Nine real companies, one framework — and a rule that turns hindsight into falsifiable judgment

This is where the whole book gets put to work. Twenty-one chapters built a toolkit — business model, financials, moat, AI fate, valuation, expectations — and this chapter runs it across nine real companies. Three ground rules up front, because they’re the point. This is not stock picking: every case demonstrates how to think, not what to buy. It presumes no winner: each company is dissected with the same framework, famous or not. And every company’s data is cited as verified from earlier chapters, with no specific market multiples — those are market data requiring a dated source.

The nine sort into thesis types. Long-term compounders (ServiceNow, Veeva) — high margin, deep moat, already profitable, AI likely thickening the moat; closest to a true compounding machine, so their biggest risk isn’t the company, it’s paying a perfect price. High-growth, high-valuation (CrowdStrike, Snowflake) — enticing growth, but each with a worry (CrowdStrike’s ~23% SBC, Snowflake’s procyclicality and lower margin); where “good company, bad stock” risk concentrates. Platforms (Microsoft, Salesforce) — the heaviest artillery in the AI entry-point battle, whose fate turns on the will to self-disrupt. Vertical + payments (Toast) — split software from payments and value each separately, or be fooled by a fat top line. Turnaround candidates (Okta, Twilio) — the thesis is repair, not growth, and the job is distinguishing a mispriced opportunity from a deteriorating trap. And the value trap — deliberately unnamed, because calling a company “permanently declining” is a strong claim; instead the chapter draws its profile (low multiple + moat erosion + slipping retention + worsening cash flow + management “managing” the metrics).

Read across the nine and a pattern emerges that’s more useful than any single verdict: the same framework asks the same questions of every company, regardless of how the story sounds. Is the moat deepening or eroding? Does AI thicken or thin it? What’s the cash-flow quality net of SBC? What does the price already assume? And what evidence would change my mind? A company can move between types over time — today’s compounder can become tomorrow’s turnaround — but the questions never change.

The single most important habit the chapter enforces: every case ends with kill criteria — the specific evidence that would prove the thesis wrong. That’s the weapon against hindsight storytelling. A case analysis without kill criteria is just a plausible story reverse-engineered from today’s price; one that names, in advance, what would falsify it is a real, testable judgment. Learn the method, not the verdicts — because verdicts go stale (today’s winner can become tomorrow’s distress), and only the inquiry doesn’t.

This is Chapter 22 of “Compounding in the Cloud.”