Actionable insights — Best Buys February 2026

1. Turn a sector-wide narrative into a shopping list

1. Spot the case where a whole sector derates on one story while the index it belongs to sits near highs — the dispersion itself is the opportunity.
2. State the bear narrative in its strongest form, in the market's own words, before answering it.
3. Concentrate a whole issue's work on that one sector rather than diversifying attention — the mispricing is where the panic is.

• Narratives that are actually right for part of the sector — the screen must be able to say which part.

2. Concede the narrative where it is true — then define who it doesn't apply to

1. Grant the disruption case explicitly for the businesses it fits: "If a company's only value is completing a basic task (like simple data entry or basic reporting), it might be in trouble."
2. Then test each candidate against three moats a general-purpose model does not defeat:
   (a) Proprietary data — decades of private, specialised data that competitors cannot obtain. "AI is only as good as the data it trains on."
   (b) High switching costs — a system whose replacement threatens the operation. "Most CFOs won't risk their entire operation to save a few dollars."
   (c) Operating leverage — the company is already deploying AI inside its own product and cost base, so the technology shows up as margin rather than as competition.
3. Require the candidate to pass at least one strongly, ideally two.

• Evidence the moat is being tested in the numbers — churn, renewal rates, net revenue retention — rather than argued in prose.

3. Weight switching costs by the cost of being wrong, not by the price of switching

1. Ask what happens to the customer if the software makes a mistake. Where an error means a regulator, a lawsuit or a stopped operation, price competition is nearly irrelevant.
2. Prefer products embedded in compliance and workflow over products bought on features.
3. Confirm with observed behaviour — retention rates, contract length, average tenure — instead of with the theory.

• A regulator or standards body legitimising an alternative — that is the one thing that lowers the cost of being wrong.

4. Use the "first drawdown of this size ever" as a rare-event marker

1. For a long-term compounder, look up its worst historical drawdown and compare today's to it.
2. A decline outside the entire historical range is either genuinely new information or a rare pricing opportunity — force yourself to name which, in one sentence.
3. Check whether the drawdown coincides with a governance event (a founder leaving) that is separable from the business story.

• Acquisition multiples and deal counts in the next few quarters — for a serial acquirer that is where a broken model would show first.

5. Assess succession on tenure, deal record and open-market buying

1. When a founder steps back, judge the successor by three things: how long they have been inside the company, what they personally built there, and how much of their own money is in the stock.
2. Weight open-market purchases above granted equity — buying with after-tax cash during a drawdown is the strongest single signal available.
3. Note whether the founder stays on the board, which preserves the capital-allocation culture without the key-man dependency.

• Insider selling during the drawdown, and any change in the decentralised operating model the founder built.

6. Look for two shareholders buying at once — the company and the insiders

1. Check the direction of the share count while the price falls. A rapidly shrinking count against rising revenue and earnings is value accruing per share regardless of sentiment.
2. Layer insider open-market purchases on top: management and the buyback pulling the same way is a two-sided vote.
3. Verify the buyback is funded by free cash flow and that revenue and earnings really are still rising.

• Buybacks that merely offset stock-based compensation — net share count is the number that matters.

7. Classify software vertical vs horizontal before judging its durability

1. Separate horizontal tools (broad, used by everyone — spreadsheets, chat) from vertical software built for one industry's specific workflow.
2. Expect vertical software to show low churn, high pricing power (critical but a small share of the customer's budget) and deep relationships — and check the recurring-revenue share to confirm it.
3. Recognise the asymmetry in an AI narrative: generic horizontal tools are the more replaceable ones, yet the panic prices both alike.

• Price increases sticking without churn — the practical test that the pricing power described is real.

8. Frame the whole thing with the voting machine — and let waiting be the position

1. Treat the narrative-driven decline as the voting machine, and the earnings trajectory as the weighing machine: "stock prices eventually follow earnings."
2. Buy high-ROIC, cash-generative, dominant businesses during the narrative rather than after it resolves — "the big money is in the waiting" (Munger).
3. Accept that the position may look wrong for as long as the narrative runs; that is the price of the discount.

• The earnings themselves. If revenue and earnings start following the narrative down, the weighing machine has changed its answer.