A Repeatable Research Process
5 steps · one page
In short
The single most useful property of a research process is that it is the same every time.
Scope. This article describes an order of work for reading a company, and why the order matters more than the contents of any single step. It does not tell a reader what to conclude, what to look for as a good or bad sign, or what to do afterwards. No threshold appears anywhere. Filing descriptions are United States federal filings, checked 14 August 2026; other markets have equivalents with different names, contents and timing.
Not that it is thorough, not that it is clever — that it is repeatable, because a process that varies with the company cannot be compared across companies, and a process that varies with the mood of the person running it cannot be compared with itself.
That is a claim about consistency, not about being right, and it is worth separating the two at the outset. A consistent process applied to a company produces a record of what was examined and what was found. It does not produce a correct answer, and no arrangement of steps will. What it produces is the ability to tell, later, whether a mistake was a failure of process or a failure of judgement — and those have completely different remedies.
The order, and why it is that order
Understand the business before touching a number. What does the company sell, to whom, and how does the money arrive? Is a sale one-off or recurring? Who else does this? The reason this comes first is not thoroughness — it is that numbers read without a business model attached are unreadable in a specific and dangerous way: they look interpretable. A gross margin of 40% means one thing for a manufacturer and something else entirely for a distributor, and nothing at all in isolation.
Then the statements, in a fixed sequence. The income statement for what was earned and how the expense lines moved against revenue; the balance sheet for what is owned and owed; the cash-flow statement for whether the earnings arrived in cash.
Then the footnotes, which is where the policies behind the numbers live, and where a figure that looked simple frequently stops being simple.
Then management's own account — the discussion and analysis, and the call — read last on purpose. Management's narrative is the most persuasive material available and the most likely to install a frame before a reader has one of their own. Read after forming a view, it can be checked against that view. Read first, it becomes the view.
Then the record of what was found, written down. Including the questions that were not answered.
The step that gets dropped is the last one, and it is the one that makes the process a process. Research that is not written down cannot be revisited, cannot be compared against what actually happened, and cannot distinguish between a view that was well reasoned and unlucky and one that was badly reasoned and lucky. A written record is also the only defence against the memory rewriting itself — the recollection of what was expected is reliably closer to what occurred than the expectation actually was. This is why the unresolved questions are recorded as prominently as the findings: a list of what a reader could not determine is a more honest description of their position than a list of what they could.
The derived measures worth computing, and the point of computing them
A small number of ratios can be computed directly from the statements without judgement, and the reason to compute them is that they make companies comparable. Using the portal's canonical fictional company, whose statements are set out in full on the Wexford Instruments page, the arithmetic runs as follows.
| Measure | Computation from the disclosed figures | Result |
|---|---|---|
| Gross · operating · EBITDA margin | 400.0 / 100.0 / 150.0 over revenue 1,000.0 | 40.0% · 10.0% · 15.0% |
| Days sales outstanding | receivables 150.0 / revenue 1,000.0 × 365 | 54.8 days |
| Days inventory outstanding | inventory 130.0 / cost of sales 600.0 × 365 | 79.1 days |
| Current ratio | current assets 346.3 / current liabilities 200.0 | 1.73 |
| Net debt · net debt to EBITDA | debt 340.0 less cash 66.3, over EBITDA 150.0 | 273.7 · 1.82x |
| Interest cover | operating profit 100.0 / interest 18.0 | 5.56x |
| Cash conversion | operating cash flow 98.3 / net income 62.3 | 1.58x |
| Free cash flow | operating cash flow 98.3 less capital expenditure 78.0 | 20.3 |
Worked example
Worked example — what eight numbers took, and what they are for. Every figure above came from three statements and required no assumption, no estimate and no view. Working-capital days and turnover use closing balances; return measures use the average of opening and closing — the convention fixed on the Pillar 24 hub and applied identically everywhere in Group IV, because a ratio computed on a different basis from the one it is compared against is not a comparison. Note what these measures do and do not do. They make this company comparable with itself over time and with others computed the same way. They do not say whether any of them is good. Cash conversion of 1.58x says operating cash flow exceeded net income by 58%; whether that is reassuring depends on why, and the why is in the cash-flow statement and the footnotes rather than in the ratio. MarketClue publishes no thresholds for any of these measures, here or anywhere.
Three failure modes the order is designed against
Starting from the conclusion. Research begun after a view has formed becomes a search for support, and the material is abundant enough that the search always succeeds. The countermeasure is not willpower — it is writing down what would have to be true for the view to be wrong, before reading further, so that the disconfirming material has somewhere to land.
Depth in the wrong place. It is possible to spend a week on a segment disclosure and never establish how the company gets paid. Time spent is not evidence of progress, and the fixed order exists partly to ensure the cheap, high-value reading happens before the expensive, low-value reading becomes tempting.
Recency. The most recent filing is the most available and the least informative about how a business behaves over a cycle. A process that reads three years of statements as a matter of course is not more diligent than one that reads the latest; it is asking a different and better-posed question.
Worked example
What a process cannot do, stated plainly so the rest of this pillar is read correctly. A repeatable process improves the consistency and the auditability of research. It does not confer accuracy, and there is no evidence that a more elaborate process produces better outcomes than a simpler one. The literature covered in Pillar 25's article on quantitative analysis is directly relevant here: more inputs bought no out-of-sample improvement in that setting, and a research checklist is not exempt from the same arithmetic. The honest case for a process is that it makes error diagnosable, not that it makes error less likely.
Frequently asked
8 questions
What makes a research process useful?
That it is the same every time. A process that varies with the company cannot be compared across companies, and one that varies with the researcher's mood cannot be compared with itself. Consistency is the property; accuracy is not something any process confers.
Why read the business model before the numbers?
Because numbers read without a business model attached look interpretable when they are not. A 40% gross margin means different things for a manufacturer and a distributor, and nothing at all in isolation.
Why read management's commentary last?
Because it is the most persuasive material in the filing. Read after forming a view, it can be checked against that view. Read first, it becomes the view.
Which ratios are worth computing every time?
The ones that need no assumptions: margins, days sales outstanding, days inventory outstanding, the current ratio, net debt and net debt to EBITDA, interest cover, cash conversion and free cash flow. They make a company comparable with itself over time and with others computed identically.
Does a good ratio mean a good company?
No, and this portal publishes no thresholds for any measure. A ratio tells you what happened; whether it is reassuring depends on why, and the why is in the statements and footnotes rather than in the ratio.
Why write down the questions you could not answer?
Because a list of what you could not determine describes your position more honestly than a list of what you could, and because the unresolved items are what a later filing can actually resolve.
How far back should the statements be read?
Further than the latest filing. The most recent report is the most available and the least informative about how a business behaves over a cycle; reading several years asks a better-posed question rather than simply a more diligent one.
Will a more elaborate process produce better results?
There is no evidence that it does. More inputs bought no out-of-sample improvement in the settings where this has been measured, and a research checklist is not exempt from that arithmetic. The case for a process is that it makes error diagnosable, not less likely.
References
- Investor.gov (SEC) — How to Read a 10-K —
- SEC — Investor Bulletin: How to Read a 10-K —
- Investor.gov (SEC) — How to Read Financial Statements —
Educational and informational only — not investment advice, a recommendation, or an offer to buy or sell any security. Investing involves risk, including the possible loss of principal. Worked examples use fictional companies and figures.