I once built a five-year projection so beautiful I almost forgot it was fiction.
It had forty-one tabs. Revenue cascaded down quarters in clean green cells. There was a cohort retention curve that bent upward at exactly the angle a venture investor likes to see, a CAC payback line that crossed into profit in month fourteen, and a discounted cash flow model with a terminal value I had pulled from somewhere between thin air and a comparable I half-remembered. I had built a machine that printed a number, and the number said I was going to be rich. I stared at it the way you stare at a photo of yourself from a good angle, and for about three days I believed it the way you believe a flattering photo: completely, and against all evidence.
Then a friend who actually ran an aerospace shop looked over my shoulder and asked one question. "What happens if churn is three points worse than you typed?"
I changed one cell. The beautiful number ate itself.
That was the day I started thinking about financial models the way an engineer thinks about a wind tunnel.
You are not predicting the weather
People misunderstand what a wind tunnel is for. They think you build one to find out what the wind will do. You don't. The wind outside is going to do whatever it wants — gusts, crosswinds, a bird, a downdraft no equation saw coming. A wind tunnel makes no claim about Tuesday's weather over Kansas. It does something narrower and far more useful. It puts a wing in a controlled stream of air and asks a single brutal question: at this speed, at this angle, does the wing hold, or does it shear off?
A financial model is the same instrument pointed at a business. It is not a forecast. Treating it as a forecast is the original sin, and almost everyone commits it, because the spreadsheet is so confident. It returns numbers to two decimal places. It does not blink. It will tell you that in March of year four you will have 11,402 customers, and it will tell you this with the same flat authority whether your assumptions are sober or insane. The decimals are theater. The model has no idea what March of year four holds and neither do you.
What the model actually does — the only honest thing it does — is take the argument you are already making in your head, the one you make to investors and to your spouse and to the version of yourself that can't sleep, and force it out of prose and into arithmetic. Prose hides contradictions. Arithmetic can't. You cannot write "we'll grow fast and keep costs lean and hire the team we need" in a spreadsheet, because the spreadsheet makes you say how fast, how lean, how many, by when, and then it multiplies those claims together and shows you what you actually believe when your beliefs are forced to share a room.
That multiplication is the wind. You are testing whether the wing holds.
The output isn't the number
Here is the part that took me an embarrassingly long time to learn, and I say that as someone who once wrote financial models good enough to fool aerospace executives, designed the logo and the packaging too, and was quietly certain the whole package was Pulitzer material. The output of a model is not the number in the bottom-right cell. The number is exhaust. The real output is a feeling you have about the number, and that feeling has a name. It's conviction, or it's the absence of conviction, and learning to read which one you've got is the entire skill.
Watch what happens when you do it honestly. You build the model. You get a result. And then you do the only test that matters: you reach over and break it. You drop conversion by a third. You double the time-to-close. You assume the channel that's working today quietly stops working, the way channels do. You make the wind ugly.
One of two things happens.
Either the business still stands — uglier, slower, less rich, but standing — and something settles in your chest. That settling is conviction. It isn't optimism; optimism is what you feel about the pretty version. Conviction is what's left after you've tried to kill the thing and failed. It's the residue of survived stress tests. You can build a company on it because it doesn't evaporate the first time reality disagrees with your forecast, which it will, in week two.
Or the business shears off. One realistic gust and the wing comes apart. And here's where the model earns its keep, because the spreadsheet just did you the enormous favor of crashing in private. It cost you a weekend instead of three years and your investors' money. People talk about models as tools of ambition. Half their value is as tools of refusal — quiet machines for finding the no before the no finds you.
Conviction and delusion wear the same coat
I want to be careful here, because there's a failure mode that looks exactly like the success mode, and 2023 is thick with it.
Conviction and delusion are both unshakeable. That's the trap. From the inside they feel identical — the same steadiness, the same impatience with skeptics, the same certainty that the doubters lack vision. Founders are praised for this certainty right up until the morning it's revealed as the other thing, at which point the same trait gets renamed in the post-mortem. The press calls it vision on the way up and hubris on the way down, and it was the same coat the whole time.
The only way I know to tell them apart is the wind tunnel. Delusion is conviction that has never been stress-tested, or worse, conviction that fled the room the moment the test got hard. The deluded founder builds the forty-one-tab model and admires it. He never changes the churn cell, because some animal part of him already knows what the cell will do, and not knowing feels better than knowing. Conviction is what you have after you've changed every cell that scares you and looked directly at the wreckage and concluded the thing flies anyway.
So the discipline isn't building the model. Any analyst can build the model. The discipline is being willing to break your own beautiful thing — to treat your favorite assumption as the prime suspect rather than the protected witness. That's hard, and it's hard for a reason that has nothing to do with math. We don't model our businesses to find the truth. We model them to feel the way the green cells made me feel for three days. The spreadsheet is a flattering mirror until you decide to make it an honest one, and that decision is an act of will, not of skill.
What the wing is really telling you
The questions that matter in a startup are rarely the ones the model answers. The model will not tell you whether to start the company. It can't tell you whether you'll still want it at year three, when the chase has gone tedious and the novelty's worn off — and I say that as someone constitutionally prone to wandering off when the interesting part ends. It won't tell you whether the world needs the thing or whether you just want to have built it.
But it does something quieter and more durable. It separates the part of your ambition that can survive contact with arithmetic from the part that's pure want. Both parts are allowed. You need the want; nobody crosses a desert on a sensitivity analysis. But you should know which is which before you commit a decade, because the desert doesn't care how you felt about the photo.
I think this is why the founders who last tend to have a strange relationship with their own models. They take them seriously and they don't believe them at all. They run the numbers obsessively and quote them to no one as gospel. They've internalized that the spreadsheet was never an oracle — it was a sparring partner, and the point of sparring was never to win. The point was to find out, before the real fight, whether they could take a punch.
I still build the models. I build them more carefully now, with fewer tabs and uglier assumptions, and I trust the number in the corner about as far as I'd trust a weather forecast for a year from Thursday. What I trust is the feeling I'm left with after I've done my level best to break the thing and watched it refuse to break.
That feeling is the only forecast worth having. Build the tunnel. Then go and try to kill the wing. What's still standing when you're done is not your prediction.
It's your conviction, and now you've earned it.