What Running Product Inside a Global Bank Teaches You That a Startup Cannot

I have run product at both ends of the size spectrum. At HSBC I was a manager of managers inside a 50–70 person product organisation, with a global function of ten technical product managers reporting into me. At Irys, a Series A startup, I was CPTO — the sole product and technology decision-maker, with nobody between me and any call that mattered. In between sat Shieldpay, a regulated scale-up where I ran technical programmes and watched a company live through the awkward middle of that spectrum in real time.

The common belief about this range is that it is one job in different amounts: more people, more meetings, more zeros, same fundamentals. Fifteen-plus years in product and technology have convinced me that is wrong in an important way. At a certain size the job does not get bigger — it changes kind. Your work stops being making decisions and becomes building the system that makes decisions without you. Most product leaders discover this too late, usually by trying to run the large organisation the way they ran the small one and wondering why they have become the bottleneck in everything they touch.

How a decision travels

At the startup, a decision is a conversation. Someone raises a problem, the people who understand it are already in the room or one message away, a call gets made, and by the afternoon something has changed in the product. The decision and the decider are the same thing. The feedback loop from choice to consequence is days, sometimes hours, and the cost of being wrong is usually a revert.

Inside the bank, a decision is a journey. It has to be written down, because it has to travel further than your voice reaches. It has to survive being retold by people who were not in the room when it was made — a product manager in one market explaining it to an engineering lead in another, who explains it to an architect, who raises it with risk. It has to be legible to functions that were not part of forming it: legal, compliance, security, audit. And it has to hold its shape across time zones and across the months between being made and being fully acted on.

This is why large organisations produce so many documents, and why startup people find that ridiculous until they have worked at scale. The document is not bureaucracy. The document is the vehicle. A decision that only exists in your head, or in a conversation, cannot travel — and at scale, a decision that cannot travel has not actually been made. It will be re-made, differently, by everyone who never heard it.

The deeper shift is this. At the startup I made perhaps a dozen consequential product calls a week, personally. At the bank, the organisation around me made hundreds — and I saw a small fraction of them. My real output was no longer decisions; it was the quality of the decisions made by people two levels away from me, most of which I would never review. That meant my actual work was strategy people could apply without asking, decision rights that were explicit rather than folkloric, escalation paths that surfaced the genuinely hard calls and absorbed the rest, and managers hired and developed to judge well when nobody senior was watching. When that system works, things go right in rooms you are not in. When it does not, no amount of personal brilliance saves you, because you are one person and the organisation makes its choices at a rate no individual can inspect.

That is the qualitative change. Below a certain size, you are the decision-maker. Above it, you are the designer of a decision-making system — and those are different crafts, requiring different instincts, and being excellent at one tells you very little about the other.

What alignment actually costs

Startups use "alignment" to mean a meeting that should have been an email. Inside a global bank, alignment is a genuine budget line, paid in the scarcest currencies the organisation has: senior attention and calendar time. Getting a significant change agreed across product, engineering, architecture, risk, legal, compliance and the markets it touches consumes real capacity — and the honest accounting, which almost nobody does out loud, is that a meaningful share of a large organisation's total effort goes on keeping itself pointed in one direction.

The startup instinct is that this is pure waste. That instinct is wrong, and holding onto it is the fastest way to fail at scale. When many teams share one brand, one balance sheet and one regulator, alignment is what stops them building three versions of the same thing, or two features that individually make sense and jointly break a customer journey, or anything that quietly commits the firm to a risk nobody priced. The cost is real, and so is what it buys.

What scale actually teaches you is the discrimination the startup never forces on you: the difference between alignment that de-risks execution and alignment that launders accountability. The first kind puts the right people in front of a decision before it becomes expensive to reverse. The second kind is a meeting whose true purpose is to get everyone's fingerprints on a decision precisely so that nobody's are — consensus as insurance. They look identical on a calendar. Learning to tell them apart, and to decline the second kind without becoming the person who "doesn't understand how things work here", is one of the most transferable skills the bank gave me. Every organisation above about fifty people has both kinds, whatever it calls itself.

Regulated environments make you better, and slower

Working under regulation made me a better product person for one blunt reason: it forced articulacy. Every decision had to survive an audit, which means every decision had to be explainable — what we chose, what we rejected, what we knew at the time, what would happen if it went wrong. A decision you cannot explain to a regulator is a decision you cannot ship. Do that for years and second-order thinking stops being a virtue and becomes a reflex. You stop asking only "what happens when this works" and start asking, automatically, what happens when it fails, who is harmed, what data it touches, what recourse the customer has, and how you would defend the call in front of someone whose job is to distrust you. Startups talk about customer empathy; regulation industrialises a harsher and more useful version of it — empathy for the customer on the day something goes wrong.

The same environment also made me slower, and honesty requires separating the two kinds of slowness. Some of it buys safety at a price worth paying: when the blast radius of a mistake is people's money, a deliberate pace is not timidity, it is proportion. But some of it is sediment — controls that outlived the risks they were built for, sign-offs that exist because removing them is nobody's job, process that accumulated the way process does. The best operators I worked with inside the bank were not the ones who followed every rule or the ones who ignored them; they were the ones who knew which rules were load-bearing. That discernment — respecting the constraint you understand, challenging the one nobody can justify — turns out to be precisely the skill the current wave of AI governance demands, which is one reason people with regulated-product scar tissue are suddenly interesting to companies that never previously hired from banks.

What the bank does better than startups admit

Startup culture treats big-company practice as a cautionary tale, which makes it easy to miss what the bank genuinely does better.

It survives people leaving. The bank is designed on the assumption that anyone might be gone tomorrow — documented decisions, understudied roles, managers developed on purpose rather than promoted in a hurry and abandoned. Startups run on heroics and call it culture, and the bill arrives the day the hero resigns. I have seen a critical system at a startup whose entire operating knowledge lived in one head, and I have seen the bank version, where the third person to hold a role could reconstruct the reasoning of the first. The second is not glamorous. It is how things last.

It manages risk instead of merely avoiding or ignoring it. The caricature says banks are risk-averse; the reality is that a bank prices risk professionally — takes the ones it is paid for, declines the ones it is not, and knows the difference. Most startups do not manage risk at all; they simply do not look at it, and the ones that die of it rarely get to write the retrospective.

And it takes the development of managers seriously, because it has to. When your output is other people's judgement, you cannot treat management as an afterthought bolted onto your best individual contributor. The bank's machinery for this is imperfect and slow, but it exists, and most startups have nothing in its place.

What the startup does better than the bank will ever manage

The startup's advantage is proximity to truth. The user is close, the feedback loop from decision to consequence is short, and — this is the part that matters — the consequence lands on the person who made the call. At Irys, if I got a product decision wrong, I felt it within weeks and I fixed it, and both the error and the correction made me sharper. Inside the bank, decision and consequence are separated by enough layers and enough quarters that the learning diffuses. Something underperforms a year after it shipped, attributed to a committee, absorbed into a portfolio review. Nobody's judgement improves, because nobody's judgement was legibly on the line.

This is not a failure of will, and the bank cannot fix it by exhortation. The layers that slow the learning are the same layers that contain the blast radius; they are structural, and they are there for reasons. Which is exactly why the startup years matter so much in a product leader's formation: they are where judgement gets trained at full feedback speed, with your own name on every outcome. The bank then teaches you to scale that judgement through other people. Neither environment teaches both.

What transfers when you go back down — and what does not

Coming down from the bank to the Series A, I found the transfer was asymmetric in ways worth naming.

What transfers: writing as a decision technology, even when the whole company fits in one room, because written decisions survive the retelling and the pivot. Explicit decision rights, because "we're small, everyone just knows" is false at fifteen people and catastrophic at forty. The audit-shaped reflex of knowing how you would explain a call before you make it. Calm in front of scrutiny — a board, a due-diligence process, an enterprise customer's security review — because nothing a startup can show you is more adversarial than a bank's internal challenge done properly. And real delegation, the kind where you design what someone owns rather than hovering over how they do it, because manager-of-managers work burns the hovering out of you.

What does not transfer: the cadence. Move at bank speed in a startup and the startup dies while you are aligning stakeholders it does not have. The invisible scaffolding does not transfer either — inside the bank, a specialist function had already thought about half the things I never consciously worried about, and at Irys every one of those unowned things was mine: the security posture, the vendor contracts, the data protection stance, the incident process. Nobody else was coming. The bank quietly trains you to assume someone else owns the thing you have not thought about, and unlearning that assumption is the single hardest part of the descent.

The failure modes are symmetric. Bank habits at a startup become process theatre — governance for an organisation that does not exist yet. Startup habits at a bank become the lone hero making fast calls that cannot travel, admired in the room and undone everywhere else. I have watched both, and been guilty of at least one.

The product leaders worth trusting with a company are the ones who can read which organisation they are actually standing in — who can build the decision-making system when the organisation is big enough to need one, and dismantle it, without grief, when it is not. Scale is not a ladder you climb once and then know. It is a dial, and the whole job is reading it correctly.