Actionable insights — Fair Isaac: a 114-page deep dive

1. Restate every forward multiple for stock-based compensation before calling anything cheap

1. Find stock-based compensation as a percentage of net income. Anything above roughly 10% needs the adjustment; above 20% it dominates the answer.
2. Deduct the full SBC charge from earnings and recompute the forward multiple on the adjusted number.
3. Compare the adjusted multiple, not the headline one, to the company's own history — because the history needs the same adjustment.
4. Apply this consistently across every candidate, not just the ones you are inclined to reject.

• Buybacks that only offset dilution. A shrinking share count alongside heavy SBC means the cash return is smaller than it looks — FICO reports both a 30% decade-long share reduction and a 22% SBC bill.
• Your own selective application. The adjustment is only a discipline if it also disqualifies names you like.

2. Turn every "no" into a price

1. When you decline a business you rate highly, state the multiple at which you would own it — on the adjusted earnings, not the headline.
2. Convert that multiple into a share price and publish it.
3. Record the gap between that price and the current one, so the size of the required move is explicit.
4. Set an alert at the price rather than revisiting the name on a schedule.

• A named price that quietly moves. The value of publishing $901 is that it can be checked later.
• Prices set against forward earnings that themselves fall. A 25x target on shrinking estimates is a moving target.

3. Ask whether AI substitutes for the product or lowers the barrier to competing with it

1. Separate the two AI risks explicitly: does the technology replace the customer's need, or does it make a rival cheaper to build?
2. For data and analytics businesses, the second is usually the live risk — the moat is the cost of building a credible alternative model.
3. Identify who benefits from that cost falling, and whether they have distribution.
4. Look for the regulatory or institutional gate that has been keeping the rival out, and monitor that gate rather than the technology.

• Moats made of habit. "Institutional inertia" is real and also the first thing a regulator can remove with one decision.
• Management responses that describe activity rather than defence — FICO 10T, direct licensing and cloud investment are recorded here without being assessed.

4. Score quality and decide on price as two separate outputs

1. Complete the quality assessment to a number before opening the valuation work.
2. Publish the quality verdict even when the answer to the buy question is no, so the two do not contaminate each other.
3. Where the two disagree, say which one is binding and why.
4. Keep the high-quality declines on a watchlist, because the quality work does not need redoing when the price moves.

• A quality score that has already absorbed the price. If the same name's score moves when the share price moves, the two axes have merged.
• The gap between a published ranking and the portfolio decision — FICO is Best Buy #4 on 7 June and still not owned.

5. Screen for tollbooths: a mandatory fee on a transaction you do not have to fund

1. Look for a business paid per transaction rather than per unit of its own capital — a fee on someone else's activity.
2. Establish its share of the transactions it touches; a genuine tollbooth appears in the overwhelming majority.
3. Check the fee is small relative to the total cost of the transaction, which is what makes it hard to dislodge.
4. Confirm the business is capital-light, so growth in transaction volume drops through to cash rather than requiring reinvestment.

• A tollbooth whose road can be re-routed by a regulator. That is precisely the FHFA/VantageScore situation.
• Share statistics quoted from the company's own investor presentation, as these are.

6. Consume outside research by keeping the decision layer in-house

1. Use third-party deep dives for the descriptive work — business model, history, competitive landscape — where breadth is expensive and errors are visible.
2. Never outsource the two things that determine the outcome: the adjustments you make to the reported numbers, and the price at which you would act.
3. Re-derive at least one headline figure yourself as a sanity check on the source.
4. Attribute the source plainly, so a reader can weigh it.

• Length taken as rigour. A 114-page case and a two-line objection can both be right, and here the two lines carry the decision.
• Guest research where the author's own position is not disclosed.