11.2 · The research process¶

Why this matters: research without a process either never finishes (there is always another note to read) or finishes too early (the story was convincing). A stage-gated process fixes both: each stage has a time budget, a deliverable and a question that decides whether the next stage is worth the time. This lesson gives you the three stages — kill test, triage, deep dive — the checklist that runs through them, and the research log that makes the work cumulative.

Learning objectives — after this lesson you can:

Run a 30-minute kill test, a 3-hour triage and a 2–3-week deep dive, each with defined outputs and gate questions.
Prioritise primary sources and sequence the reading.
Keep a research log that supports later review and compliance.
Manage time, avoid research-as-procrastination, and know when to stop.

Prerequisites: 11.1 Idea generation, 03.2 Anatomy of an annual report · Time: ~60 min

1. Stage gates¶

Stage	Budget	Inputs	Deliverable	Gate question
0. Queue	5 min	The idea and its source	One line in the queue	Is there a plausible reason for mispricing, inside (or near) my circle?
1. Kill test	30 min	Screener page; `tools/fi` dashboard; shareholding pattern; last annual report's auditor's report	A one-page kill-test sheet (business in three sentences; 5-year numbers; people; forensic quick scores; valuation snapshot; mispricing reason)	Any disqualifier? If not, what is the one question triage must answer?
2. Triage	3 hours	Latest annual report (MD&A, auditor's report, RPT note, contingent liabilities, borrowings, receivables ageing); last four concall transcripts; latest rating rationale; two competitors' latest calls	Triage note: unit economics and KPIs; industry position; say-do table; first reverse DCF; a one-sentence variant perception with the evidence that would refute it	Can I state a variant perception, and is the payoff asymmetric enough to justify three weeks?
3. Deep dive	2–3 weeks	Five annual reports; all concalls; DRHP if recent; industry data; scuttlebutt; the model	Full case file; three-statement model with scenarios; forensic checklist; governance checklist; scuttlebutt log; valuation triangulated; catalysts; pre-mortem; thesis-breakers	Do the numbers, the people and the price all pass? What specifically breaks the thesis, and by when would I know?
4. Memo & decision	1 day	The case file	One-page thesis; sizing; journal entry	Would I hold this at 2× the size for three years without new information? If not, why is it a position at all?

The disqualifiers at the kill test (from 09.7 and 05.6): a Beneish flag without explanation; CFO/PAT under 60% for three years (non-lender); pledge over 20%; unexplained auditor resignation; substantive SEBI action against promoters; a business you cannot explain in three sentences; no plausible mispricing reason. Most ideas die here in twenty minutes, and that is the process working.

2. Sequencing the reading¶

Primary sources first, in this order — because each step tells you what to look for in the next:

Annual report, back to front: auditor's report (opinion, KAMs, CARO), then the notes that matter (RPTs, contingent liabilities, borrowings, receivables ageing, revenue policy), then the financial statements, then MD&A, then the chairman's letter last (so the narrative does not frame the numbers).
Concall transcripts, oldest to newest for the last two years: build the say-do table as you go (05.5); note what analysts asked and what was dodged.
Rating rationale: liquidity, covenants, key sensitivities (04.5).
DRHP if listed within five years: KPIs, risk factors, related parties, pre-IPO history.
Competitors' concalls and reports: the same market described by people with different incentives.
Industry data (05.7): the regulator's statistics, trade bodies, tender portals.
Sell-side reports last: to learn the consensus and its arguments, once your own view exists (03.5).
Scuttlebutt: designed after the reading, because now you know which claims need testing.

3. The research checklist¶

A single page that travels with every case (the casebook's 01-full-case-study.md is built on it):

Section	Questions	Modules
Business	What does it sell, to whom, why; unit economics; value chain; concentration	05.1
Industry	Structure, forces, life cycle, capital cycle, regulation, substitutes	05.2, 05.4
Moat	Source; evidence in 10 years of ROIC, share, pricing; erosion risks	05.3
Numbers	Growth, margins, returns, working capital, leverage, quality; 10-year dashboard	04.1–04.7
Filings	Auditor's report, notes, policy changes, say-do table	03.2–03.4
Forensics	40-point checklist; scores; RAG and verdict	09.7
Governance	Promoter, RPTs, pledges, board, auditor, issuances, regulators	05.6, 09.5
Scuttlebutt	Plan, conversations, data checks, log	05.7
Valuation	DCF (scenarios, sensitivities), multiples vs peers/history, reverse DCF, expected value	06.3–06.8, 07.x
Thesis	Variant perception; catalysts; risks; pre-mortem; thesis-breakers with dates	11.3
Decision	Size; entry plan; monitoring KPIs; journal	11.4–11.6

4. The research log¶

A dated, append-only record of what you read, whom you spoke to, what you learned, what surprised you and what you expected — one file per company in the casebook. It serves three purposes: it makes the next quarter's update an increment rather than a restart; it lets you audit your own process after the outcome (was the thesis right for the right reasons?); and it is your compliance record — what you knew and when, and that nothing you learned was UPSI (05.7 §5).

Minimum entry: date · source (document/person type) · claim or fact · confidence · what it changes · next action.

5. Time management¶

Timebox each stage and stop when the budget ends; extend only with a written reason.
Kill fast. The cost of a wrong kill is a missed idea (there are thousands); the cost of a wrong keep is three weeks and, worse, an anchored view.
Research as procrastination: the tell is reading a fifteenth document without a changed conclusion. If the last three sources changed nothing, you are done — write the memo.
Parallelise cheaply: run kill tests on five ideas in an evening; run one deep dive at a time.
Diminishing returns are steep: the first annual report and four concalls give 80% of the information; the fifth annual report gives 5%. Spend the marginal hour on scuttlebutt or the model, not on the tenth note.
Calendar the process: results seasons (Apr–May, Jul–Aug, Oct–Nov, Jan–Feb) are for updates; the quiet months are for deep dives.

6. When to stop¶

Stop researching and decide when: the variant perception is stated with evidence and a refutation test; the valuation is triangulated and the price sits clearly on one side of the distribution (or clearly does not); the forensic and governance checklists are scored; the thesis-breakers are written with dates; and the last three sources changed nothing. Stop and kill when: a disqualifier appears at any stage; the variant perception cannot be stated; the payoff is symmetric; or the time budget has doubled without convergence.

Trader's lens

A stage-gated process is a stop-loss on research time. Traders know that the hardest discipline is cutting a position that has consumed effort; analysts face the same sunk-cost pull with a company they have spent three weeks on. The gate questions are pre-committed exits — decide the criteria before the work, and honour them.

India notes

Exchange announcement feeds (BSE/NSE corporate announcements, filterable by company) are the research log's raw timeline; download them at the start of a deep dive.
Concall transcripts are mandatory disclosures (LODR, 2023 amendments) and sit on the exchange sites and company IR pages; older ones are on aggregator sites.
The results calendar is public (companies announce board-meeting dates); build the monitoring schedule from it.

Common mistakes

Starting with the chairman's letter and the sell-side report.
Skipping the kill test because the idea came from someone credible.
Deep-diving three ideas at once and finishing none.
No research log — the same document read three times across two years.
Confusing "I have read everything" with "I have a view".

Key terms¶

Term	Meaning
Stage gate	A decision point with a predefined question that must be answered before more time is spent
Kill test	A 30-minute screen for disqualifiers and a mispricing reason
Triage	A 3-hour read to state a variant perception and payoff asymmetry
Deep dive	The full multi-week research producing the case file and model
Say-do table	Record of management guidance vs delivery
Research log	Dated record of sources, findings, confidence and next actions
Timebox	A fixed time budget for a task
Research-as-procrastination	Continuing to gather information after conclusions have stopped changing

Check your understanding¶

An idea passes the kill test with a "yes" on every disqualifier check but you cannot write the one question triage must answer. Proceed?

Answer

No. Without a specific question there is no hypothesis, and triage becomes open-ended reading. Return it to the queue with "no mispricing hypothesis" and move on.

Why read the auditor's report before the MD&A?

Answer

The auditor's report and notes are the least narrative-laden parts of the report; reading them first means the MD&A's story is tested against facts you already hold rather than framing how you read the facts.

Three weeks into a deep dive, the valuation is ambiguous (price near the probability-weighted value) and the thesis-breakers are clear. What is the output?

Answer

A watchlist entry with a target entry price, the monitoring KPIs and the thesis-breakers — a complete, valuable output. The research was not wasted; the decision is "not at this price".

Which four documents give most of the information for a triage, and in what order?

Answer

The latest annual report (auditor's report and key notes first), the last four concall transcripts (oldest first), the latest rating rationale, and two competitors' latest concalls.

Go deeper¶

The casebook's PROCESS.md — the same stages with the private-repo templates.
Michael Mauboussin, Think Twice — on process vs outcome and pre-commitment.
Atul Gawande, The Checklist Manifesto — why checklists work in expert domains.
15.1 The case-study method — the process at full length.

← Previous: 11.1 Idea generation · Module index · Next: 11.3 Writing an investment thesis & memo →