By Heorhi Tratsiak
There is a particular kind of silence that falls over a bank when everything is about to change.
I know this silence well. I lived inside it.
For seven years, from 2015 to 2023, I worked at one of Belarus’s largest banks — moving through roles that took me from the early stages of customer operations to the inner mechanics of international payment systems: Visa, Mastercard, Apple Pay, Samsung Pay, SWIFT. I watched money move across borders, watched institutions hold themselves together under extraordinary pressure, and watched human beings make financial decisions in circumstances that no textbook had ever prepared them for.
Those years gave me something that no MBA program can teach: an intimate, ground-level understanding of how people actually make financial decisions — not how they should make them, but how they do, when uncertainty is in the room.
Now, watching artificial intelligence reshape the financial world, I find myself returning to those years constantly. Because the question I keep asking is not whether AI is smarter than us. It clearly is, in certain ways. The real question is: what does AI still not understand about money — and what does it understand better than we ever could?
When Fear Enters the Room
During my years in banking, I witnessed something that no risk model fully captures: what happens to financial behavior when institutional confidence breaks down.
It didn’t matter what triggered it — whether economic news, political uncertainty, or rumors spreading through a community faster than facts could follow. The pattern was always the same. Clients who had been completely rational the week before would arrive at branches with a specific kind of urgency. Not panic, exactly. Something more controlled than that, and more dangerous: a quiet, determined decision to take back control of the one thing they could.
Their savings.
I remember queues forming not because a bank was failing, but because fear had arrived — and fear does not wait for facts. The institution honored every request. Cards kept working. Loans were still being processed. Some clients even opened new deposits during those periods, which struck me as either extraordinary trust or extraordinary resilience.
What I was witnessing was a masterclass in behavioral finance that no professor had ever taught me.
People were not making rational decisions. They were making survival decisions. Decisions shaped by incomplete information, social contagion, and the deeply human need to feel in control of something — anything — when the larger environment has become unpredictable.
This observation stayed with me. It shaped everything I came to understand about the relationship between AI and human financial judgment.
What a Bank Actually Teaches You About Money
Most people think of finance as a numbers game. After seven years inside it, I can tell you: finance is a psychology game with numbers attached.
The most important thing I learned was not how to calculate interest rates or reconcile a SWIFT settlement. It was how to read the person — or the institution — in front of me. Why does this client want this loan, really? Why is this business applying for this credit at this specific moment? What is the gap between what someone says and what their financial behavior actually reveals?
Credit decisions, at the ground level, are not made by spreadsheets. They are made by people reading other people — looking for the signal beneath the application, the human variable that determines whether the numbers on the page translate into real-world outcomes.
This is what separates a good banking professional from a form-processing function: the capacity to detect what the data doesn’t show. The client whose income statements are clean but whose cash flow tells a different story. The business owner whose confidence masks a structural problem. The borrower who will repay every cent because reputation means more to them than any collateral clause.
For years, this human variable was untranslatable. It lived in experience, in institutional memory, in something close to professional intuition. Machines could process applications. They could not read the room.
What AI Changes — and What It Doesn’t
Here is where it gets genuinely interesting.
Modern AI systems are beginning to do something remarkable: they are building models of human financial behavior that are, in certain measurable ways, more accurate than human judgment. Not because they understand emotion — they don’t — but because they have processed enough data to recognize the patterns that emotion produces.
Where a loan officer sees a nervous client, an AI sees a statistical cluster. Where I observed clients withdrawing savings during periods of institutional uncertainty, an AI might flag that behavioral pattern weeks earlier as a leading indicator of systemic stress — because it had seen the same signal in 40 other financial environments across 30 years of global data.
This is the genuine superpower of AI in finance: it operates without the cognitive biases that make human financial decision-making so unreliable. It does not experience loss aversion. It does not anchor to the last number it heard. It does not hold a losing position too long because admitting the loss feels like admitting failure.
I know this last pattern intimately. I spent years following financial markets alongside my banking career — technical analysis, fundamental research, risk management. And the single greatest obstacle I ever faced was not the market itself. It was the gap between what the data showed and what I was willing to accept. The moment a position moved against me, something irrational engaged. Not stupidity — something more primitive. The brain’s ancient circuitry, designed for physical survival, taking control of a financial instrument it was never built to operate.
AI does not have this problem. It processes the signal and responds to it — not based on how the decision feels, but based on what the evidence indicates. This is not a small advantage. In financial decision-making, it may be the entire game.
What Seven Years Taught Me About the Limits of Optimization
And yet.
I keep returning to those periods of institutional uncertainty I witnessed. To the clients who stayed. To the loan officers who made judgment calls on borderline applications — not because the numbers supported it, but because a career of watching real people manage real financial lives had given them a form of pattern recognition that no training data set yet contains.
These are not failures of rationality. They are expressions of something that AI, at least as it currently exists, does not possess: contextual judgment. The capacity to say — the data suggests one thing, but I have seven years of ground-level experience watching how this kind of situation actually resolves, and that changes how I read this application.
My own decision to leave banking in 2023 was not a calculation. I had stability, expertise, and a clear path inside an institution I understood deeply. I left because I had decided — with my full professional judgment, not just a cost-benefit analysis — that the next phase of my career required building something rather than operating within an existing structure.
An algorithm evaluating my situation would have recommended staying. The variables pointed clearly in that direction. And the algorithm would have been wrong — not because the data was bad, but because the decision involved a kind of forward projection that requires something beyond optimization: a vision of what you are trying to build, and a willingness to accept short-term risk in service of a longer-term direction.
This is the category of judgment that AI is not yet equipped to replicate.
The Payment Systems Layer Nobody Sees
One dimension of AI’s impact on finance that receives less attention than credit scoring or fraud detection is its effect on the operational infrastructure that makes financial transactions possible.
For years, I sat at the intersection of this infrastructure — managing the settlement relationships between our bank and Visa, Mastercard, Apple Pay, and Samsung Pay. Every month: transaction reconciliation, fee calculations, discrepancy resolution across multiple currencies and reporting systems. And managing the financial obligations of smaller banks that operated through our infrastructure as a principal member.
This work was precision-intensive and operationally complex. Small errors compounded. Timing discrepancies between systems created reconciliation gaps that required expert interpretation to resolve. The professionals I worked alongside at the payment networks were extraordinary — experienced, methodical, and precise in ways that made the system function reliably at scale.
AI is beginning to transform this layer of finance significantly. Reconciliation processes that required days of expert analysis can now be processed in minutes. Anomaly detection that once depended on individual expertise can now flag discrepancies across millions of transactions in real time.
But the judgment layer — the professional who can interpret an anomaly, understand its institutional context, and determine whether it represents a system error or a pattern worth escalating — still requires human expertise. What AI changes is not the need for that expertise. It changes the questions that expertise is asked to answer.
Where This Leaves Us
The financial industry is being rebuilt. AI is not approaching it from the outside — it is already inside it, already reshaping credit decisions, already transforming fraud detection, already processing the settlement infrastructure that most banking professionals never see directly.
This is, on balance, a good development. The biases that human judgment carries — developed over careers but also shaped by blind spots — have produced real errors for real clients over a long period. A system that evaluates financial behavior based on comprehensive data patterns rather than individual cognitive shortcuts is, in many contexts, a more accurate system.
But accuracy is not the same as wisdom. And optimization is not the same as sound judgment.
What seven years in banking — through periods of institutional stress, market volatility, regulatory change, and the day-to-day complexity of financial operations — taught me is that financial systems are human systems. They are trusted or abandoned based on human perception. They hold together or break down for human reasons that don’t always appear in the data until after the fact.
The most important professional skill in finance has never been calculation. It has been the capacity to be present with complexity — to understand what a client, a counterparty, or a market is actually communicating, and to make decisions that account for the full situation rather than just the measurable variables.
AI is becoming genuinely extraordinary at the measurable layer. The question for the next decade is not whether it will transform finance — it already is. The question is how the professionals who understand the human layer of financial systems will learn to work with it, and what they will bring to that collaboration that no model can replace.
I spent seven years watching financial decisions get made under real conditions. What I observed was not irrationality to be corrected. It was judgment being exercised in the presence of uncertainty.
That kind of judgment is still ours to develop. And in an era of increasingly capable AI, it may be more valuable than it has ever been.
Heorhi Tratsiak spent seven years in banking and payment operations, managing settlement relationships with Visa, Mastercard, Apple Pay, Samsung Pay, and SWIFT. He holds a Yale Certificate in Financial Markets and writes about the intersection of financial systems, technology, and decision-making.
