Payments Settlement, Network Effects, Free Money
To understand the motivation for retail banking clients, it helps to think not about the process of issuing a loan, but first, the process of spending it. Imagine that Alice has just received a loan for \$10,000 from Acme bank. She uses \$8,000 to purchase a used commercial oven for her bakery, from private seller Bob, who banks at FizzBuzz. Alice writes a check and gives it to Bob, who both delivers the oven and deposits the check at his bank.
Now, there are many check clearinghouses and payment settlement systems, but one of them is the Federal Reserve (in the US...other countries have a corresponding central bank that plays a parallel role). Let's suppose the check makes its way there for clearance and settlement. The Fed determines that the check is legitimate, and moves funds between Acme and FizzBuzz. In particular, both banks have an account at the Fed (which is a bank for banks...the central bank is a bank whose customers are themselves banks or the national treasury). It thus debits \$8,000 from Acme's account, and credits the same to FizzBuzz's account. Assuming that Acme is not overdrawn, the transaction completes and all is well.
But what happens if a lot of Acme customers had written checks recently, and all of them just happened to get cashed (deposited, settled, etc.) today? Well, it turns out that Acme can overdraw its account! And when that happens, the Fed allows it, but forces Acme to borrow the balance via the overnight Fed Funds mechanism. That is, they get a one-day loan to balance their Fed account at whatever interest rate is set by the Fed (one of the mechanisms the Fed uses to influence economic activity). The next day, Acme must pay back the loan by depositing the necessary funds into its Fed account, with interest. Don't worry, though! In most cases, Acme's customers will also be receiving checks from other banks, which helps replenish Acme's reserve account with the Fed. As long as Acme retains adequate reserves, it will be on either side of the overnight loan equation about the same amount of the time.
Reserve Requirement
Now, for large banks, the Fed requires, by law, that the bank maintain a minimum balance with the Fed (just like, retail customers with fancy accounts with lots of perks may also be required to maintain a minimum balance). And it turns out that this balance is a fraction of the deposits the bank has on its books (10% is a common number, but some jurisdictions actually have no reserve requirement). And from this, we get the commonly cited claim that banks "loan out the money deposited by bank customers". However, there's a problem here.
First, the Fed doesn't care where the reserves come from. The bank can raise money by issuing stock, selling bonds, selling cupcakes, playing craps at the casino...all the ways that anyone else can raise money (well, within the regulations imposed on banks). So it's entirely possible for a bank to fund its reserve account with money raised entirely in capital markets (that is, from investors, not retail depositors). Indeed, a bank funded this way can issue loans without a single "outside" deposit (meaning, from money which comes from a customer).
Second, note that the reserve does not need to be a constantly growing pool of money. It's primary function is to facilitate inter-bank payments, and it really only needs to be big enough to cover all the outgoing payments drawn against its accounts on any given day, with the stipulation that the reserve will also be replenished by payments arriving into its accounts. On average, the reserve should maintain a somewhat stable size. Frankly, it doesn't really matter if the number and size of deposits is growing or shrinking, as long as the reserve is adequate to cover payments settlement, as well as the Fed reserve requirement (which is mostly just a safety check more than anything).
Third, it should be noted that a the Fed only imposes a reserve requirement on banks with more than $16.3 million in assets. Thus, a small bank literally does not require depositors. And yet, it can still issue loans.
Net Settlement
The Federal Reserve is a bit strict in that it requires all the daily checks flowing from Bank A to Bank B to be handled as one transaction, and all the daily checks flowing from Bank B to Bank A to be handled as another transaction. This artificially inflates the reserve requirements. For this reason, banks often use other settlement systems which have been set up between them (especially regional banks). These systems allow "net settlement", which means that if Acme Bank owes \$25 million in payments to FizzBuzz, but FizzBuzz owes \$29 million to Acme, the settlements system will say: "You can handle the \$25 million in transactions internally. We are only going to create a single payment for \$4 million from FizzBuzz to Acme." Notice how this dramatically lowers the reserve consumption at the central bank. This avoids a significant amount of the overnight borrowing imposed by the Fed, and results in lower fees for the banks.
But wait, there's more! We sort of hand-waved what is going on during net settlement. We know that thousands of checks got drawn against Acme accounts to pay FizzBuzz customers, and vice versa. But what happened to that \$25 million that never actually moved between banks? Well, it's simple. One of those checks is from Alice (Acme) to Bob (FizzBuzz). But another check was written by Darren (FizzBuzz) to Candace (Acme), for \$2,000. Rather than move \$8,000 from Alice's account to FizzBuzz, Acme just moves \$2,000 from Alice's account to Candace's, and then looks for more incoming payments to dispose of the rest of Alice's check. Of course, that's not literally what's happening. In reality, the accounts just get credited and debited as if the proper transaction occurred. But logically, the money to pay Paul comes from Peter, because both Peter and Paul are customers of the same bank, and this avoids intermediation by fee-charging third parties. Of course, the settlement houses still charge fees for their service, so they haven't really gotten away from fees...they have just reduced them by a significant margin.
Now, we get to the coup-de-grace: the "free money". It's actually only one of the free monies. It turns out there's a lot of free money in banking (hence, why it's so popular!). What happens when Candace writes a check to Alice? Since they are both customers of Acme, the check never has to leave the bank! Acme will quite happily settle the check internally, without engaging any third-party settlement service. When this occurs, a bank officer does a little happy dance. And how often does this dance occur? Well, about 30% of the time!
Network Effects
By this point, I hope it is obvious why banks want retail customers: when both the payer and the payee are customers of the bank, the bank saves money processing the transaction. This is especially true when a bank customer receives a loan from the bank, and "spends" it at that bank (by purchasing goods and services from other customers of the bank). The more customers use the same bank, the more the bank can lower its fees and raise its savings rates. This is the "network effect" which drives customer-seeking. Remember, when a customer makes a transaction with another bank, both banks pay fees to a third party to process the transaction. If the reserve is exhausted, they pay even bigger fees for overnight funds. So, keeping transactions "in-house" is extremely attractive and financially lucrative for banks. Every bank thus seeks to obtain every possible bank customer, just as corporations generally seek to obtain a monopoly to maximize their profits.
Overnight Funds
Of course, the sequence of payments is a more or less random process, as far as the bank is concerned. It cannot control when its customers write or deposit checks and other payments. On any given day, the transactions will mostly balance and no overnight funds will be required. However, when funds are required (due to an unlucky series of outflowing payments without a balancing inflow), the bank has to somehow raise money. Of course, there's lots of ways a bank can do this. It can sell liquid assets it has on hand. It can take a short-term loan. It can issue CDs, etc. But why bother with all that hassle when it is sitting on a huge pile of "free money"? That's right, I'm talking about customer deposits. Since a majority of checking accounts have zero or close-to-zero interest rates, this is literally the cheapest temporary money a bank can obtain. In most cases, it doesn't need the money for more than a day or two. It just smooths out the blips in the random sequence of payments between banks.
So you see, the deposits don't facilitate the origination of loans; they facilitate the efficient transfer of payments between banks (of which the majority of money happened to come from a loan, but that's incidental, rather than a necessary requirement).
Money Creation
Now, this question arose from an investigation about money creation, so I think it is worthwhile to consider that question briefly. First, we must consider the nature of a deposit account. When a customer deposits money into an account, they give up legal claim to that money. What they gain is a promise from the bank to return any portion of the money up to the account balance, upon demand. Basically, a deposit account at a bank is a kind of instant loan issued by the customer to the bank, with an indefinite, variable period and an absurdly low interest rate, callable at any time (within the restrictions set out in the account agreement...for savings accounts, there is a maximum number of withdrawals per time period). Of course, the customer wants to issue this loan because the bank will facilitate payments to third parties.
The important fact to note is that, as far as the bank is concerned, a deposit account is a liability. It is something they owe. When money flows into such an account, the liabilities of the bank increase by the amount of the deposit, but so do the assets. The bank now receives the deposit as cash (quite literally, a credit to the "Cash Account", if you like), to spend as it sees fit. As noted above, one of the important ways it spends the deposit is as a short-term cover for payments settlement. But what about loan origination?
When the bank issues a loan, it "deposits" (credits) the loan amount into the customer's deposit account. It then debits the customer's loan account, which is an asset of the bank (it's money owed to the bank in the future). Some folks have implied that this proves that customer deposits fund the loans, but no such thing has occurred. You see, if customer deposits were used to "pay for" the loan, then one more entry would be added: a debit to the bank's "Cash Account". In this case, the actual money that the bank received from customers would be used to fund the loan. But no such transaction is recorded! That is the important point made by the article you cited.
In fact, what has actually happened is that the bank has created an "IOU" and handed it to the loan borrower. The borrower can then spend this IOU as if it were real money, because our legal and financial system says that it is real money. But the reality is, this money was created in the instant that the bank credited the customer's deposit account, without debiting any corresponding cash account. As it turns out, when the customer pays down the loan, payments will then "destroy" the money created by the loan, as they cancel out.
Time Travel
While the money creation/destruction model works just fine for describing what happens in loan origination, there is an alternative way to view the situation which I would like to share. If you were paying close attention, you would notice a small sleight of hand I used in the description above. I said that money creation occurred because the bank credited the borrower's account without debiting a "cash account". But the bank didn't "cheat". It did debit an account: the borrower's loan account. Of course, a loan account is like the inverse of a deposit account: the bank is the one which gets to demand money, up to the amount of the loan. Of course, they are not perfectly symmetrical, but they are sufficiently symmetrical to balance the books. So the money to pay for the loan came from somewhere...not the other depositors, but definitely from somewhere...but where? Well, it should be obvious: it came from the future! After all, that's where the loan account dollars flow into the bank. A bank loan is a time travel device which transports future dollars into the present for use by bank customers! And, just as time loops violate causality and allow a person to be their own grandpa, a dollar in a time loop may also violate causality and become its own "grandpa".
A time traveler which creates himself violates conservation of energy laws because the mass of the time traveler is removed from his future when he travels back to the past, and is added to the past, which is now "heavier" than the previous past by the amount of the time traveler. The duration in which the self-created time traveler exists causes the universe to be "fatter" by the mass of that person. In the same way, the money supply of a universe with time travel is not conserved. The duration in which a money time loop exists (i.e., a loan) causes the money supply to be larger by the amount of the loan. This, IMO, is the best way to see that a loan truly, literally, creates money out of thin air.
Now, we have a problem. What if the borrower breaks the time loop? What if the loan payments do not ultimately destroy the loan, and the borrower defaults? Will this cause a rip in the fabric of spacetime? Well, yes, of course it does! This is why banks demand collateral: the bank tries to repair the rip any way it can, including taking property that has nothing to do with the loan, and using it to patch the rip.
Reification
If the money was created out of thin air, why is it such a big deal for loans to be repaid? Can't we just say: "Oh, it's ok, that was just made-up money anyway. No loss"? Well, yes and no. There are two major problems with loan defaults. The first one is obvious: banks make revenue on loans. Banks are not charities run by volunteers. They are profit-making institutions run by people who expect to be (highly) compensated. And the operating expenses of the bank are generally paid by the interest earned on loans. So, a failing loan == failing revenue == operational shortfall. Bad News.
The other problem has to do with inter-bank payments. If a borrower and all her payees are customers of the same bank, the IOU which is the loan can exist as a kind of magical play money that bankers are allowed to manifest at will. Moving this money between customer accounts just amounts to credits and debits in the accounts on the books. That's because when the borrower says: "I'm cashing in part of this IOU", the bank says: "No problem, the person you're paying is also my customer, so this becomes a kind of IOMe." The problem is when a payment goes outside the bank. At this point, "real money" must exist to cover the payment, and that money must exist in the Fed account for the bank. It is at this point that the "play money" created out of thin air by the loan origination process becomes "real money". I call this process "reification" (but nobody else does). The problem should be obvious: just because Acme bank declares: "This new money exists. I say so." does not mean that FizzBuzz is obliged to recognize that statement when processing a payment. FizzBuzz says: "Nuh-uh. Don't give me that monopoly money. Give me the real stuff!" This is why the Federal Reserve account must be funded with "real money". Otherwise, banks would just always declare that they have the money needed to cover their payments, and never borrow from anyone. Essentially, they would print money indefinitely, and with reckless abandon.
In this narrow sense, the depositor funds play a role. They provide the cold, hard cash money to facilitate the payments between banks, when a borrower tries to spend the newly minted money somewhere else. The depositor dollar "reifies" the future dollar manifested by the loan and spent outside the bank. It becomes the "physical" stand-in for that future dollar that will (hopefully) eventually land on the bank's books. And it is only in this sense that depositor funds are "necessary" for loan origination. However, recall that the reserve account need not be funded by deposits, and thus, deposits only reduce the cost of payments settlement.
Of course, this only works because a bank, on average, has as many payments coming in to the bank as going out. If a bank issues loans, but has no customers who receive payments from other banks, then the bank is clearly going to have a liquidity problem in short order. By acquiring retail depositors, a bank helps ensure the balanced bidirectional flow of funds with other banks. The more balanced the flow, the less reserve required at the Fed. This is the essential role of depositors, and why they are absolutely essential to a bank which issues loans. When a borrower spends a loaned amount, that money becomes a deposit somewhere else, as it flows out of the bank. In order to minimize the amount of base money required by the bank, the bank wants to be a target of loan spending as much as it is a source. Thus, it wants lots of depositors receiving payments from other banks. The netting process causes most of these payments to stay "in-house", with far fewer dollars being "reified" by the Fed account. It is much less concerned with literal cash deposits, which are mostly a rounding error.
Note that all of the money could, in theory, come from loans, with nobody actually depositing M1 into any of the banks at any time! If payments between banks are perfectly balanced, then no reserve is required, either. All the money moving between customers could theoretically be loan-originated money with no currency or M1 involved at all. This is why "broad money" is such an important category in macroeconomics.