How to Write an Evidence-Led Finance Research Note

An evidence-led note answers the question in its first sixty words and supports the answer afterwards, in the order Barbara Minto set out as the Pyramid Principle in 1978. Build it in seven steps: define an answerable question, log every source, grade each claim, compute in one traceable file, draft answer first, review against the log, and record what would change the answer.

A note is not a record of the work. The note is the answer arranged for somebody who will never do the work, and a record and an answer are two different documents with two different orders. Discovery is untidy, and arrives in whatever sequence the post and the telephone deliver it, so the working file runs in the order things were discovered. A person making a decision needs the answer first and the road to it afterwards, and the note runs in that order. The difference between the order of discovery and the order of use is what all seven steps below are arranged around. Barbara Minto set the second order out as the Pyramid Principle in 1978, and every rule of answer-first orderPutting the conclusion in the opening lines and the support after it. below traces back to her.

In what order do the seven steps actually run?

The seven steps run in the order below, once, forward. Two of them invite reordering in particular. Gathering feels like progress and defining feels like delay, so the first temptation is to gather documents before writing the question down. The second is to leave the source log until the end, when the work is nearly finished and the list can be assembled tidily in one sitting. Both temptations cost far more than they save. A writer needs to tell whether a step is finished or has merely taken time, and each section below says what its step produces.

One thing governs the whole list. The order of writing is not the order the reader meets. The reader opens the note, meets the answer, meets the three reasons behind it, and meets the last line. Each of those three reasons is a supportA reason the answer holds, stated so it can be checked on its own., and a reader takes them in almost exactly the reverse of the order in which they were produced. Two of the seven steps leave no visible line in the finished note at all, and they are still the two that decide whether the rest of it is worth reading.

1. Define the question so that it can be answeredOne sentence carrying a measure, a period and a threshold. Written down before anything at all is opened.
   Finished when: the result that would end the work can be named.
2. Open the source log first, and log each item as it arrivesEvery item gets its line the moment it arrives, whatever is later thought of it.
   Finished when: nothing in the working file is missing from the log.
3. Grade every claim at the moment it is writtenFact, assumption, inference or scenario, marked on the sentence as the sentence is made.
   Finished when: no sentence in the draft sits there ungraded.
4. Compute in one file, with the inputs on their own tabEvery raw number sits on the input tab beside its item number, and the formulas point at that tab.
   Finished when: no number is typed inside a formula anywhere.
5. Draft the answer first, in sixty wordsWrite those sixty words before a single line of support, and resist every urge to soften them.
   Finished when: the answer stands alone and reads as an answer.
6. Read back against the log, not against the draftStart at the log, take each item, and find where it is used. Then take each number and trace it home.
   Finished when: every number reaches an item and every item is accounted for.
7. Close with what would change this answerName the things the answer stands on, and say how far it moves if any one of them turns out wrong.
   Finished when: the last line makes the answer conditional in the open.

Step one: what makes a question answerable?

Three things go into one sentence: a measure, a period and a threshold. Nothing else counts as done. On 3 March a worry arrived at the Kavery research desk, not a question. Latha Menon asked whether the desk should be worried about Meenakshi Tubes, and Sharada Iyer spent forty five minutes turning that into something a person could finish. She wrote down this: has the operating margin of Meenakshi Tubes stayed above the 11 per cent floor the credit policy of this desk sets, in the year just reported, and what single thing could move it most.

Read the rewritten version again and notice what it now contains. The measure is the operating margin. The period is the year just reported. The threshold is 11 per cent. A question carrying all three has a stopping point built into it, and a question missing any one of them will run until somebody loses patience. A household choosing a school meets the same difference. Is this school any good has no end to it, and nine schools can be visited with the household still undecided. Will the fees fit inside twelve thousand rupees a month, for this coming year, at a school within four kilometres, can be answered on a Saturday.

Step two: when does the source log get opened?

Before the first document is opened, and not one minute later. Timing is the whole of this step. Open the log, then start work, and give every item its line as it arrives rather than at the end. Opening the log first is what makes it a record of everything looked at, and opening it last turns it into a list of whatever happened to be used. Sharada Iyer opened hers on the morning of 3 March with nothing in it at all, and it closed on 11 March carrying fourteen items.

Nothing else about the log belongs to this step. The columns of the log, the marking of one item as standing behind another, and the handling of an item that a better one later beats are all set out under building a source log for financial research, and none of them changes the timing. Step two needs two things only: the log open before the work starts, and nothing touched along the way skipping it.

Step three: when is each claim graded?

At the moment of writing, in the same keystroke, never afterwards. Every sentence in a working draft is one of four things, and which one is known at the moment it is typed. Sharada Iyer marked hers as she went. The claim that revenue rose 15.2 per cent came off the two filed statements, and she marked it as a fact. No filing anywhere states that realisation per unit fell about 4.8 per cent. She worked the fall out herself from a volume series set beside the revenue growth, and marked it as an inference. The sixty forty split of total cost between variable and fixed was marked as an assumption, and the split is row four of the register. The margin of 5.0 per cent that follows if one order does not repeat was marked as a constructed scenario.

Attach the log item numberThe identifier that ties one sentence in the note to one item in the source log. in the same movement. The sentence and its number travel together from that point on, through the draft, through the review, and into the finished note. Grading afterwards means reconstructing one's own reasoning from memory, and memory is generous: it upgrades assumptions into facts and inferences into things somebody else supplied. Two days after the fact the writer genuinely remembers reading that the split was sixty forty. Nobody filed it. The writer chose it.

Step four: what makes a computation file traceable?

One file, one input tab, and no number typed inside a formula. Three rules are the whole of it. A traceable fileA computation file where every input can be followed back to its source. keeps every raw figure on a tab of its own, each on its own row, each beside the item number it came from. Every other tab in the file points at that one. When somebody later asks where 21,20,00,000 came from, the answer is a row, and the row names the item, and the item names the filing.

Here is the input tab from the assignment, cut down to the rows the conclusion actually used.

The value of the tab is that row five points at rows three and one, so the pairing of profit with revenue is visible on the face of the sheet instead of buried inside a cell. The visible pairing is what let the review pass in step six catch a first draft that had divided the profit of the year just reported by the revenue of the prior year.

Step five: what goes in the first sixty words?

Four things go in, and everything else waits. The measure, the period, the answer, and the comparison that makes the answer mean something. Here are the sixty words the note actually carried, exactly as they were written: the operating margin of Meenakshi Tubes for the year just reported is 10.8 per cent. The margin is below the 11 per cent floor in the credit policy of this desk, and down from 12.0 per cent in the prior year. Revenue grew 15.2 per cent in the same period. Three supports follow, and one assumption could move the figure far.

Notice what is not there. No description of the business and its trade. No paragraph on why the sector has been difficult. No account of how the work was done, how many people were spoken to, or how long it took. A reader who has the answer will read the rest, and a reader still hunting for it will stop. Every one of those belongs somewhere in the note, and none of them belongs in the first sixty words. The sixty words come before anything else, written on the day the number arrives rather than the day the prose is polished.

Now put the same finding into the other order and watch what happens to the reader. Two paragraphs on the business and its trade, one on why the sector matters, and the figure arriving in paragraph four. The words are the same words. A reader stops at a different place in each, so the document is not the same document. Answer-first and background-first are not two styles of the same note, they are two notes that get read to completely different depths.

Step six: what is the draft checked against?

The log, not the draft. Checking against the log is the single instruction in the whole sequence that people reverse most often, and reversing it makes a check feel thorough while finding nothing. A review passA separate reading whose only job is to find what the writer cannot see. that starts at the top of the draft and reads down is testing whether the note is convincing. A review pass that starts at the log and works into the note is testing whether the note is true. The two tests are different, and only the second one catches anything.

Prakash Nadar ran four checks on the assignment, in this order. Trace every number in the note back to an item in the log. Check that every claim carries an item number. Check every axis and every unit on every chart. Read the answer on its own, with the supports covered, and see whether it still stands up. The first check found a transcription that had divided the operating profit of the year just reported by the revenue of the prior year, giving 12.5 per cent. A figure of 12.5 per cent reads as a rise, where the correct 10.8 per cent is a fall. The second found one sentence carrying no item number at all. The third found a chart whose axis began at 9 per cent, so a movement of about 1.2 points filled the entire frame. The fourth passed.

Step seven: what is the last line of the note about?

The last line is always about the same thing: what would change this answer. Not a summary, not a restatement, not a flourish. The last line names the things the answer is standing on and says how far the answer moves if one of them turns out wrong. Every note ends this way, whatever it is about, and that sameness is the point. A reader who knows the last line is always the same kind of line goes looking for it.

The assignment closed on three of them. The filed accounts are final. The December order repeats. Input cost stays inside the range of the two reported years. Each is a load-bearing assumptionOne that changes the conclusion if it is removed., meaning the conclusion changes if any one of the three is dropped, and the note said so plainly. Then it put a number on the middle one. Of the Rs 21,20,00,000 filed, one order produced Rs 2,96,80,000, or 14 per cent of the revenue of the year. Take it out and revenue becomes Rs 18,23,20,000. Hold fixed cost at Rs 7,56,00,000 and scale variable cost with revenue to Rs 9,75,24,000. Operating profit is then Rs 91,96,000, a margin of 5.0 per cent. The closing line turned a single figure of 10.8 per cent into a stated range running down to 5.0 per cent, and the sixty forty cost split it rests on is assumption row four, chosen by the desk and filed by nobody.

How long should the note be, and what decides that?

The finished note ran to 1,100 words on a single sheet. Nothing about the size of the work decided that. Nine hours of production went into it and twenty one documents sat behind it, and none of that appears in the length. The number of supports the answer needs is what decides the length of a note, and three supports at roughly three hundred words each, plus sixty words of answer and a closing line, land at about eleven hundred words every time.

The instinct runs the other way. A long assignment feels as though it deserves a long note, and a note that is short after a hard week feels like an insult to the week. It is not. The week bought the right to be brief. An answer that genuinely needs six supports takes six and runs to two thousand words. An answer that needs one takes one and finishes in four hundred. Length follows the argument, never the effort behind it, and a note padded to look proportionate to the work is a note whose reader now has to find the argument inside the padding.

How long did the whole assignment actually take?

Nine hours of production and a review pass of forty minutes. Defining the question took 45 minutes. Gathering took 4 hours. Computing took 2 hours. Drafting took 2 hours and 15 minutes. Grading happened inside the drafting rather than as a block of its own, and the closing line was written inside the last stretch of it. The review pass added 40 minutes on top, under 7 per cent of the 9 hours and 40 minutes the whole thing took. Those 40 minutes caught the one error that would have reversed the answer.

Gathering took nearly twice what drafting took. Almost nobody plans an assignment like this that way round. The usual plan gives the writing most of the time and the gathering whatever is left, and then the gathering runs short and numbers start arriving from recollection rather than from filings. The plan works the other way round: the hours for gathering are booked first, two hours are taken for computing, and the drafting has what it needs at the end. The sixty words are already written and the supports are already graded, so the drafting needs less time than feared.

Who else writes a note in exactly this shape?

Almost everybody who has to hand a judgement to somebody who will not repeat the work. A credit file has to show later what was known on the day rather than what turned out to be true, so a lender writes it as a credit file entry. An analyst covering a listed issuer writes it as a published note, where the first screen is the whole product and the rest is the defence of it. The people in a fund committee room have read four papers before this one and will read three after it, so a committee paper for a fund runs on the same skeleton.

The shape works at household scale too, and household scale is the honest test of whether a structure is real or professional decoration. A person deciding whether to take a loan for a wedding can write this on a single sheet: the answer in sixty words, three supports, and a last line saying what would change it. Writing down what would change a decision converts it into something that can be revisited without arguing about it again from the start, so the last step is the one that transfers most cleanly to a household. Six months later the question is not whether the decision was right. The question is whether anything on that last line has actually happened.

What happens when the answer arrives in paragraph four?

The Kavery research desk, Kavery Capital Services Private Limited, Meenakshi Tubes Private Limited, Sharada Iyer, Prakash Nadar and Latha Menon are invented.
Educational material. Not advice on any investment, tax, budget or market position.