is where the smartest lenders spend a lot of their resources – getting their hands on even more data, applying lots of smart math to that data, observing the results of their lending decisions, and then doing it again and again and again as quickly as they can.

This is the realm of credit risk management, where nerdy giants like Capital One reign supreme.

(Editor’s note – Francisco Javier Arceo has a great overview of how the operational and analytical parts of lending money work in his newsletter, which you should subscribe to if you haven’t already.)

However, there is a different, less-well-understood path to finding these positive outliers.

Selling Money in Someone Else’s Store

Let’s go back to the central challenge of adverse selection in lending.

You are operating a money store. You know that everyone walking into your store is looking to buy money. You know that these prospective customers have hidden information that may give them an advantage over you. You have to efficiently expend your limited resources to uncover that hidden information and discover which of them you want to work with.

Allow me propose an alternative.

What if you were selling money inside someone else’s store?

Now we’ve changed the equation. The denominator is different. The population of folks in your store isn’t made up of people looking to buy money. It’s made up of people looking to do something else entirely.

These folks still have an informational advantage over you. However, two things are different. First, the store that you are selling within has information on these prospective customers. Information entirely outside the realm of what direct lenders have access to (this is what Matt Harris was referring to in his essay). And second, the impact of that information asymmetry (which no lender can ever fully eliminate, no matter how sophisticated they are) may be different because the underlying characteristics of the store’s population may be different. In other words, the store may have already positively selected for a segment of consumers that you would, all things being equal, prefer to lend to.

That second point is really important, so allow me to elaborate with a real (but anonymized) example.

There’s this bank.

Its main business is auto lending. Near-prime auto lending. Borrowers who have FICO scores between 600 and 680. Most of the financing it provides is for the purchase of used cars.

These borrowers are sourced through relationships that the bank has built with independent car dealerships. These dealerships are typically not big enough to have their own dedicated F&I guy, so the bank functionally plays the role of an outsourced F&I office for the dealerships, especially on busy weekends when the owner needs to be out on the lot selling cars.

When a customer needs financing, the dealership literally calls the bank on the phone and then hands the phone across the desk to the customer. The underwriter from the bank then talks the customer through the entire underwriting process, in real-time, and closes the loan.

During that process, the underwriter (who is experienced, well-paid, and working out of a U.S.-based call center) helps the customer navigate any potential difficulties – everything from helping them set up two-factor authentication to helping them calculate their annual income based on their weekly take-home pay. They also expend a lot of time verifying information about the customers, which can include such manual tasks as looking up the listed work address on Google Maps and verifying that it visually matches the business type listed on the application.

The bank employs data scientists and has sophisticated risk models, but a lot of the discretion for approving the loans is left to the underwriters. The terms for approved loans are generally fair, but not unusually competitive.

To most lenders, this wouldn’t make any goddamn sense.

Near-prime auto lending through independent used car dealerships is a risky arena to play in. Capital One, that paragon of analytic sophistication, ramped up its lending in this area over the last couple of years but it has recently slammed on the breaks and appears to be preparing for significant losses.

If you are going to survive in this market, conventional wisdom would suggest that you need to be incredibly efficient at sorting goods from bads. Automation and analytic sophistication should be your biggest priorities.

But that’s not what this bank’s underwriting process prioritized at all!

Getting on the phone with each customer? Helping them with basic tech support and looking up business addresses on Google Maps? You might do this for an especially promising mortgage customer. You don’t do it for near-prime auto loan customers buying used cars.

This model shouldn’t work.

But it does.

This bank has been operating with this model for decades, and it consistently delivers results that outperform its competitors by a significant margin (its delinquency rate is consistently below 3%, which is extremely good in near-prime auto lending). Heck, it has been thriving in a market that the mighty Capital One could barely dip its toes into.

How is it doing this?

Well, the biggest reason is that it’s very picky about the dealerships that it works with.

The bank understands that a 660 FICO score is just an average. It finds the 660s that perform like 670s (and 680s and 690s) by building relationships with the dealerships that attract those customers. Those customers who are, on average, older, less tech-savvy, responsible, and dependent on their cars for their livelihoods.

Imagine a master electrician, 62 years old, who earns good but not great money, is a bit sloppy with his credit card payments, but is extremely serious about always having a reliable (not flashy) truck that he can carry all his equipment around in. This is the type of customer that this bank wants, and it works extremely hard to earn the loyalty of the dealerships that positively select for these customers.

That last part is key because it explains the bank’s anachronistic underwriting process.

It invests in a process that works well for these customers (and the dealerships that attract them). If the customers prefer talking on the phone with a human to fiddling with a mobile app, no problem! If they need help parsing questions on the loan application or setting up two-factor authentication through their email, the underwriter will be patient as a saint in walking them through everything they need to do, step by step.

These investments don’t make sense for the average lender because the average lender doesn’t have this bank’s positive selection advantage. Everyone else is selling money in a money store. This bank is playing a different game.

And it knows it.

This bank understands what its unique advantage is. It invests in the areas that sustain and strengthen this advantage – paying its loan underwriters above-market salaries, for example – and, importantly, it doesn’t invest in areas that don’t.

This is more difficult than you might imagine.

Everyone who works in the financial services industry has an ego. We all want to be thought of as cool. We want to partner with the sexiest brands. We want to have the shiniest technology and the smartest algorithms. We want to be interviewed on stage about how our company is on the cutting edge of (insert your fintech buzzword here).

However, in order to seize the full advantages offered by embedded lending, we must suppress these instincts, individually and organizationally, as much as we possibly can.

Beware Ego

American Express provides an instructive example.

In the late 1980s and early 1990s, AmEx was struggling. Merchants were aggressively pushing back on the company’s high interchange fees and the company’s brand was struggling. Kenneth Chenault, the newly appointed CEO of AmEx, turned things around. He landed a series of new partnerships with large retailers and expanded AmEx’s merchant footprint substantially.

His biggest win was Costco.

In 1999, AmEx signed a deal with Costco to be its exclusive credit card partner. The Costco American Express card functioned as the retailer’s most cost-effective membership option and AmEx was the only credit card brand accepted at Costco’s 221 stores.

It was a perfect fit for AmEx’s closed-loop issuer/network model and Costco, which commanded insane loyalty among its affluent customer base, proved to be a huge growth engine for AmEx. By 2004, the number of Costco warehouses in the U.S. had increased to 327. And the total number of AmEx cards rose from 46 million to 65 million. By 2015, 10% of the 112 million AmEx cards in the market were Costco-branded.

But eventually, ego got in the way:

The Amex people, most of whom had MBAs, sometimes found it amusing to deal with Costco veterans who spoke about starting out stocking warehouse shelves. Less endearing was the habit Costco executives had of referring to Amex as a “vendor.” That made the Amex people seethe. After all, they represented one of America’s oldest corporations. But they smiled and said nothing, and the corporate marriage endured for 16 years.

In 2014, Costco put its credit card business up for bid, in an attempt to get the best pricing possible. If you know anything about Costco, you know this is standard operating procedure. The company is fanatical about negotiating the best deals possible with all of its vendors, in order to pass those deals on to its customers without selling anything at a loss. It’s not personal. It’s simply how they uphold their value proposition to their customers.

AmEx took it personally:

Amex wasn’t happy about competing with global banks such as Citigroup and JPMorgan Chase and its archrivals Visa and MasterCard. But Chenault fought for the deal—even though his company might actually lose money in some cases when Costco customers swiped the card. As the negotiations dragged into January 2015, however, he became agitated and called his counterpart to remind him that Amex hadn’t only furnished Costco with its prestigious card; it had been Costco’s “trusted partner.” Jelinek interrupted, according to people who were briefed by Chenault about the call, and told him that as far as he was concerned, Amex was another vendor, just like the one that sold Costco ketchup. “If I can get cheaper ketchup somewhere else, I will,” he said. As rumors about the call spread, the rank and file who heard about it couldn’t believe someone from Costco had the nerve to insult Amex like that. Ketchup! Chenault called Jelinek a few weeks later to say Amex was pulling out.

In the immediate aftermath – which was devastating to AmEx’s share price – the company tried to spin the breakup as a coldly rational economic decision. However, everyone close to the negotiation said that it was, at the end of the day, an emotional decision.

AmEx simply couldn’t tolerate the idea that its success was, even partially, due to something other than its own storied brand.

And so it got itself kicked out of what was, perhaps, the best possible store in which to sell money.