What Seven Years Inside a Financial System Taught Me About Money, Power, and the Limits of Technology
By Heorhi Tratsiak
I have been trying to write this chapter for three months.
Not because I lacked material. The preceding seven chapters contain more material than I originally thought I possessed — more precision, more accumulated observation, more of the texture of what it means to work inside a system that is simultaneously the most sophisticated engine of human cooperation in history and the most powerful instrument of coercion ever built without a single soldier.
I struggled with this chapter because I understood, before I began it, what it needed to do. And what it needed to do was harder than analysis. Analysis is a form of distance. You describe the mechanism. You name the concept. You place the event in its historical context. The description is accurate, the naming is useful, the context is illuminating. And the distance that makes all of this possible also makes it possible to complete the work without being fully changed by it.
This chapter cannot be written from that distance. This chapter is the one where I have to account for what I actually know — not just what I observed, not just what I analyzed, but what I am responsible for having understood. What that understanding obligates. And what it means that I continued, for a period after understanding it, to sit at my terminal and process the transactions.
I do not write this in self-accusation. I write it because the moral question of this book is not the question I thought it was when I started writing it.
I thought the question was: how does the financial system work, and why has no one explained it from the inside?
The actual question, which the writing of the preceding chapters forced me toward, is different. It is: given that the system works the way it works — given everything described in these pages — what are we responsible for? What does it mean to know this? What does the knowledge oblige?
I don’t have comfortable answers. I have the answers I have, which are incomplete and which I offer not as conclusions but as the most honest account I can give of what seven years inside a financial system left me with.
What the Machine Cannot Know
Somewhere in the history of financial services, a decision was made that has never been fully examined. The decision was to treat the automation of financial processes as equivalent to the improvement of financial processes. To assume that a system that processes transactions faster, cheaper, and with fewer human interventions is, by those properties alone, a better system.
I want to challenge this assumption, not from sentiment, but from observation.
Over the course of seven years, I processed thousands of transactions and made thousands of credit-related assessments. In the last years of that period, I watched machine learning systems begin to do what I had been doing — and in many respects, do it faster and more consistently. The pattern recognition was genuinely better in certain domains. The processing speed was not comparable. The scale at which the systems could operate was orders of magnitude beyond any human capacity.
What the systems could not do — what no system I encountered was designed to do — was notice what it was processing.
This sounds like a philosophical observation. I mean it as an operational one.
When I processed a wire transfer for a woman trying to send money to her daughter studying abroad, I knew, in a way that did not appear in any field of the transaction record, what I was processing. I was processing a mother’s care. A student’s rent. The material form of a relationship that existed in a context I could partially imagine, which made the outcome of the transaction — whether it processed correctly, whether it cleared on time — something that carried moral weight for me.
The machine processed the same transaction as a data object. Name. Account number. Amount. SWIFT code. Compliance flag status. Approved or declined. The machine’s indifference to the human content of the transaction was not a failure. It was a design specification. The machine was designed to process the data object, not the care embedded in it.
This is fine when the machine is making routine decisions in a stable environment. The efficiency gains are real. The consistency is genuinely valuable. The elimination of certain human biases — the loan officer who unconsciously favors borrowers who remind her of herself, the compliance officer who applies rules more stringently to names that sound foreign — is a genuine improvement.
What happens when the environment becomes unstable, when the policy context changes, when the machine is suddenly processing data objects that represent, in the real world, a completely different kind of situation than the one for which it was trained, is a different question.
In the weeks following February 2022, automated compliance systems processed transactions the same way they had processed identical transactions the week before — because the data objects were identical. The compliance flags had not yet been updated. The risk classifications had not yet been revised. The transactions that should have been flagged for human review were processed, or declined, on the basis of parameters that were already obsolete.
This is not a failure of AI. It is a demonstration of its nature. AI systems optimize for the parameters they are given. When the world changes in a way that makes the existing parameters no longer adequate, the system continues to optimize for parameters that are no longer adequate, until a human intervenes to revise them.
The human who intervenes to revise the parameters is making a judgment. About what the situation now requires. About what values the system should serve. About what outcomes are acceptable and what outcomes are not. This judgment cannot be automated, not because the technology is insufficiently advanced, but because it is a moral judgment, and moral judgments require a subject — someone who can be held responsible for them.
The most dangerous development in financial technology is not the replacement of human labor with automated systems. That replacement is real and significant, but its consequences are primarily economic. The most dangerous development is the disappearance of the human subject who bears responsibility for outcomes. When no one is deciding — when the system is deciding — then when the system produces a wrong outcome, there is no one to hold accountable. The outcome exists without an author. The harm is real, but it belongs to no one.
I wrote in the fifth chapter about acceptable costs. The phrase comes from policy analysis, where it describes the harm to individuals that is judged tolerable in pursuit of a policy goal that is judged to produce greater benefit. I want to return to it here, because it is the phrase that most precisely names what I believe is the central moral problem of modern financial infrastructure.
The concept of acceptable costs requires someone who accepts them. A decision-maker who can be identified, who can be asked to justify the acceptance, who can be held accountable if the acceptance was wrong. In the distributed systems of global finance — in the compliance cascade I described in Chapter Five, in the institutional veto I described in Chapter Six — this accountability has been so thoroughly distributed that it has effectively ceased to exist.
Nobody accepted Alina’s mother as an acceptable cost. Nobody made that decision. It was the aggregate output of a thousand decisions in which she did not appear as a variable. The outcome was as complete as if someone had decided it and as unaccountable as if it had happened by natural law.
This is not a problem that more sophisticated AI will solve. It is a problem that more sophisticated AI will accelerate, because the automation of financial decision-making at scale is the automation of consequence without accountability. It is the construction of systems in which millions of human outcomes are produced by processes that have no subject, no judgment, and no possibility of being asked: did you mean for this to happen?
Seven Years: What Each Chapter Actually Cost
I want to move through what this book has been, from the inside.
Chapter One was written with the pleasure of someone explaining something they understand well. The architecture of global payments is genuinely elegant — a system that evolved over decades into something that works, that coordinates the financial activity of billions of people, that turns the abstract obligation of a debt into the concrete movement of value across borders and institutions. Describing it precisely, in language accessible to someone who has never seen a SWIFT terminal, felt like an act of tribute. I admired the system when I worked in it. That admiration was not naive. It was the admiration of someone who understood what it took to build something that reliable at that scale.
Chapter Two cost me the memory of faces. The trust layer is not a metaphor. It is the accumulated weight of every person who sat across a desk from me, or whose file crossed my terminal, and whose life circumstances I was required to translate into a risk assessment. Some of them I helped. Some I couldn’t help, because the risk was real and the alternative was unfairness to the institution and ultimately to other clients. Some I should have helped and didn’t, because I was tired, or because the system made the cautious choice the defensible one, or because I was applying a policy that was wrong in their specific case. I do not know, for most of them, what happened. The system does not provide that information. I made a decision and the file moved on and I processed the next file. They remained people, somewhere, whose financial lives continued to be shaped by decisions I had made and forgotten.
Chapter Three required me to hold two things simultaneously: the professional pride of having been inside an institution that held together under maximum pressure, and the knowledge that the crisis that tested the institution was a human catastrophe that extended far beyond the balance sheet. I described the operational resilience with accuracy and with genuine respect for the people who maintained it. What I could not fully describe was what it felt like to understand, in real time, that the events producing the crisis had destroyed the lives of people who had nothing to do with the financial system — that the operational story I was equipped to tell was the narrow edge of a much larger human story I was not equipped to tell. I told the story I had. It was true. It was not complete.
Chapter Four was the chapter of historical reckoning. The thirty-year arc from neutral infrastructure to geopolitical weapon. Writing it required me to trace, in reverse, the history of decisions that produced the system I had been trained to operate as if it were simply technical. The decisions were political. The people who made them were pursuing interests they believed were legitimate. I did not resolve that question. I described the mechanism and named the progression. The moral weight of the history is real whether or not I adjudicate it.
Chapter Five was the hardest to write. Not because the material was difficult to recall — it was vivid, more vivid than I would have preferred — but because writing it required me to be honest about something I had avoided being honest about while it was happening. While I was processing the questions from correspondent banks, while I was telling clients that their transfers were experiencing delays due to the compliance environment, I knew more than I said. I knew the delays were not temporary. I knew the trajectory. I knew that the language I was using — accurate, institutional, defensible — was also incomplete, and that the incompleteness was causing harm to the people I was using it with.
I did not lie to them. But I did not tell them what I knew. And I told myself, during that period, that this was professionalism. That there were things I could not disclose for legal and institutional reasons. That the system required this and the system served a larger purpose and individual disclosures would create more problems than they solved.
All of that was true. And it was also a way of not having to live, day by day, with the full weight of what I knew.
Chapter Six was the chapter of failure — the three products that were killed by no decision and the mechanism I built from the pattern of their deaths. Writing it required me to be precise about something I had, for several years, allowed to remain imprecise. The institutional veto is a real mechanism. It is not bad luck or bad timing or insufficient product quality. It is the operation of a system structured to prevent what I was trying to build. Understanding that clearly was, in some ways, a relief. It took away the comfort of self-blame. But it also took away the comfort of hope — the possibility that the next iteration would be different, that the right partnership or the right timing would produce the breakthrough.
Chapter Seven was the first chapter I wrote with something approaching hope. The alternatives are real. The change is happening. The weapon is losing its monopoly. These are not consoling fictions. They are accurate assessments of a transition that is underway. But writing that chapter from inside this one, I am aware of what the hope costs: it is hope that arrives too late for Alina’s mother, too late for the businessman whose relationship with his German supplier did not survive eleven days of compliance review, too late for everyone who needed the money to move in 2022 and could not find a workaround in time.
The future is better than the present. This is true. It is also true that the people for whom the present is the only thing that matters have no use for a better future.
Who Is Responsible
I want to try to answer the question directly, because I have circled it for eight chapters without landing.
The sanctions regimes that produce compliance cascades are designed by policymakers who are pursuing goals — geopolitical goals, human rights goals, security goals — that they believe are legitimate. In many cases they are legitimate. The policy apparatus that produces them includes lawyers, economists, national security specialists, and in some cases genuine moral deliberation about proportionality and targeting.
None of these people are responsible for Alina’s mother.
The correspondent banks that update their risk classifications and impose additional documentation requirements and slow their processing times are operating rational compliance functions. Their decisions are legally defensible. In most cases they are legally required. They have no way to know, from the data available to them, that one of the transactions in their review queue represents a daughter’s care for a mother across a closed border.
None of these institutions are responsible for Alina’s mother.
The card networks that suspended services to affected issuers were executing decisions made at the policy level, using infrastructure built for exactly this purpose, on a timeline that was operationally impressive and commercially ruinous to no one except the people whose cards stopped working.
They are not responsible for Alina’s mother.
Alina’s mother is not responsible for Alina’s mother. She was a sixty-seven-year-old woman whose rent came due on the twenty-eighth of the month and whose daughter had sent the money as she always had.
The architecture of distributed consequence — the same architecture that makes the compliance cascade so effective as a coercive instrument — is also the architecture of distributed responsibility. When everyone is following their rules, and the rules are collectively producing harm, and no single actor can be identified as the author of that harm, what we have built is a system of consequence without accountability. A system in which human suffering can be produced at scale, reliably, predictably, and without anyone being responsible for producing it.
I am not going to tell you that this is unique to financial systems. It is not. It is the defining moral problem of complex modern institutions in every domain. The hospital patient who dies because every individual department followed its protocol correctly. The civilian killed because every node in the targeting chain made a locally defensible decision. The climate that changes because every individual actor acted rationally in their own interest.
The specific claim of this book is that the financial system — because it is the infrastructure through which the material conditions of human life are mediated, because it reaches into every corner of every economy, because it has been deliberately engineered into the most powerful non-military coercive instrument in history — is the domain where this problem is most acute and least examined.
Every analysis of financial sanctions focuses on the macroeconomic impact. The GDP contraction, the inflation rate, the reserves drawdown, the currency depreciation. These are real and they matter. What is almost never examined is the specific human experience of each transaction that didn’t clear. Not as a statistic, not as a coefficient in a macroeconomic model, but as the concrete moment when someone who needed money to move discovered that it could not move, and tried to understand why, and could not get an answer that named what had actually happened.
This book has tried to name what actually happens. At the level of the SWIFT terminal and the correspondent bank queue and the compliance cascade and the card network suspension and the institutional veto and the woman standing at the kitchen window waiting for a notification that does not come.
Naming it does not fix it. But the failure to name it — the comfortable consensus that financial infrastructure is technical and therefore morally neutral, that policy is made at levels too abstract for individual human stories to be legible, that the system is too complex for any individual to bear responsibility for its outcomes — is the precondition for the harm continuing unexamined.
The Reader’s Position
You are holding this book. Which means something about you.
You have a bank account. Almost certainly more than one. You use a payment card — Visa or Mastercard, almost certainly — and you use it without thinking about the certification processes that make it work, or the correspondent relationships that underpin the clearing, or the geopolitical decisions that determine which cards work in which countries. The infrastructure described in the first chapter of this book is, for you, invisible in the way that all truly successful infrastructure is invisible: you notice it only when it fails.
For most of your life, it has not failed.
This is a form of privilege that is almost impossible to perceive from inside it. The smoothness with which money moves for you — the tap of a card, the transfer received, the international payment cleared — is not a property of the universe. It is a property of your location within a system that has been built, maintained, and in certain significant respects weaponized in ways that benefit people in your position at the expense of people in other positions.
This is not an accusation. It is a description. You did not design the system. You did not decide to weaponize it. You are, like almost everyone, living in a structure you did not build and cannot individually dismantle.
But you now know how it works. You know what correspondent banking is and how the compliance cascade propagates through it. You know what the card network suspension looks like from the inside of a bank and from the inside of an airport terminal. You know about the institutional veto and about the structural trap that prevents the countries most harmed by the system from building alternatives to it. You know that the alternatives are being built anyway, by people whose need was more urgent than their access to institutional support.
Knowledge creates a different relationship to complicity. Before you knew, your participation in the system was unreflective — the participation of someone who uses the water without considering its source. After you know, the participation is chosen. You continue to use a system whose coercive capacity you now understand. You continue to benefit from its smooth operation while knowing that its smooth operation for you is not separable from its coercive operation against others.
What does that oblige?
I am not going to pretend that I have a clean answer. There is no individual action that dissolves the structural conditions this book has described. Closing your bank account does not reform correspondent banking. Refusing to use your Visa card does not change card network policy. Reading a book — even this book — does not make anyone’s blocked transfer clear.
What I can say is this: the first obligation of knowledge is not action. The first obligation of knowledge is honesty. The refusal to go back to not knowing. The refusal to participate in the comfortable consensus that the financial system is technical and therefore neutral and therefore not a moral domain.
It is a moral domain. Every transaction in it carries human meaning. Every policy decision that affects it produces consequences for specific people who have names, who have mothers, who have rent due on the twenty-eighth of the month. The abstraction of those people into acceptable costs is a choice — made or not made, by each person who has the knowledge to make it differently.
You have that knowledge now.
What Remains Human
I want to end this chapter — and this book — with what I believe is the most important thing I learned from seven years inside a financial system, which is not about SWIFT message formats or correspondent bank relationships or compliance cascades.
It is about what cannot be automated.
Not because the technology is insufficient, but because the thing itself is not a process. It is a capacity that belongs specifically to beings who can be harmed, and who therefore understand, from the inside, what harm is.
The capacity I mean is this: the refusal to treat a human being as an acceptable cost.
This refusal cannot be written into an algorithm. Algorithms optimize for objectives. If the objective includes minimizing harm, the algorithm will minimize harm to the degree that harm is measurable within the objective function. But the harm that matters most — the harm of a sixty-seven-year-old woman who has to borrow from a neighbour for the first time since her husband died because a risk classification she has never heard of made her daughter’s transfer impossible — is not easily measurable. It does not appear in a compliance log. It does not generate a flag in a risk system. It is a form of harm that requires a human being to notice it, because it requires a human being to understand that dignity is a thing that can be damaged, and that its damage is not an acceptable cost.
The financial systems being built now — the CBDC infrastructure, the DeFi protocols, the stablecoin rails, the upgraded hawala networks described in the preceding chapter — are better, in many respects, than what they are replacing. They are faster. They are cheaper. They are more accessible. In certain important respects, they are less susceptible to the specific weaponization I have described.
They are not, in themselves, moral improvements. A blockchain is not more ethical than a correspondent bank. A smart contract does not have a conscience. The new architecture solves the technical problem of centralized control. It does not solve the moral problem of what happens when efficient systems process human beings as data objects and produce harm that is real, distributed, and unattributable.
That problem is not solved by any technology. It is solved, to whatever degree it is solved, by the people who operate inside financial systems maintaining the understanding that what they are processing is not data. It is people’s lives, in material form. It is a mother’s care for a daughter, transmitted as a SWIFT message. It is a student’s ability to continue her education, denominated in euros. It is an old woman’s dignity, measured in a rent payment that arrives on time.
The human variable is not a limitation of financial systems that better technology will eventually remove. The human variable is the only thing that stands between a financial system and the full, efficient, automated expression of its capacity to treat human beings as acceptable costs.
I was the human variable, for seven years. I processed the transactions with the knowledge of what I was processing. I made the credit assessments with the understanding that I was deciding about real people’s access to real things. I stayed in the conversations with clients longer than the procedure required, sometimes, because the procedure was ending the conversation before the person had been heard.
This is not heroism. It is the minimum that human participation in a system processing human lives should look like. It is the minimum, and it is not guaranteed, and it is not automated, and it is not preserved by any technology. It is preserved only by people who choose to preserve it.
I do not know whether I preserved it adequately. I know that there were times I did not. I know that the institutional pressures I have described in this book — the compliance culture, the distributed accountability, the structural incentives for inaction over engagement — made it progressively harder over time to maintain. I know that when I left the institution, one of the things I was leaving was the daily experience of those pressures and what they were doing to my capacity to notice the people on the other side of the terminal.
I carry that. I carry it as information about what systems do to the people inside them over time, which seems to me to be information worth having.
The Final Account
This is the account I promised in the beginning: the inside view. The operational reality. The things that the academic literature and the financial journalism and the policy analysis do not contain, because none of those genres are written from inside a SWIFT terminal by someone processing the transactions while the system around them changes into something they were not trained for.
I have tried to be precise. I have tried to be honest. I have tried not to claim more authority than my position justifies, and not to claim less authority than my position earned.
What I know, I know from the inside. The architecture of the system. The texture of its normal operation and the different texture of its operation under stress. The mechanism of the compliance cascade. The institutional veto. The human cost of each. The alternatives being built by people who have more to lose from the system’s coercive capacity than I did.
What I believe, I believe from having lived inside this and having looked at it clearly enough, I hope, to describe it without the comfortable distance of abstraction.
I believe that the financial system is one of the greatest achievements of human coordination in history, and that its weaponization is one of the most significant moral developments of the twenty-first century, and that both of these things are true simultaneously and cannot be separated.
I believe that the alternatives being built will matter, and that they will arrive too late for some of the people who needed them now, and that this too is true simultaneously with the hope their existence provides.
I believe that no infrastructure, however powerful, can permanently contain the human need to transact, to trust, to build relationships across borders and across political divisions. The evidence for this is historical and it is recent and it is the forty seconds it took for a payment to arrive through a system that did not exist fifteen years ago.
And I believe that none of this — not the analysis, not the history, not the alternatives, not the hope — removes the obligation to see the person in the transaction.
Her name, as I have written, is not something I know. She is a sixty-seven-year-old woman in Minsk. She has a kitchen window. She stands there on the twenty-eighth of the month and waits for the notification that tells her that her daughter thought of her, that the care transmitted itself across the distance between Berlin and Minsk in the material form that distances require.
When the notification came, she did not think about SWIFT. She did not think about correspondent banks or compliance cascades or the geopolitical decisions that determined whether the infrastructure her daughter’s care had to pass through was open or closed that month. She thought about her daughter. She thought about lunch. She did what people do when the system works: she forgot the system existed.
When the notification did not come — when the system had been used, at levels of abstraction she could not have conceived of, to prevent the movement of her daughter’s care — she noticed the system the way everyone notices it when it fails. Not with understanding. With the specific helplessness of someone who trusted a thing that turned out not to be what they thought it was.
I was inside the system when it did that to her.
I understand why it happened.
I do not accept that it was acceptable.
That refusal — to understand and still not accept — is not an analysis. It is not a policy recommendation. It is not a framework or a concept or a term that will appear in academic citations.
It is the only thing I know how to do with what I know.
And it is, I believe, what the knowledge requires.