What's next in the personal loan market
Part 1/2 - who's the real customer, myths, and a shitty profit pool analysis
This is a bit of a change of pace from my previous posts. It does not signal a new direction. My first draft of this was written over two years ago and my 'conclusion' (in the next post, sorry!) still hasn't happened, so maybe I'm wrong... I'll mostly be picking on LendingClub ("LC") here since they have the most public data.
The personal loan market has blown up in the last ten years. 20.2M Americans have unsecured personal loans as of 3Q19 with $156B in outstanding balances, thanks in large part to QE, LendingClub, and Prosper.
Over the years, we’ve learned a few lessons:
P2P doesn’t work at scale because there are very few individuals who both have money, and want to actively pick loans
After Prosper failed to raise/sell in 2016, it essentially became a captive origination platform for institutional buyers
Upgrade (LendingClub founder’s second try) went full institutional
LendingClub has gone from 16% self-directed retail (1Q16) to 4% SDR (3Q19)
Institutional pickers, however, are still quite hungry and opinionated
GS decided to stop buying from LendingClub and build its own consumer division (Marcus)
Entire businesses like Theorem have emerged to buy consumer credit
LendingClub has launched LCX, a real-time bidding platform for loans
Lesson from Product School re: online car sales: Who’s the real customer?
Alternative data for underwriting is, so far, a dud. Everyone above a certain scale basically uses FICO.
ML is tricky to use in a legibly/provably-not-racist way
A lot of data are still unavailable in easily consumed ways and introduces friction
Plaid / {utility,cable,cell bill} / TurboTax needs consumer consent and a login
However, Credit Karma and LendingTree are slowly accumulating this same data
What also matters is how the product is sold - if you’re securitizing a block of ‘prime’ it’s a lot easier for your capital markets team to point to FICO than to say with a straight face ‘our unseasoned model thinks that this person who’s at 620 FICO is actually more like 710 equivalent’
All the money that was saved by not having branches… went toward marketing expense.
Retail/branches are CAC. Or CAC is the new branch cost. Same as e-commerce.
In 3Q19, 44% of LC fee revenue went to marketing expense
So, what is the value that these originators provide? They get a ~5% fee on originations, of which ~3% is marketing expenses and ~1.5% is origination expenses. And a lot of those origination expenses are either undifferentiated/commodities (credit pull, The Work Number), or not core competencies / outsourcable. Looking at LendingClub, ~40% of its non-core workforce is outsourced, which is probably most of the call center.
In the third quarter, 19% of our total workforce resided in our BPO partnerships, while 48% of our workforce was located outside San Francisco - LC earnings transcript, 3Q19
So, in summary:
Buyers of consumer credit will largely be sophisticated bidders. The real customer in this market is the institution, not the consumer.
There’s not much secret sauce in underwriting, and a lot of the useful data are difficult to aggregate
Originators (or origination partners of WebBank / Cross River) fail to keep most of the value (approximated by revenue) that they generate, losing it either to customer acquisition (Credit Karma, LendingTree, Nerdwallet, Google, Axciom, credit bureaus, and the good ol’ USPS ) or origination expenses (BPO)
What does this lead to? What other market does this start to look like? Stay tuned for part 2 (hopefully tomorrow)…