.png)
Everyone’s very pleased with the archaeologists, “Oh, well done, you’ve discovered all this ancient stuff,” but in fact it’s shaming, because they should have found it ages ago. It’s about time archaeology finally caught up. We need to be finding the things that have become buried in mud almost immediately.
- David Mitchell
Something like 8,000 to 10,000 years ago, during a time when the world was radically different from how it is now, a young man concentrated on exactly the same task that I sit down to concentrate on today.
His methodology is different from mine, but the outcome is precisely the same. A semicircular clay shape in his hand is almost finished, if he can straighten out a kink in its lip. A handful of small clay tokens, different in shape and size and some with special inscriptions, lie on his workstation. Once he’s done with the semicircular piece, he deposits these tokens into its basin. He finds the other semicircular piece and joins the two halves together, forming a hollow sphere with the tokens nestled inside. The important work done, he only needs to decorate the sphere, called a bulla. He takes the bulla to the farmer, and cautions him: It should be stored somewhere cool and dark where the sun won’t crack the clay. If the bulla’s outer shell is broken, it will be treated as invalid come the winter.
This ancient device is one of our first recorded material transactions: The small tokens within the bulla encode the quantity of grain that the farmer deposited in the communal silo. The bulla’s integrity verifies the authenticity of the tokens. When the farmer comes to retrieve his grain from communal silo, the young man will crack open the clay of the bulla and make a count of the tokens within. They represent the amount of grain the farmer can withdraw from the communal silo.
I, meanwhile, sit down at my desk some thousands of years later to examine a very similar packet. Instead of clay, we’ve got software—fine, both materials are designed to be malleable. Where the young Sumerian uses clay tablets, I can use little packets of JSON. Where the young Sumerian uses a bulla for authentication of the tokens within, so I use OAuth 2.0. Potato, potato (this never works in writing, I don’t know why I keep trying it).
Both of us are involved in the transaction chain. We are, in the end, accountants. Sure, the young Sumerian may technically be a craftsman and a sculptor, just as I am a software developer, but we are both in the service of accountancy. If I brought this young Sumerian from nearly 10,000 years ago into my office today to show him how his profession evolved, I wonder what he’d say. I suspect it would sound something like: “What the hell happened here?”
The young Sumerian keeps his records very straightforward. They record all of the information that he requires—he must know the owner, and the quantity of the grain. Two fields. Easy. I might tell him that we generalized his system of accounting to work for products other than grain—livestock, ale, wine, honey, whatever. I suspect he’d nod and observe that a slight diversification of the shapes and markings for the tokens inside the bulla would solve this problem. This is, in fact, how the ancient Sumerians did solve the problem.
I might tell him that the quantities get very large. Rich men and kings have quantities of material wealth that require a large number of tokens to represent—perhaps too many tokens to fit inside a structurally-sound bulla! Maybe I’ve stumped him. Then again, he could just go and invent Cuneiform tablets, which is, again, exactly what the Sumerians did.
The invention of writing opened up the doors to a lot of complexity in economics, but our translation of cuneiform doesn’t always help clear things up:
3(N48) 1(N34) 6(N14) 1(N01) 1(N39a) , U4x(3(N14).7(N01)) SZEa DUBa LAGABbxLAGABb (KUb1 SZIMa)a SZAm2
Great. To translate (remembering that there are still some things modern scholars have not deciphered):
“5617 1/5 of something, exchange barley, 37 months, Kushim’s final account. ca. 135,000 liters."
The young Sumerian would recognize the outline of his bulla’s purpose in this message. It is still, despite its relative complexity, just a transaction: The writer received 5617.2 of something from a fellow named Kushim; the tablet writer, to pay for this something, has transferred 37 months worth of barley to Kushim. Kushim’s account, following the transaction, stands at 135,000 liters of.. I guess… barley?
I’m impressed! Despite one or two missing words (which, important to note, are only illegible in hindsight; they were perfectly plain in Kushim’s time), I can make sense of this transaction. I can make out amounts and parties. I might be able to calculate the exchange rate, just as soon as the archaeologists work out what the something in question is. The information even conveys that the tablet-writer is managing a sort of bank, since they are able to keep tabs on Kushim’s running account balance.
Obviously, our economic systems are different. We run on a fully monetized system, while Kushim and his contemporaries are running on a system of debts or a gift economy where goods are exchanged instead of currencies. That means that Kushim’s transaction doesn’t fit neatly into modern transaction data, but it’s close enough that we could give it a go. Here’s a visualization of what Kushim might see, if he were using a modern banking API:
{
"user_id": "kushim",
"description": "transfer to bank of uruk",
"type": "transfer",
"exhange": {
"out": [{ "product": "something", "amount": 5617.20, "unit": "unknown" }],
"in": [{ "product": "barley", "amount": 37, "unit": "months" }],
},
"running_balance": { "unit": "liter", "amount": 135000 }
}
This is a little complicated, but it gets the information across. Our modern software could handle this kind of thing (it would help to know what something is, but we’ll just have to use our imagination for now). As a modern engineer, though, I’m eager to introduce Kushim and Sumeria to the notion of currency. That will make the data so much cleaner, I say. I show Kushim what the transactions look like after the introduction of a standard currency:
{
"user_id": "kushim",
"description": "transfer to bank of uruk",
"type": "transfer",
"amount": 200.00,
"running_balance": 700.00
},
{
"user_id": "kushim",
"description": "transfer from bank of uruk",
"type": "transfer",
"amount": -200.00,
"running_balance": 500.00
}
It’s going to take some explaining to help Kushim understand that the commodities have been standardized with prices, and that a “trade” probably needs to be split across multiple transactions, but I think he’d understand. Where he might have reservations, though, is in the loss of information. The itemization has disappeared from the records, papered over by currency. How do we know what the money was exchanged for, he asks? Great question, I say, my palms getting sweaty. We encode that into the description. Kushim notes: “The descriptions don’t seem to have all the information anymore.” I say, “It’s only an example.” He asks, “Do you have any other examples?”
Oh, no. I do have other examples.
I show Kushim a smattering of transaction descriptions. Some of them are relatively understandable. Seeing others, however, he’s livid. The following examples of transaction descriptions are reproduced from real transaction data that FinGoal has encoutered:
$AVE800 1234656 COMBINED 801-234-2939N00921321 PO BOX 4059 SEE REMITTANCETRN221556061*3632221682~What do these tell us? The first suggests the money may have gone to someone called $AVE800 and offers us a phone number. The second, infuriatingly, ostensibly directs our attention to a physical letter where the actual transaction details might reside. The final one, perhaps the most obtuse of all, tells us only that it is a transaction - the only piece of information that we already knew.
Compare these transactions, which were written to the ledger of history within the past year, to our Sumerian examples written lifetimes ago. The difference is shocking: It is easier to understand a transaction from thousands of years ago, than it is a transaction from last week.
I sit down with Kushim and let him cool off before explaining. First, I note that some information—like the date and the amount—are saved in other transaction fields, so we haven’t lost them. Then I explain that the introduction of currency means that the actual items changing hands matters less to the general accountancy system than it did in Kushim’s day. While the business in question keeps an itemized receipt—important for their purposes—the larger accounting system doesn’t; whatever $N$ number of items at $P$ price was bought is losslessly represented by the amount, $N * P$ . We know the date it happened on (kind of, some interpretation of ‘know’ in this sense could be the topic of a different article, if I’m honest). Beyond that, most of the other information is cosmetic rather than essential. The system doesn’t need it to operate.
“Begging your pardon, but what a load of ox shit,” Kushim says. If it was his transaction, he’d like to know what exactly he bought with his money. If the accounting system of the business he transacted at preserves that information, why shouldn’t it be communicated to him? Why should he, looking at his banking ledger, not know what the money is going for? Why isn’t it required to be legible?
I read the anger in Kushim’s eyes and grasp for an explanation. Ultimately, I settle on the horrible truth: Our transaction is legible. It’s just not designed for humans anymore.
I don’t think Kushim or the bulla sculptor could have conceived of vast sections of modern technology. They certainly wouldn’t guess that the human would someday be mostly removed from accountancy. And they wouldn’t guess that we’d all be just fine with it. In Kushim’s time, cuneiform tablets needed to be legible at two endpoints: One, the accountant at the bank of Uruk; the other, Kushim. There was no machine, no reader to plug the tablet into. Therefore, human legibility was the only legibility that mattered where accounting was concerned.
Of course, now we do have machines—readers and printers both. To explain it to a Sumerian is a ham-fisted analogy: On one end, a machine uses an automatic stylus to press a cuneiform tablet with requisite transaction data when a transaction occurs. When it’s done, it launches this information catapult-style to a reader, which catches it. Sometimes, the catching machine is also a printing machine. It reads the tablet it received and presses another one. It may be last machine in the chain, but it may also be only a link. The tablets might be catapulted across five machines before they reach Kushim’s banking application. At each stage, the reader reproduces what information it needs from the tablet it receives to press its new tablet; the original tablet is not sent along with the new tablet. Thus, an elaborate game of telephone is born. Critically, no human (by design!) needs to be involved in this communication. So, the machines only prioritize the fields that ensure each re-pressing of a new tablet is accurate. Any other data can be let go.
Obviously, I’ve misled the Sumerian reader significantly in my technical description of our system, but I think the spirit is right. The only additional callout is that, while we use these machines, we don’t entirely trust them. That is, perhaps, the sole reason that the transaction description still exists at all: In the event that we mistrust the accuracy of the ledger, an auditor will want to know what exactly the transactions were for.
To summarize: Sumerians made accounting systems for Sumerians; Modern people, contrarily, make accounting systems for machines.
Machines and Sumerians have one major thing in common: They write in systems that we cannot easily read. The Sumerians have been extinct for eight millennia. The death of their language, writing system, government, and the material wear on their records are all pretty good excuses for their modern inaccessibility. What excuse do the machines have?
Modern banking needs to leverage modern technology, no doubt. But have we over-prioritized technical literacy over accountancy? The need for transaction cleansing from companies like Yodlee would suggest yes. The finance industry writes billions of transactions every day—knowing perfectly well at every step along the way what the transactions are—and losing that information at each stage, voluntarily. The data just… drops out of the transaction packet. Eventually, the severely curtailed transaction makes its way to Yodlee for cleanup and FinGoal for transaction tagging.
It’s not really an exaggeration to compare modern transaction data to fragmentary cuneiform records. We’re missing shards. The information is dropped from the record. When we receive a transaction from a banking core, it is usually full of holes. We get an original description like 000030304045958 20250116 WINOT COFF. Sometimes, these transactions are cut short because of software limitations in fifty-year-old systems—literally, our bulla isn’t big enough to hold all of the tokens we need, and we refuse to invent a bigger bulla (this is slightly unfair; the real issue is that millions of distributed and non-standard machines and services would also need to be updated to use that bigger bulla. Some of those machines are written in humanity’s new and more dangerous version of the dead language, COBOL).
Yodlee scrapes out the fluff and leaves us with WINOT COFF. Obviously, the string is truncated; a shard is missing. Yodlee makes an assumption (in this case, a safe one, though it’s not always so clear cut) and extends the description to WINOT COFFEE. From there, FinGoal looks at Winot Coffee to try to categorize it, to characterize the behavior, to guess what might have been purchased and why. In short, the system works to deduce properties of the transaction that were intentionally lost—lost, in some cases, literally overnight.
This kind of data loss is bad across the board; it has negative effects that ripple across the entire financial industry:
You’re asking: Do I have a solution for this? Kind of! My team of engineers, and the great data engineers at Yodlee, work hard to act as expert archaeologists, to get the proverbial mud off these transactions just as soon as they’re dropped. The advent of open banking, with Yodlee supporting a growing list of open-banking enabled sites, may also prevent the loss of the information to begin with. Both of these initiatives are valuable and helpful. However, the ancients are often more radical than we are. I think my Sumerian friends would offer a better and more lasting solution, one which (for some reason) seems impossible to a modern audience: Just make a bigger bulla.
.png)
.png)
