Notes from a work-in-progress on valuing businesses that promise you everything

There’s a certain kind of company you’ve probably met.

It has a perfectly respectable core business — makes real products, earns real (if modest) profits, has done so for years. Then, over eighteen months or so, it announces that it is also entering five frontier industries. New materials! AI supply chains! A joint venture with a famous research institute! The stock triples. The P/E ratio hits 300.

And then the arguing starts. Value investors call it a bubble. Growth investors call the value investors dinosaurs. Sell-side analysts publish target prices that differ from each other by a factor of three. Everyone is very confident.

I’ve spent the past few months building a valuation framework for exactly this kind of company — partly as an intellectual project, partly out of self-defense, because I own one. Here’s the tour, minus the equations.

Why the usual tools whiff

P/E ratio? Useless. The earnings come from the boring old business; the market cap comes from the five new stories. Dividing one by the other is like judging a “house + lottery tickets” bundle by the rent on the house. The 300x tells you nothing — not that it’s expensive, not that it’s cheap. It’s just noise wearing a number’s clothes.

DCF? In theory, perfect. In practice, each new venture requires you to guess technology success, customer adoption, ramp-up speed, and pricing — and multiply your guesses together. The output range spans an order of magnitude. At that point the model isn’t disciplining your judgment; it’s decorating it.

Sum-of-the-parts? This one looks purpose-built for multi-business companies, which is what makes it dangerous. In practice, analysts value each new venture as if it will succeed, assume the ventures don’t compete with each other for anything, and add everything up. Under this arithmetic, every new press release raises the target price. Announcing a sixth venture is always good news. A seventh? Even better. You see the problem.

The flip: stop asking who wins

Here’s the move that unlocked everything for me.

Predicting which of the five ventures will succeed is nearly impossible — even the founder doesn’t know. But you don’t need to. Ask a different question instead: what’s the probability that all of them fail?

Say each venture has just a 40% chance of working — pretty pessimistic. If the five were independent, the chance that every single one flops is 0.6⁵ ≈ 8%. Which means there’s a 92% chance at least one becomes a real business. You don’t know which one. You don’t need to know. That’s the beginning of a value floor — a reason the company is worth more than its boring core, without requiring you to pick winners.

Nice, right? Now let me ruin it.

The buffet-plate problem

That 92% assumed the ventures are independent and that adding more ventures is free. Neither is true, because every venture draws from the same pot: the founder’s attention, the engineering team, the balance sheet.

So I modeled it. Give the company a fixed amount of “management attention,” split it across N projects, and let each project’s success odds shrink as its share of attention shrinks. Then something wicked happens, and it depends on how attention-hungry each project is:

  • Light projects (partnerships, licensing, minority stakes): spreading out genuinely helps. Cast a wide net. The math approves.
  • Heavy projects (build-your-own factory, furnaces that can never be switched off): the math turns hostile. In my baseline calibration, running five heavy projects gives you almost exactly the same all-fail probability as focusing on one — around 40% either way — except now you’ve spent five times the capital and quintupled your organizational chaos.

Diversification’s benefit gets entirely eaten by dilution. I call it the diversification–dilution trade-off. My grandmother called it “your eyes are bigger than your stomach.” Same theorem.

The practical corollary is my favorite line in the whole paper: a company’s real strategy is written in its capital expenditure, not its press releases. Five announced “strategic directions” mean nothing. Where the money and the dedicated senior hires actually go — that’s the strategy.

The floor beneath the floor

One more wrinkle. Those five ventures usually share something deeper: the same balance sheet, the same key partner, the same irreplaceable founder. If that common foundation cracks, every venture dies at once — no matter how “diversified” the portfolio looked.

So the true all-fail probability has a hard bottom: the fragility of whatever everything depends on. You can stack ten options on top of thin ice; you cannot diversify away the ice. When I look at a company like this now, my first question is no longer “which venture is most exciting?” It’s “what’s the ice?”

Where do the probabilities come from? (Not from vibes)

The fair objection to all of this: aren’t those success probabilities just… made up?

They don’t have to be. Drug companies solved this decades ago — nobody guesses whether a Phase II molecule will succeed; they look up the historical transition rate across thousands of trials, then adjust for specifics. Industrial companies have an equivalent database hiding in plain sight: competitors’ IPO prospectuses, which document — under legal liability — how long factory ramp-ups actually took and when profits actually arrived. (Spoiler: two to three years, basically always, no matter what management says on the earnings call.)

My working calibration for a new venture going from “tech works” to “actual profits”: roughly 9% by default, 25% with partial hard evidence, 64% only after orders and certifications are publicly confirmed. The rule that keeps the whole thing honest: the 64% must be earned with evidence, never assumed from enthusiasm. The gap between what the market believes and what the evidence supports even has a name in my ledger: narrative premium.

Keep receipts

The last piece is embarrassingly simple: a prediction ledger. Every tip, every management forecast, every “my source says next year they’ll do 3–5x revenue” gets written down with a date. When the date arrives, settle the account. Sources that deliver earn smaller discounts on their next claim; sources that whiff earn bigger ones. Your information network stops being a rumor mill and becomes a scoreboard.

This sounds petty. It is the single highest-return habit in the entire framework. I recently watched three “independent” channels tell me the same exciting story — and the ledger revealed they all traced back to one original source. Three echoes, zero confirmations.

The punchline

The framework doesn’t spit out a magic price. What it does is convert a shouting match (“it’s a bubble!” / “you just don’t get it!”) into a checklist of things you can actually verify: What’s the boring core worth? What’s the ice? Where is the capex really going? Which claims have hard evidence, and which are just well-traveled echoes?

And it redefines where the edge comes from. In these stocks, the money isn’t made by believing stories earlier than everyone else — that’s a game of musical chairs. It’s made by verifying stories earlier than everyone else. Belief is free. Verification compounds.

The full paper — with the actual math, the tables, and an anonymized case study — is still being polished. If any of this made you look at a stock in your own portfolio and quietly whisper "…what’s the ice?" — then it’s already working.

This post describes a research framework, not investment advice. All companies mentioned are stylized composites. The furnaces, however, are real, and they really cannot be switched off.