The Earnings Teardown Playbook: find the line the headline is hiding, in 20 minutes

A company reports. The headline says beat on the top and bottom line. The stock falls eight percent. Somebody on television calls it an overreaction.

It almost never is. What happened is that the press release was written to be read in ninety seconds and the filing was written to be legally complete, and the number that moved the stock lives in the second one. The gap between those two documents is where the entire game is played — and closing it is a repeatable, roughly twenty-minute job that has nothing to do with being a finance professional.

This is that job, written down. Six passes, a time budget for each, a scorecard, and the failure modes.

One caveat up front, and we mean it: this is a reading method, not investment advice. We are not your financial adviser. Everything below helps you understand what a company said. What you do about it is yours.

Before you start: get the right two documents

You need the press release (fast, promotional, has the headline numbers) and the earnings presentation or 10-Q (slow, complete, has the tables). Open both. If a company publishes only the release and makes you wait for the filing, that is itself information — note it and move on.

Set a timer. The discipline of twenty minutes is what stops you reading the CEO’s adjectives.

Pass 1 — The scoreboard (2 minutes)

Write down four things and nothing else:

• Revenue: reported vs. consensus
• EPS: reported vs. consensus
• Guidance: raised, held, or cut
• What the stock did in the first hour

That last one is your treasure map. If the first three are good and the fourth is bad, something in the filing contradicts the release, and the remaining eighteen minutes have a specific job: find the contradiction. If everything agrees, you are doing maintenance reading, which is faster and less interesting.

Pass 2 — The growth-quality check (4 minutes)

This is the single highest-yield pass, and almost nobody does it. Put three growth rates side by side:

1. Revenue growth, year over year
2. Gross profit growth, year over year
3. Operating income growth, year over year

Healthy compounding looks like an ascending staircase: gross profit growing at least as fast as revenue, operating income faster still. When the staircase inverts — revenue growing faster than gross profit — the company is buying its growth. Every incremental dollar of sales is arriving with more cost attached than the last one.

Worked example, this week: Shopify grew revenue 34% and gross profit 31%, and guided the third quarter to revenue growth in the low thirties against gross profit growth in the mid-to-high twenties. The gap isn’t a rounding error, it’s a widening design. More volume crossing the platform, less of each dollar staying.

Do the same arithmetic on the guide, not just the quarter. Companies signal the inversion a quarter before it shows up.

Pass 3 — The segment split (3 minutes)

Consolidated numbers are an average, and averages hide the thing you want. Find the segment table — it is usually four to six pages into the presentation — and ask one question: does the biggest segment behave like the total?

If the total grew 24% and the segment that is 60% of the business grew 4%, then something small and probably lower-quality is carrying the print.

Worked example: Uber’s gross bookings rose 24% and beat consensus. Inside the segment table, Mobility — the core business — grew bookings 22% while Mobility revenue grew 1%. Volume up, take rate down. That single row is why the stock fell on a beat.

Rule of thumb: when a bookings/volume figure and a revenue figure for the same segment diverge by more than about ten points, the company is either discounting, changing accounting, or mixing toward a lower-margin format. The call will tell you which. Usually only if someone asks.

Pass 4 — The cash lines (3 minutes)

Three numbers, all found in the cash flow statement, none in the press release headline:

• Capital expenditure — compare to the same quarter last year and to what analysts modeled. A capex line that doubles unannounced is a strategy change disclosed in a table.
• Free cash flow — operating cash flow minus capex. If net income is up and FCF is down, find out why before you believe the net income.
• Stock-based compensation — as a percentage of revenue. Rising SBC with flat headcount means you are paying more for the same company.

Worked example: AMD beat on every reported line and fell about 8%. Capital expenditure came in at $808 million against roughly $299 million modeled. Nobody had that in a spreadsheet, and it re-dated every free-cash-flow forecast on the street by a year.

Pass 5 — The guide, decoded (4 minutes)

Guidance is the only forward-looking thing in the document, so it carries most of the weight. Three checks:

1. Rate, not level. Convert the guided range into a growth rate and compare it to the rate just reported. “Raised guidance” is compatible with decelerating growth, and usually is.
2. Width. Compare the width of the range to last quarter’s. A range that widens is management telling you they have less visibility, in the only language they are allowed to use.
3. Which lines got guided. If a company guided revenue but stopped guiding margin, that is the answer to a question you were about to ask.

Pass 6 — Share count and the adjusted bridge (4 minutes)

EPS is a fraction, and the denominator is a choice. Check diluted share count against the year-ago quarter. If EPS grew 12% and the share count fell 9%, the business grew about 3%.

Then find the reconciliation table — every company that reports an “adjusted” figure must show the bridge from GAAP. Read the add-backs. One-time restructuring charges that appear for six consecutive quarters are not one-time; they are operating expenses wearing a costume.

The nine-line scorecard

Paste this into a note. Fill it in as you go. Nine lines, one quarter, permanent record — and when you do it for four quarters in a row on the same company, the trend does the analysis for you.

1. Revenue growth % → ____
2. Gross profit growth % → ____  (lower than line 1? flag)
3. Operating income growth % → ____
4. Largest segment growth % vs. total → ____ / ____
5. Capex, this quarter vs. year ago → ____ / ____
6. Free cash flow, this quarter vs. year ago → ____ / ____
7. Guided growth rate vs. reported growth rate → ____ / ____
8. Diluted share count change % → ____
9. Biggest single add-back in the adjusted bridge → ____

Three or more flags and the headline is not the story. Zero flags and a falling stock means the story is macro, sector, or positioning — not the company.

The AI prompt that does the grunt work

Extraction is mechanical; judgment is not. Hand a model the filing and this prompt, then check its arithmetic against the source before you trust a single figure:

The instruction to write NOT DISCLOSED is the important half. What a company declines to break out is frequently the most informative thing in the release, and a model asked to be helpful will otherwise estimate it and hand you a number that does not exist.

Six ways people get this wrong

• Reading the release and skipping the tables. The release is marketing with a legal review. The tables are the product.
• Anchoring on beat/miss. Consensus is a bar set by people who also read only the release. Beating it is a fact about the bar.
• Treating “record quarter” as information. Any company growing at all has a record quarter every quarter.
• Ignoring the year-ago comparison. A number is not big or small; it is bigger or smaller than the same number twelve months ago.
• Believing the first hour. The move at 4:05 p.m. is an algorithm reading a headline. The move over three sessions is people who read the tables.
• Doing this once. One teardown tells you about a quarter. Four consecutive teardowns tell you about a company, and that is the only version that compounds.

Run it on a schedule

Pick three companies you actually care about — an employer, a customer, a holding — and put their report dates in your calendar with a twenty-minute block attached. Three companies, four quarters, twelve blocks a year, four hours total. That is the entire commitment, and it is the difference between having opinions about markets and having evidence.

Pair it with the fifteen-minute market review for the weekly cadence, and the verification trust ladder for deciding how much of the extraction you let a model do unchecked.

The line is always in there. It is just never in the headline — because if it were, it would not be worth twenty minutes.