Ontology Is the Idea Finance Has Been Missing

There is a number that reframes almost everything about how we work, invest and govern, and hardly anyone has a feel for it. In 2025 the total amount of data created, captured and copied in the world reached roughly 181 zettabytes, on the widely cited estimates from the research firm IDC. A zettabyte is a trillion gigabytes. Written out, the figure is meaningless, which is precisely the point. No human being has any intuition for it, because no human being was ever built to hold it.

Artificial intelligence is the reason the number keeps climbing, and it climbs from both ends. AI consumes enormous quantities of existing information to train, and then it produces more of its own, running day and night, never tired, never asking for the weekend off. A single large bank now generates more internal information in a week than its analysts could read in a career. The gap between what is recorded and what is understood is not closing. It is widening every single day.

I have come to believe this is the defining problem of the decade, and that the answer to it is an idea most people have never heard of. The idea is called ontology, and I want to explain in plain language why it sits at the centre of everything we build at KXCO, and why I think it is the most important and most underrated development in finance, banking and regulation today. I have written the full technical version of this argument for engineers and analysts on the KXCO blog, in an essay called Mapping Data So People Can Understand It. This is the shorter version, written for everyone.

Three machines and a human being

Think of the modern flow of knowledge as a chain with three links.

The first link is artificial intelligence, and its job is to produce material. AI reads everything, watches everything, transcribes everything, and turns the unstructured mess of the world into signal at a scale no army of analysts could ever match. It is the supply side of knowledge, and it is effectively infinite now.

The second link is processing, and this is where quantum computing is beginning to matter. I want to be careful here, because quantum is the most over-hyped word in technology and I have no wish to add to the noise. Quantum computers are not magic and they are not general replacements for the machines we use today. What they are is extraordinarily fast at a specific and important class of problems, the kind of search, optimisation and pattern-finding that sits underneath making sense of a complicated web of relationships. In December 2024 Google showed the point vividly when its Willow chip ran a benchmark in about five minutes that the company estimated would take one of the fastest classical supercomputers on the order of ten septillion years to finish. That is a length of time so far beyond the age of the universe that the comparison stops meaning anything.

So the first link makes the data effectively unlimited, and the second link is making the processing of hard, structured problems effectively instant. Two of the three classical bottlenecks are falling away. What is left standing, untouched by any chip, is the third link. A person still has to look at the result and understand it. A quantum computer can find the pattern in a septillionth of the time, and it will still be useless to a bank, a regulator or an investor unless a human being can see the pattern, trust it, and decide what to do about it.

The whole game, in other words, comes down to the last link. And the last link is a person.

Understanding is a picture

Here is the belief that shapes how we design everything at KXCO. I stated it in the technical essay, and I will repeat it here because it is the heart of the matter.

Imagery is the primary form of understanding. All language, written or spoken, is subtext.

I do not mean that as a flourish. I mean it close to literally. When you truly understand something, you can see it. You hold a mental picture, a shape, a model in the mind's eye, and the words you use to describe it come afterwards, as a way of pointing at the picture you already hold. Language is how we pass understanding to one another, and it is a magnificent tool, but it is a slow and lossy channel laid over the top of something faster and older. Ask any great investor, any engineer, any chess player how they actually think, and underneath the vocabulary you will nearly always find a picture.

The science has been pointing this way for half a century. In 2014 the psychologist Mary Potter and her colleagues at MIT flashed images at people at speeds down to thirteen milliseconds per picture, faster than the eye can consciously settle, and found that people could still grasp the meaning of a target image at that speed. Thirteen milliseconds is about a hundredth of a blink. The brain was pulling meaning out of a picture in less time than it takes to become consciously aware of seeing it. There is no equivalent feat for text. You cannot read a sentence in thirteen milliseconds. The visual system is the fast path to comprehension, running far ahead of the verbal one.

That is not an accident. Vision is the single largest consumer of resources in the brain. Neuroscientists commonly estimate that around a third of the cerebral cortex is involved in visual processing, far more than hearing or touch. We are, physically, visual animals first. When you show a person a picture, you are speaking to the largest and fastest part of their machinery for understanding. When you hand them a page of dense text or a table of numbers, you are routing everything through a narrow and comparatively recent channel and asking the visual system to sit the round out.

The oldest and most important study for my purposes is about reasoning rather than memory. In 1987 Jill Larkin and the Nobel laureate Herbert Simon published a paper with a title that says almost everything, "Why a Diagram is (Sometimes) Worth Ten Thousand Words". Their argument was rigorous rather than poetic. They showed that a diagram and a written description can contain exactly the same information and still not be equally useful, because the diagram groups related facts together by their position in space. When the things you need to reason about are placed next to each other, you no longer have to hunt for them or hold them all in your head at once. The diagram does part of the thinking for you. The text, however complete, forces you to do all of that work yourself.

This connects to one more classic finding. In 1956 George Miller published his famous paper on the limits of working memory, the origin of the idea that a person can hold only around seven things in mind at once. The exact number matters less than the shape of the limit. Our capacity to juggle separate facts is tiny and fixed, and no amount of extra data changes it. A picture is powerful precisely because it lets us escape that limit. Instead of holding a hundred facts in a seven-slot memory, we hold one picture and read the facts off it as we need them. The picture becomes external memory, and reasoning that was impossible in the head becomes easy on the page.

A list is not a map

Let me make this concrete, because the gap between a list and a map is the entire argument in miniature.

Take a handful of ordinary facts about the technology economy, the kind you could read in any business paper. Nvidia invests in OpenAI. OpenAI buys enormous quantities of computing power from Oracle. Oracle buys its chips from Nvidia. Nvidia also invests in Anthropic. Anthropic buys its computing power from Microsoft. Microsoft buys its chips from Nvidia. Written as a list, each fact is clear, each is true, and the list is complete. Read it and you learn a set of individual relationships. What you almost certainly do not see, because the format hides it, is that the money is moving in a circle. Capital flows out from the chipmaker into its customers, and a large part of it flows straight back in the form of chip purchases. The list contains that loop. It does not show it.

Now draw the same facts as a map, with each company a point and each relationship a line, and the loop is not just visible, it is impossible to miss. Your eye finds the closed circle before you have consciously read a single label, because closed shapes are exactly the kind of thing the visual system is tuned to catch instantly. The fact that took a paragraph of careful reading to suspect is now understood in a fraction of a second. Nothing was added. The information is identical. Only the form changed, and the change in form is the difference between data you happen to possess and insight you actually have.

This is what Larkin and Simon meant, made physical, and it generalises far beyond a story about chip companies. Every important system that human beings need to understand and cannot easily see is a system of relationships. A supply chain. A corporate ownership structure. A web of financial exposures. A network of transactions. The flow of a legal obligation from one party to another. In every one of them the facts are usually available and the pattern is usually hidden, for the simple reason that we store them as lists and rows rather than as maps. The information is not missing. The picture is.

What an ontology actually is

The word sounds academic, and it comes from philosophy, where ontology is the study of what exists and how things relate. In the world of data it means something concrete and useful. An ontology is a formal map of the things in a domain and the relationships between them. Instead of storing rows in a table, it stores facts as small, complete statements: a subject, a relationship and an object. Nvidia invests in OpenAI. Because the relationships are treated as first-class citizens rather than buried in a database, the whole thing is a graph by nature, and a graph is a picture waiting to be drawn.

At KXCO we add one discipline to this that I consider non-negotiable, and it is the part that makes an ontology trustworthy rather than merely tidy. Every claim in the model carries its provenance. It is not enough to assert that Nvidia invests in OpenAI. The claim has to travel with three things attached: a source, so you can see where it came from, an as-of date, because a fact that was true last quarter may be false today, and a confidence level, because some things are documented in a filing and others are reported second-hand, and an honest model does not pretend those are the same. Where a fact is genuinely unknown, the model says so plainly rather than guessing.

That discipline is what turns a pretty graph into infrastructure you could build a bank on. When every line in the picture is a sourced, dated, confidence-scored claim, the picture stops being an artist's impression and becomes an auditable model of reality. You can click any relationship and see exactly why it is there and how much to trust it. And because the whole thing is readable by machines as well as people, an artificial intelligence can reason over the very same structure a human is looking at, which means the person and the machine are finally working from one shared, inspectable picture of the world rather than two divergent ones.

Why finance needs this most

I have spent my career in and around markets, and I can tell you that finance is not, at bottom, a business of numbers. It is a business of relationships, and the numbers are simply the current readings on those relationships. Who owns what. Who owes whom, and by when. Which exposures move together and which move apart. Which counterparty sits quietly behind three others you thought were independent. Every one of these is a relationship, and the entire discipline of risk management is really the discipline of understanding a web of them well enough to survive when it moves.

This is exactly the kind of understanding that lists and tables destroy and that maps preserve. A bank's risk system can hold every position, every counterparty and every exposure in a database and still be blind to the one structural fact that matters, because that fact is not in any single row. It lives in the shape of the connections between rows. The most expensive lessons in modern financial history have been, over and over again, failures to see a structure that was present in the data all along.

The global financial crisis of 2008 is the case that should be taught in every data class, not just every finance class. The information needed to understand the danger was, to a very large extent, already recorded somewhere. Institutions held the mortgages, the securities built on them, the derivatives built on those, and the web of counterparty exposures that tied everyone together. What almost nobody had was a picture of how it all connected. Firms could not see their own total exposure to a failing counterparty across all their desks and subsidiaries, let alone their exposure to the counterparties of their counterparties. When one large institution wobbled, no one could tell in time who else would fall, because the relationships were scattered across thousands of systems as disconnected rows. The facts were present. The map was absent, and the absence of the map cost the world somewhere in the trillions.

The regulatory response to that failure is, when you look closely, an attempt to force finance to build the map. In 2013 the Basel Committee published a set of principles known as BCBS 239, on effective risk data aggregation and risk reporting. Strip away the language and what it demands is that a large bank be able to pull its risk data together across every silo, quickly and completely, so that senior people can actually see the institution's total risk as a single coherent view. That is a request for an ontology in all but name. More than a decade on, many banks still struggle with it, precisely because they are trying to satisfy a demand for a map using technology built for rows.

There is a quieter revolution that makes the whole thing newly possible. The world's payment and messaging systems are migrating to a standard called ISO 20022, a structured, meaning-rich format for financial messages that is replacing the terse and ambiguous formats banks have used for decades. The significance is that money is starting to move with proper, machine-readable context attached: who, to whom, for what, under which obligation. Structured messages are the raw material an ontology feeds on. As the plumbing becomes semantic, the chance to build a live, connected, comprehensible model of financial reality moves from theoretically appealing to practically achievable. We have designed KXCO for this world on purpose, mapping our model to the standards that regulated finance is converging on.

There is a whole category of risk that exists specifically to defeat the list-and-table view of the world, and it is where an ontology earns its keep most dramatically. Money laundering, sanctions evasion and the concealment of who really owns a company all work by exploiting the fact that institutions store reality as disconnected records. A network built to move illicit money is designed so that no single account, no single filing and no single jurisdiction shows anything wrong. The wrongness lives entirely in the relationships between them. Ask an anti-money-laundering team what they are really fighting and it is this. They are trying to see a structure that someone has spent real effort making invisible. An ontology is close to the ideal instrument for that fight, because it makes the structure the primary object. A shell company that looks innocent as a row becomes obvious as a hub with too many faint connections. A hidden owner behind three layers of holding companies becomes a node you can trace to with your eye.

Why the regulator needs it even more

If ontology matters to a bank, it matters even more to the people who supervise banks, and this is the part I feel most strongly about. A regulator's entire job is to understand a system too large and too interconnected for any individual to hold in their head, and to understand it well enough to spot danger before it becomes a crisis. That is, almost word for word, the problem this whole piece is about.

And yet supervisors are handed the worst possible tools for the task. They receive vast quantities of data from the firms they oversee, delivered as filings, returns and reports, which is to say as lists and tables and documents. They are asked to detect systemic patterns, contagion paths and hidden concentrations, which are relationships, using formats that hide relationships by design. We ask our regulators to see the shape of the financial system, and then we send them the information in the one form guaranteed to obscure the shape. It is no wonder that crises are so often visible only in hindsight, when someone finally draws the map.

This is why I argue that ontology is not merely a financial technology but a regulatory one, and potentially the most important development in supervision since the move to electronic reporting. A regulation is, at heart, a statement about relationships that must or must not exist. This entity may not be exposed to that one beyond a limit. This owner must be disclosed. Enforcing such a rule requires, first and above all, the ability to see the relationship the rule is about. A rule the regulator cannot picture is a rule the regulator cannot enforce, no matter how well it is written. Ontology is the layer that makes rules legible in reality rather than only on paper.

The practical difference is not abstract. Picture two ways the same question gets answered inside a large institution. In the first, a risk officer asks whether the firm is dangerously exposed to a single stressed counterparty, waits days while three teams reconcile four systems, and receives a spreadsheet that answers a slightly different question than the one that was asked. In the second, the officer opens a live map of the firm's exposures, sees at a glance that a cluster of positions all trace back to the same counterparty, clicks a line to check the source and the confidence behind it, and makes the call in minutes with the evidence in front of them. The first is where most of the industry still lives. The second is what becomes possible when you treat the relationships as the primary thing and show them as a picture. As activity explodes under AI, and as automated agents begin to transact at machine speed, the distance between what supervisors are responsible for and what they can actually see will become the defining risk in the system. A picture a supervisor can read in real time is the only serious answer I know of.

None of this requires KXCO to be a regulated institution, and we are not one. We are a software company. What we provide is the infrastructure, the semantic layer and the visual engine that lets the institutions and, in time, the supervisors who do carry those responsibilities finally see the systems they are accountable for.

See it working

I dislike arguments that stay abstract, so we built this one into something you can open right now. The KXCO Ontology Engine is a live, public model of the artificial intelligence economy. It maps roughly a hundred companies, people, chips and capital flows as typed, sourced claims you can click on and inspect, and it renders the whole thing as a map you can explore and question. Open it and ask the questions built into it. Who depends on whom. Which company invests in its own customers. How much capital moves in a circle. Follow the supply chain down and watch it narrow, through the chipmakers, to the handful of firms, and ultimately the single Dutch company, that the entire edifice rests on. Every answer is a claim you can inspect, with its source and its confidence attached. Where a fact is undisclosed, the engine tells you so rather than inventing a number.

We chose the AI economy as the public demonstration because it is the most consequential and least understood web of relationships in the world today. The same method, applied privately, points at whatever a particular institution needs to understand: its own exposures, its supply chain, its ownership structures, its market.

One idea, running through everything

People sometimes ask why a company with a signing product, a treasury product, a security product and a blockchain also talks so much about ontology, as if it were a side project. It is the opposite of a side project. The ontology is the spine, and each product is a different organ hanging off the same nervous system.

KXCO Nexus is where identity, legal execution and proof live. Every signature and every verified identity is a statement about relationships between parties, written into the ontology as a claim, which is what lets a signature become more than an image. KXCO Treasury, our economic operating system, turns a pile of balances into a live map of where value sits and how it moves. KXCO Sentinel, our quantum-resistant cloud, protects the integrity of the model itself, because a beautiful map of reality is worthless if it can be quietly forged. Armature L1, our settlement layer and public record, anchors the most important claims so that they are not just asserted but provable to anyone, permanently. And Meridian, our private-capital deal network, treats a deal as a structure you can see rather than a folder of files you have to reconstruct. Five products, one idea, made five times: take a domain that institutions currently experience as a pile of disconnected records and turn it into a connected, inspectable picture of reality.

That picture is the design principle of KXCO stated at its simplest. We do not build for the machine, because the machine is no longer the bottleneck. We build for the person at the end of the chain, whose comprehension is the scarce resource that everything else exists to serve. Artificial intelligence fills the model with signal. Quantum-grade processing and our cryptography keep it fast and trustworthy. The ontology organises it into relationships. And then, at the last step, the one all the others are for, we render it as a picture a human being can understand at a glance, question with a click, and act on with confidence.

Imagery is the primary form of understanding, and everything upstream of the picture is subtext. It is extraordinary and necessary subtext, but subtext all the same. If you want the full engineering account of how we build it, the discipline underneath the claims, and the evidence behind every study cited here, it is all in the companion essay on the KXCO blog, Mapping Data So People Can Understand It. And if you would rather simply see the idea in motion, open the live ontology and start asking it questions.

Dr. Shayne Heffernan is the founder of KXCO. KXCO is a software company that builds the reality infrastructure for the human and AI economy. Nothing in this article is investment or regulatory advice.