Actionable insights — Growth Investing Done Right

1. Screen an unprofitable company with a single combined number before you look at anything else

1. Compute revenue growth and free-cash-flow margin for the same period, as percentages.
2. Add them. Require the sum to exceed 40 — the venture-capital "rule of 40".
3. Read the mix, not just the total: a company at 60% growth and a −10% margin scores 50 and is interesting; the same 50 built from 10% growth and a 40% margin is a different business entirely.
4. Check the direction of the margin over several periods. The rule tolerates losses; it does not tolerate losses that are getting worse.
5. Reject the trivial way to pass it — "it's easy to grow revenue fast if you sell $1 for 90 cents" — by asking what the growth cost.

• A rule imported wholesale from private-market software into public markets of every kind — the 40 threshold was calibrated on SaaS, not on distributors or insurers.
• The missing half: this screen has no price in it at all, so it can only tell you a company is good, never that it is worth what it costs.

2. Write the sell conditions before you own the thing, one set per company

1. At the moment you take a position, write down the specific, numeric conditions under which you will trim or exit it — company-specific, not portfolio-wide.
2. Phrase them as business conditions ("at least 15% revenue growth"), never as price conditions.
3. Allow a documented override, and define what an override looks like in advance: a broken rule plus credible guidance that reverses it is not a sell.
4. Distinguish a mistake from a deterioration. "Every company makes them. You have to be a bit tolerant. But blind tolerance is just hoping."
5. Keep the asymmetry in front of you when you apply them: the cost of selling a multibagger early is unbounded; the cost of holding a loser too long is capped at the position.

• An override clause broad enough to swallow the rule. Count how often it is used; a rule overridden every time is not a rule.
• Rules that were rewritten after the fact. The discipline only exists if the version being applied is the version written at entry.

3. Re-mark a quality score every quarter, so slow deterioration shows up as a number

1. Fix a list of criteria — here 17 — spanning management quality, growth, and whatever operating metrics matter for the model (his example: sales efficiency).
2. Score every holding on the full list after every earnings report, whether or not anything looks wrong.
3. Keep the history. The signal is the trajectory of the score, not its level in any one quarter.
4. Pair it with the sell rules: a falling and low score, or a broken rule, triggers a review — not an automatic sale.
5. Use it deliberately as an emotional instrument during drawdowns, which is the one time you cannot trust your own reading of the business.

• Score inflation: an owner marking his own holdings every quarter has an obvious incentive, and no third party audits the marks.
• Criteria that cannot fall. A score built mostly of qualitative judgements will drift downward only when the holder is already willing to sell.

4. Rank the portfolio by what you put in, not by what it is now worth

1. Keep a second view of the book weighted by original cost of each position.
2. Apply the position-size cap to that view — here roughly 8%, stretched to 10% occasionally.
3. Let the market-value weights run wherever the winners take them; a 20% market-value position is acceptable if you can hold it.
4. Decide the market-value tolerance personally and honestly: "Can you sleep well at night with a 20% position? I can, but it doesn't mean you can."
5. Trigger additions off business performance, never off the price: "I add when I see a company doing well, and I mean fundamentally, not the stock price."

• The cost of the rule: measuring risk at cost understates how much of today's money sits in one name. A cost-weighted 10% can be a market-weighted 25%.
• Adding to a name that is "executing" while the price stalls — right often enough to be a policy, and also exactly what averaging into a broken thesis feels like from the inside.

5. Test personalisation for its marginal cost, not for its existence

1. Ask whether the product is tailored to the individual customer at all — if it is a single undifferentiated offer, scale is now a disadvantage rather than a moat.
2. Then ask the harder question: what does one more customised customer cost? Personalisation delivered by software is nearly free; personalisation delivered by people or inventory is not.
3. Reject businesses where differentiation raises unit cost — "too much differentiation leads to higher costs and less profits".
4. Require both halves at once: unscaled and spread at large scale.

• Criteria admitted to be subjective and then used as if they were not — the mission-statement test is presented with no way to fail it.
• Recommendation engines counted as personalisation when the underlying catalogue is identical for everyone; the test should bite on the product, not the display.

6. Underwrite the drawdown before you buy, using base rates rather than resolve

1. Accept the historical price of the outcome you are aiming at: of 365 stocks that rose 100x, none avoided a 50% fall and most fell 75%+ more than once.
2. Assume in advance that you will experience that on any winner you own, and decide now what you would do.
3. Identify the real exit trigger, which is narrative rather than numeric: "it's not only the price drop that makes you sell, it's the stories surrounding the stock."
4. Put an instrument between yourself and the story — the score and the rules — and pre-commit to consulting it before acting.
5. If both are intact, treat the fall as a buying condition rather than a warning: "when it hurts the most, it's often the best time to buy."

• Survivorship in the evidence: the 365 hundred-baggers and Gardner's seven are the winners: the same "hold through anything" behaviour applied to the losers is what the base rate punishes.
• Unaudited performance claims attached to a subscription product — "+400% since then" and the June 2022 list "up 367% on average" are both self-reported here, with no source.