Actionable insights — Our Top Picks: 2025 Results

1. Compute the value-versus-price gap explicitly, in percent

1. Measure the change in intrinsic value over the year using a per-share operating figure — owner's earnings, FCF per share, NPATA — not reported EPS.
2. Measure the change in the share price over the same period.
3. Subtract: the difference is how much cheaper (or dearer) the business became, independent of what it is worth in absolute terms.
4. Only then ask whether the derating is explained by something real. If nothing operational broke, the gap is the opportunity.

• Value measured on a metric the company chose. Owner's earnings and NPATA both add back real accounting charges — legitimate for a serial acquirer, and worth restating yourself once.
• A gap that keeps widening for years. At some point the market is pricing something the metric does not capture — which is the EVO.ST question.

2. Confirm a cheapness call with capital, not commentary

1. When you think a quality name is oversold, look for three independent parties putting money in at today's price.
2. Rank them by information: an executive buying on the open market, then the company itself repurchasing, then a specialist investor with a matching style adding.
3. Require the purchases to be open-market and recent — grants, options exercises and scheduled plans carry no signal.
4. Treat the absence of any such buying as the more informative case.

• Buyback authorisations counted as purchases. An authorisation is permission, not execution — check the shares actually retired.
• Following a respected investor without knowing their position size or holding period.

3. When a cheap company is not buying back stock, find out why before concluding anything

1. Notice the absence — a founder-run business that calls its own shares undervalued and repurchases none is a contradiction worth resolving.
2. Look for management's stated reason in a transcript or meeting, rather than inferring one.
3. Distinguish the three possibilities: no capital, better uses for the capital, or no real conviction.
4. "Better uses" is only credible if the alternative deployment is visible in the accounts — acquisitions completed, not a pipeline described.

• A founder who is capital-constrained at the company level and also personally leveraged against his own shares — the combination that blows up in April 2026.
• "Overwhelmed with opportunities" as a permanent condition. It should show up as completed deals within a year or two.

4. Separate a slowing metric from a slowing business — then check the aggregate

1. Identify the single metric the market is reacting to, and define exactly what it excludes.
2. Find the aggregate figure that includes what the metric leaves out, and see whether the total is still growing.
3. Decide which of the two is the real driver of value for this business model.
4. If the excluded part is doing the work, state the condition under which the market's metric would start to matter.

• A store-opening programme that ends. Once the country is covered, like-for-like becomes the only metric there is.
• A company-level case that has quietly become a country-level one. The Dino defence rests on Polish convergence, which is a macro exposure.

5. Answer a disruption narrative with a mechanism, not with a denial

1. State the bear thesis in its strongest form and identify what would have to be true for it to hold.
2. Locate the part of the business the technology actually touches, and the part it does not.
3. Test the untouched part for the source of pricing power — switching cost, domain expertise, relationship, regulation.
4. Note whether the company itself is treating the risk as live. A dedicated management call is evidence about the threat's seriousness, not about its outcome.

• An assertion standing in for an argument. "I personally don't believe" is a position, not evidence — hold the author to the version with the mechanism attached.
• The same reasoning being applied selectively. It defends Topicus and Constellation, and is not applied to the software names outside the book.

6. Read a reader survey as two different experiments, and report both

1. Rank the responses by popularity and by outcome, and publish both cuts.
2. Compare what the crowd believes with what actually paid — they are usually different lists drawn from the same people.
3. Check whether the most-held conviction did well. Conviction intensity is the variable most likely to be wrong.
4. Keep the survey year on year, so the same names can be scored across a full cycle rather than one December.

• Headline figures that do not reconcile between the two write-ups (+17.1% vs +16% for the index; +19.0% vs +17.3% for readers). Different cuts of the same year — neither number travels on its own.
• A survey whose respondents are subscribers of the person publishing it. The "wisdom of crowds" framing assumes independence the sample does not have.