Actionable insights — How One Frustrated Investor Democratized Wall Street's Data

47:21 1. Build a per-company KPI template — the eight things you track for every name

1. Accept the real constraint first. Access to everything is not the same as insight: "you can drown in — I love that I have the ability to pull all the data, but… I also want to just pull up a view and be narrowly focused on what I think matters."
2. Write a fixed, short list — roughly eight items — of what you will track for every company you own, and make it your starting template for each new name rather than rebuilding a view from scratch each time.
3. Put company-specific operating metrics on the list, not just the headline financials. The metrics that decide the business (trips, transaction volume, cards in force, gross bookings, take rate) are usually disclosed but sit outside the three statements.
4. Keep the template stable across names so that the comparison between two companies is apples-to-apples, and so a change in one line is visible without hunting.
5. Treat everything outside the template as noise by default. The template's job is to make ignoring headlines a decision you already made, rather than one you have to make under pressure.
6. Expect the template to simplify you, not sophisticate you: "it's forced me a lot as an investor to really isolate what I think is important and what I should track and block out a lot of noise that you'll see in headlines."

• A template that has quietly grown to thirty lines; a holding whose template you have never actually written down; a metric on the template that the company has stopped reporting (see insight 4).

1:04:59 2. Cross-compare one KPI across the whole industry to test a narrative

1. State the narrative you believe out loud, in a form that can be falsified by a number: "I think company X is growing the fastest."
2. Pick the single operating metric the industry all disclose — gross bookings for travel platforms, transaction volume for card networks, whatever the sector's common unit is — rather than a financial-statement line that accounting choices distort.
3. Chart it for every significant peer on one axis, not just the name you own. Save the chart as a standing comparison you revisit, rather than a one-off.
4. Ask the three qualifying questions together, because growth rates alone lie: which is growing fastest, off which base, and at what take rate? A high growth rate off a small base at a collapsing take rate is not the same story.
5. Let the answer overrule the prior. When the chart disagrees with you, that is the output — not a reason to re-cut the chart.
6. Use it defensively too: when the one company you own tells you the industry is hard, check whether its peers are reporting the same conditions. If they are growing volumes and pushing price while your company cannot, the problem is company-specific.

• A peer whose growth diverges from the group in either direction; a company whose reported industry conditions do not match its peers'; a take rate falling while volume grows (share bought with price).

39:50 3. Take rate up and volume up — the rare combination worth acting on

1. For any platform or marketplace, isolate the two variables that usually trade off: volume (units, trips, transactions, bookings) and take rate (the share of each transaction the platform keeps).
2. Read them together over several quarters, not one. Volume up with take rate down is share bought with price. Take rate up with volume down is harvesting a declining asset. Neither is a signal.
3. Flag the rare case where both rise materially at once — pricing power and demand confirmed simultaneously: "take rates doubled while the trips tripled. That's a rare kind of KPI you see from a business, to be able to double their pricing power and usage triple."
4. Draw the profitability inference the market has not yet drawn. If a business with fixed-ish costs is taking a bigger cut of a much larger volume, margin follows: "Oh wow, this company's actually about to get really profitable."
5. Explicitly check the combination against the consensus story rather than the price. His trigger was worth acting on precisely because "the consensus on the street was that Uber's never going to make a dollar."
6. Be willing to reverse a public position on it. He had called the business a zero-rate artefact with terrible unit economics for two years before the data turned him.

• The combination appearing in any marketplace you follow; the reverse combination (take rate rising as volume stalls) in one you own; a company that changes how it defines take rate mid-series.

41:11 4. Put the exit trigger on the KPI — including its disappearance

1. Write the sell condition in the same units as the buy condition. If the thesis is a KPI, the exit is that KPI breaking — not a price level and not a headline.
2. Name the specific threat and say where it would first appear in the data. For a ride platform facing autonomous vehicles: "if they get disrupted by some autonomous vehicle threat, I'm going to see it there."
3. Add the second, less obvious trigger: the company ceasing to disclose the metric. Voluntary KPI disclosure is dropped far more often when it flatters management to drop it.
4. Treat non-disclosure as more urgent than a bad print, because it removes your ability to monitor at all: "or even worse, they stop reporting or stop disclosing the number… I would be very much so looking for the exit before other people are."
5. Automate the detection if you can, or diary it if you cannot — a discontinued line item is easy to miss quarter to quarter because nothing appears, and nothing is what you must notice.
6. Keep the trigger honest by writing it down while you still like the position, not while you are defending it.

• A KPI silently absent from this quarter's release or supplemental deck; a metric redefined or restated rather than continued; management moving a number from the press release into a footnote.

1:02:30 5. Screen to exclude before you screen to select

1. Reject the usual purpose of a screen. It is not a shopping list: "not because, oh, here's a list of things that I should invest in, but here's a list of things that I know I might be interested in."
2. Invert the filter — "it's really helpful to filter not what's in the screen, but filter out what's maybe not a good use of my time." The scarce resource is attention, not ideas.
3. Write down your own style constraint first, explicitly, so the exclusion is principled. His: growth at a reasonable price, long holding periods.
4. Convert the style into a hard exclusion rule. No steady history of topline growth means he will not be a long-term shareholder, "unless this is some special situation or spinout situation" — the carve-out is named, not implied.
5. Exclude whole sectors whose behaviour does not fit your style, regardless of individual merit — for him, metals and mining, because "junior mining… is just so boom bust, hard to make money. It kind of feels more like a lottery ticket unless you're in the industry."
6. Re-run the reduced screen often and spend the reclaimed time on fewer names: "a fresh list every time, and spend more time on fewer names that actually make sense."

• An exclusion rule that is really a bias — check whether the sector you exclude is genuinely mismatched to your style or merely unfamiliar; a "special situation" exception being used often enough that it has become the rule.

58:49 6. Audit the number back to the filing before you act on it

1. Treat every aggregated number as a claim until you have seen its source. Data vendors clean, standardise and consolidate — each step is a place a number can drift.
2. Before sizing a position on a figure, click or look it through to the primary document: the 10-Q, 10-K or equivalent for non-US filers.
3. Prioritise the numbers that look wrong. A figure that strikes you as implausible is either an error or an unusual fact, and both are worth the two minutes: Visa's September-25 quarter net revenue "almost looks fake. It is 40 billion. Not 40 billion 362. It is 40,000,000,000" — and the filing confirmed it.
4. Apply the same test to a big single-period move — "it's like the debt doubled or whatever it is" — before you build a story around it.
5. Understand why this is the trust mechanism rather than a nicety: "there's no such thing as a perfect data set in the world… trust is built over me showing you instead of me telling you." Provenance is what makes a number usable when you have to defend a decision.
6. Extend the check to non-US names, where standardisation error is most likely; he specifically names Canada, the UK and Europe as covered, and points at the asymmetry between a US company and "some micro cap in Taiwan" as the industry's remaining gap.

• A number you cannot trace to a document; two vendors disagreeing on the same line; a "standardised" metric whose definition differs from the company's own; suspiciously round figures that turn out to be either genuine or rounded — you cannot tell without the filing.

50:22 7. Check the score quarterly — not daily, not never

1. Set the review frequency to the reporting frequency. The KPIs update quarterly, so the decision cycle is quarterly: "the right amount of frequency to stay on top of the story every 3 months without being spoonfed it every single day so that you have decision paralysis."
2. Between reviews, route headlines through the template rather than through your position. If the news does not touch a template metric, no action is required and none should be taken.
3. Use the quarterly review to answer one dynamic question rather than a static one — the host's version, learned from training interns: not "is this good or bad" but "what's getting better, what's getting worse?"
4. Hold the time horizon separately from the review cadence. Reviewing quarterly is compatible with "holding them for, you know, decades potentially."
5. Recognise what the cadence is actually for. The scarce commodity is not ideas but the ability to hold them — "there's a shortage of the ability to hold them long enough. And it's not just patience, it's the conviction" — and a quarterly evidence check is what refills conviction in a drawdown.

• Finding yourself checking a position daily — usually a sign the thesis is not written in KPI terms; a quarter skipped on a holding you are comfortable with, which is where a discontinued disclosure hides.

35:34 8. Refuse the score — never outsource the verdict, only the gathering

1. Separate the two things a research process does: gathering and normalising information, which should be automated aggressively, and judging it, which should not.
2. Be suspicious of any tool that hands you a verdict — a quality score, a valuation rating, a buy/sell grade. Dennis refuses to build one even though customers ask: "I don't want to build a scoring system that tells you if a stock is good or bad or valued or undervalued."
3. Apply the same rule to general-purpose AI. He is blunt that "letting GPT manage your portfolios hasn't really been linked to good returns" — and note that this is the person selling the AI-native product saying it.
4. Where you do use a model, narrow its context to one document rather than the whole corpus: "I don't want to reach in the mind of the entire LLM. I want to reach into the mind of the LLM's extraction of this specific document." A narrower context is a more auditable answer.
5. Keep the judgment step in your own template and your own words, so the reasoning is inspectable later when the position is under pressure.
6. The user-feedback corollary, worth stealing for any process: "it's so important to listen to users. It's also so important to know when to ignore their recommendations."

• A screen or dashboard whose ranking you have started treating as a conclusion; a summary you cannot trace to the underlying document; a decision you could not explain without citing a score.