The Two Essential Jobs

A firm theory of banks, fintech challengers and what can AI actually replace - Part 1

What is a bank? What does it matter?

I’ve spent the last 15 years of my career working in banking, but never even had the suspicion I was. Being a “banker” implied certain qualities: conservative ties, suits, measured and thoughtful demeanor. I was not a banker. I was an engineer, a consultant, a product manager, a technology “executive”. Furthest thing from a banker if you ask me. I’ve never worked for a bank. I worked for a consulting firm, Amazon, a Fintech startup. Where is the bank in any of that?

Misadventures at the edges of banking

However, it is a testament to how pervasive banking is in our economy, that I ended up working in banking nevertheless:

1. In 2010, spending a few months in Cherry Hill, New Jersey “transforming” the mortgage backroom operations for a Systematically Important Financial Institution. What did I know about banking? Nothing. But nothing beyond working brain cells was required to solve the problems they were facing: "How come every mortgage application has less than 50% of the required documents? Look at this one, it doesn’t even have a driver’s license”. Confused mortgage specialist looking at me suspiciously “No one ever asked us for this stuff.. loan officers would get angry if we asked”. Cue Steve Carell yelling “Yup, there’s a bubble” meme. Only 3 years later. That stuff was obvious, and our “solution” was equally so: “Naming and shaming is the only thing that works” suggested the grizzled banking executive-turned-McKinsey partner: “just put a big scoreboard in the middle of the floor with the names of the mortgage specialist and their application completeness rates"“. That worked.

2. Amazon decided to get into B2B retail, and realized that businesses beyond a certain size don’t just want to put in a credit card for every purchase. They want invoices, terms, quantity discounts, etc. Cue a couple of years of doing things that looks like banking operations, which we had no idea were: “we want to give deeper discounts to businesses you say? What is a business even? I guess we need to check with a third party database to get to know these businesses.. know your business is what we’d call it..hah”. Or what do you do if someone wants to pay on invoice, but we want cash now. There is this thing called factoring. Who would’ve thought? And so on. You don’t set off to work in banking, but you end up anyway.

3. Amazon Pay was an interesting one. We loved saying that we are just a money transmitter, not a bank (big implications to that distinction), but when 60-70% of your operations bandwidth is spent filling SARs, or monitoring the CBD contents of that hemp products seller, you’re doing more banking jobs without knowing it.

4. Finally, I spent the last 4 years building the product and technology operations for a Fintech, a cobranded-card-issuer. That at least nominally feels like a bank. But Fintechs don’t usually start off thinking of themselves as banks. Every fintech pitch is some variant of: “unlike a traditional bank, here is what we will do…”. But the truth of the matter is that it is wishful thinking most of the time. For starters, most Fintechs who are in any way involved in the deployment and movement of money are licensed indirectly as banks. That’s the whole Sponsor Bank industrial complex. “Rent a [bank] charter” is a less artful, and more direct way of saying it. The successful Fintechs realize that this is not a sustainable model. They have no option but lean in: either become a bank outright, or be ready to be a bank at any point (i.e. build up all the state-deputizing-you-to-be-the-law functions -more on that in Part 2). If you’re lucky, and very good, you end up being a bank.

But if all those things (technology companies issuing credit cards, online retailers selling on invoice, payment wallets moving money) are “banking”, what does banking even mean? Does the term have any value, or is it a catchall for money-related stuff. And why do you even need to think about that if you’re thinking about building AI to take over _____ insert banking thing.

The two essential jobs

However, I’ll contend that if you look across all these experiences there are two main jobs-to-be-done in banking (and a third one that has to be done - more on that in part 2).

1. Distribution

Money is everywhere. Its in saving accounts, pension funds, corporate treasuries, etc.. The obvious job of a bank is to move the money between those who want the return (or the convenience of not stashing money in their mattresses), and those who need the capital. That’s the economics 101 answer, and it’s obvious. But the less obvious is what makes a firm good at that operation. If you think of a bank as an oil company, there is an upstream operation (getting the oil out of the ground, or getting the capital out of people who have it), and a downstream operation (getting the refined gas to the gas station, or lending the money to those who can deploy it profitably).

Because we are mainly talking about consumer banking here, the upstream and downstream operations end up looking very similar. I have a checking account, and a credit card. A lot of people do. So “distribution” in this context refers to both the upstream and downstream operations. It is a many-to-many relationship that banks sit in the middle of.

But what makes a bank uniquely good at this? Traditionally, nothing beyond real estate acumen and state granted monopolies. There was 1 chartered bank in town, and they had a convenient branch on Main Street. Before ATMs, before the Internet. That’s basically what was there to it (not exactly, but we are getting to that).

Something like 90% of Fintech, challenger banks, Neobanks set to exploit that once the internet made it possible to provide banking services outside of a branch. Some benefitted from early mover advantage (PayPal), and others benefitted from the distribution advantage in other spaces: why does Amazon or Apple have a payment wallet business? Because they have a lot of customers already, and it is easier for them to get the customers to do 1 more thing, than for a startup to convince them to setup an account, and to trust them with their money. (Some take other approaches - my previous employer focuses on cobranded cards, which you can think of as go to market partnerships with mini Amazons and Apples - in retail, hotels, airlines - to basically benefit from the same dynamic).

We are so far into that dynamic now - mobile apps replacing bank branches - that it is legitimate to ask if anyone has any new ideas. Because otherwise, it seems like the distribution market is very efficient - with any surplus opportunity likely going to the big advertisers who have the adjacent eyeballs (Meta and Google).

2. Portfolio management

Unlike oil, money is very fungible. It also has time value. If I sell you gas in a gas station, I really don’t need to know much about what are you planning to do with it. You’re also paying me on the spot, and my margin is already baked in. For a bank, the spot value of the $1 I am lending you, is $1. There is no very little business in that. The only way the trade makes sense is over time. I’ll lend you the $1 today, and get it as $1.25 in a year. But all of a sudden, I have a lot of interest in what is going on in your life - at least during that year. Because lending money is easy, but getting it back is hard, the selection of who I am willing to lend to becomes very important. If anything, it becomes THE job of the bank. Constructing portfolios of borrowers (or lenders, as the dynamics work the other way around, especially with fractional banking, where I need to predict who is going to call in their loans) is what determines the difference between success and failure.

Now - and now I am getting into controversial territory- there is a lot of ink spilled, and research papers written about risk management, underwriting, machine learning approaches and big data. I won’t contend that there is no value in any of it. But I will say that - again at least in the consumer space - there is no excess alpha in most of what you do internally beyond the data available to everyone who pays a Transunion or Experian subscription. In the consumer space, in the US, FICO scores are pretty good predictors of default rates. That’s how they are defined in the first place. I am sure others can squeeze some additional alpha from proprietary data sets, but I haven’t seen enough practical evidence that these data sets are not already correlated to plain vanilla credit score signals.

So does that mean that there is no value in portfolio or risk management? Would everyone just converge at the same performance?

No! I think there are massive differences in performance, but they are not predicted by better data, models, or internal operations. The differences in performance are driven by the human at the top playing the portfolio management role. Most of the time that’s the CEO (if they are interested), or someone very close to them. And the difference in their judgment is not about the static analysis (i.e. what will happen if everything stays the same - again, FICO is great for that), but its in their ability to predict or divine what will change in the static situation. Are we likely to go into a recession? Is it time to go risk off? Is the risk too correlated - at the face of it - across all these businesses we got into?

Again, over here, you can see the problem: most of those portfolio managers are seeing the same macro data. So, a “best”, “normative” answer of what will happen next is unlikely to yield a huge difference in performance across firms. But that’s another topic where there is enough research on both sides, with evidence suggesting that portfolio managers do matter.

AI?

So can AI replace or do any of these jobs better?

If you look at distribution first, the answer is maybe, but it is likely to have to do with changes in other areas outside of banking that will change the nature of distribution in banking (like what happened with the internet and mobile). If people are spending more time with AI “friends” over the time they are spending on social media, or there are new devices that people are spending more time with over their phones, then - yeah - there will have to be new distribution models, and there will be new winners in banking who can get the attention of these borrowers and lenders in their new setting. So likely an exciting space, but my suspicion is that the consumer attention economy will have to change first before meaningful opportunities are obvious (and of course the ones who will see the non-obvious opportunities first win)

Portfolio management is an interesting one. At the face of it, it feels like the most obvious AGI-is-going-to-be-better use case. It is a lot of numbers, a lot of linear algebra and regressions. AI can deal with much more data, more quickly. So, yeah, AI can be a better portfolio manager. But the real question is: does it matter? And my suspicion is that it doesn’t for two reasons: 1) even if AI is better, it is unlikely to provide any firm-level advantage; it is unlikely that the big banks at least won’t have access to similar AI capabilities. Yeah of course there are differences in data, and how well they integrate them, etc.. but over time these differences equalize (how many banks are driving meaningful differentiated returns from the quality of their mobile apps - to use the last big change). More importantly, 2) the portfolio manager is the principal of the firm. So even if you as an AI do a better job in managing the portfolio than the de-facto portfolio manager, you’re not replacing them. That doesn’t mean there won’t be opportunities to augment them, or streamline risk and underwriting operations - billions of dollars of value will be created in that space. But at the end of the day - barring any extreme AGI scenario where they take charge of all the money - someone has to provide the intention to “do banking” and AI is not replacing that.

So does that mean there are no big AI opportunities in banking?

Far from it!

The good news is that these two essential jobs of banks (portfolio management at the top, and distribution at the street-level) actually represent a minute percentage of all the jobs in banking. There is a huge middle layer of activity that has to primarily do with how banks “fulfill” these two jobs, as well as the societal and political role that banks have been deputized to play that is more than ripe for AI disruption.

That will be the topic of Part 2, complete with analogies to sheriffs and bounty hunters in the old west.