Bayesian thinking: Why CFOs need it more than they admit

The discomfort of being certain in public

There’s a particular discomfort that comes with being a CFO in a retail trading business. The numbers move quickly, the narrative moves faster, and just as you think you’ve formed a view, the market hands you a fresh reason to doubt it.

Which is why I keep coming back to Bayesian thinking. Not because it sounds clever in a meeting room — it doesn’t, it mostly earns polite nods — but because it describes how you actually have to operate when you’re responsible for capital and risk in a business where certainty is decorative.

I first met Bayes’ theorem in a probability course as an engineering undergraduate. One more formula to memorise before an exam, wedged in between distributions and confidence intervals, and I filed it away as algebraic housekeeping. It has taken most of a career in finance to realise it was the only thing on that course I still use.

Not the formula itself — I couldn’t tell you when I last wrote it out — but the idea underneath it. What you believe should be a function of what you believed before and what has arrived since. Nothing else gets a vote.

Start with a view, then let it be moved

I’ve never had much time for the idea that finance deals in clean answers. It’s a series of calls made on partial information, each one shaped by what came before, each one due for revision the moment the facts shift. Start with a view. Update as the evidence arrives. Try not to fall in love with the first answer. That’s the whole discipline, more or less. Having a view is the easy part; having a way to change it is the job.

In retail trading it feels less like a framework and more like weather. This isn’t a sleepy business where the inputs drift gently and the forecasts fail politely. It’s behaviour, volatility, acquisition cost, conversion, retention, spreads, regulation — and the occasional piece of complete nonsense that turns up on a Tuesday afternoon and ruins the shape of the month.

So when someone tells me they’re confident, I reach for the caveats. Confidence is only useful paired with a willingness to update quickly. A good CFO doesn’t cling to the original model because the board approved it, or because it looked elegant in the deck. A good CFO asks the more awkward question: what do we believe now, having watched what the business has actually done?

The damage rarely comes from being wrong at the outset. It comes from being slow to admit the evidence has started pointing somewhere else.

Forecasts are not truth

One of the more maddening habits in finance — and in business more widely — is treating forecasts as though they were facts. They aren’t. They’re assumptions in a nice suit.

A forecast is a view of the world based on the evidence available at the time. If the evidence changes, the view should change. Obvious, you’d think. In most businesses it turns out to be genuinely hard. People get attached. They defend the number. They explain it away. They wait for more certainty, as if certainty ever arrived on schedule. It doesn’t — or at least it never has for me.

In a trading business this bites quickly. A small shift in market conditions, product mix or customer quality can change the entire shape of the next quarter. If you’re still working from last month’s assumptions, you’re already behind.

The Bayesian habit helps because it enforces one useful rule: no number is sacred. Every one of them is provisional.

The real job is updating

This is what makes the role interesting, and occasionally exhausting. You aren’t just reporting what happened. You’re deciding what should happen next, on the latest evidence, usually sooner than you’d like.

Which means revising a view without turning it into drama. If customer economics soften, the answer isn’t panic. If a product outperforms for six weeks, that isn’t a victory lap — six weeks is a small sample, and cohorts have a habit of reverting once you stop admiring them. If a channel disappoints, the answer isn’t to bury it in the broader narrative and hope nobody asks. In each case the job is the same: update your sense of what happens next, and say so out loud.

It sounds technical, but it’s really a behavioural skill. The best finance leaders I know aren’t the ones who never change their minds. They’re the ones who change their minds for the right reasons, at the right speed.

Capital allocation is probabilistic

This becomes unavoidable once you move from forecasting into capital allocation. Every meaningful decision in a growth business carries an uncertain payoff — hiring ahead of demand, building new tooling, entering a new market, changing pricing, tightening credit terms. None of it comes with a guarantee.

So the question is never “will this work?” The honest question is what the probability-weighted return looks like, and how much we lose if we’re wrong.

Framed that way, decision-making gets a lot less theatrical. You stop hunting for perfect certainty and start looking for the evidence that shifts the odds: customer behaviour, cohort performance, channel efficiency, plain operational reality. You aren’t trying to eliminate uncertainty. You’re just trying to respect it. For a CFO that isn’t philosophy, it’s the day job.

It’s also why finance finds AI unsettling

There’s a version of this argument playing out right now, and it’s worth naming.

Traditional finance is deterministic by design. Double-entry bookkeeping, reconciliations, controls — the whole apparatus is built on the premise that the same inputs produce the same outputs, and that an answer is either right or wrong. This isn’t a flaw — it’s the whole point. You cannot run an audit on vibes.

AI doesn’t work like that. Ask a model the same question twice and you may get two different answers. What comes back is a distribution rather than a fact — and, unhelpfully, it rarely sounds like one. To a finance team raised entirely on determinism, that reads as a defect. To anyone who thinks in Bayesian terms it feels oddly familiar. It’s how the business itself has always behaved, even when the ledger pretended otherwise.

I suspect that’s part of why some finance functions are adapting faster than others. The ones getting on with it already knew that most of what matters — forecasts, valuations, capital decisions — was probabilistic all along. The ledger is deterministic. The business never was. That’s also why metering cognition unsettles people: you’re being asked to buy a distribution by the unit.

Human judgement still matters

I should be careful not to oversell this. Bayesian thinking is a way of thinking more clearly under uncertainty. It isn’t a machine that hands you the answer, and it’s no substitute for judgement. There’s still room for instinct, pattern recognition and context — arguably more room, in a business that moves quickly. The trick is using them without tipping into overconfidence.

I’ve watched businesses fall into both ditches. Some become paralysed, waiting for evidence that never fully arrives. Others become dogmatic, treating an early thesis as destiny. Neither lasts. What lasts is the habit of disciplined revision.

A CFO’s advantage

That, for me, is the real prize. Bayesian thinking gives you permission to be provisional without being weak. You can say, entirely honestly, “this is our best view for now” and mean it as a position of strength rather than a hedge.

Good decisions mostly get made before the full picture is available. The leaders who update cleanly and communicate clearly are the ones who keep an organisation moving while everyone else is still defending last quarter’s slide deck.

In a retail trading company, where the environment changes faster than the narrative, that stops feeling like theory and starts feeling like survival. The best finance leaders aren’t the ones who sound certain all the time. They’re the ones who stay useful when certainty disappears.