How to Read an Earnings Call Transcript, Step by Step

Reading an earnings call transcript starts before the transcript. The reader writes down and numbers what needs explaining first, and only then reads the prepared remarks for the frame management chose and the questions for what other readers could not get. The useful material is rarely a new figure. The signal is which question was asked twice, which answer described a number without giving it, and what was promised for next time.

Seven steps run in a fixed order. Each names where to look, what to write down and when to move on; the accounting items themselves, and what any answer means about the business, are covered separately. The seven end with a short sheet: the numbered questions on the left, and beside each one what the call actually gave back. Reading a transcript is a procedure applied to text rather than a relationship between two quantities, and the sheet is what the procedure produces.

A transcript is one of the very few documents an analyst reads that was built to be read in a particular sequence by someone who wanted a particular thing understood, so the order is not a convenience. A balance sheet does not care where the eye lands first. A call does. A call opens with a framing statement, moves through the subjects management chose, and only then admits questions. A transcript read in the order it presents itself is being read in the order it was designed to be received. Fine for receiving it. Poor for examining it.

Step one: what happens before the transcript is opened at all?

The reader sits with the published statements and writes down, in numbered order, what needs explaining. Not topics. Questions, each one attached to a figure and to a period, each one short enough that an answer would be recognisable as an answer. Four is a normal number. Two is fine. Twelve means the list is a list of interests rather than a list of questions.

Writing the list is compulsory, not advisable, and the reason is the shape of the document rather than anything about the people who produced it. A transcript arrives with a frame already built into it, and a reader carrying no list of their own will adopt that frame and then experience the adoption as understanding. Nothing announces the swap. The file gets finished, it can be summarised accurately, questions on it can be answered, and every one of those things is true of the frame rather than of the company.

Ordinary life supplies the same lesson. A shopper who walks into a shop with a written list leaves with what was wanted. A shopper who walks in without one leaves with a full bag and the pleasant feeling of having shopped well, and notices at home that the one thing actually needed is not in it. The bag is not the problem. The absence of the list is, and the bag is what stops the absence being noticed.

The questions come from the statements, so write them where the statements sit rather than where the news sits. A line that moved more than the line above it. An item that appears in one period and not the other. A total that is described in one place and split in another. A figure that management set a level for at the start of the year, and where the year finished against it. Number them one to four and leave a wide column to the right of each. The wide column is what the next hour fills.

Step two: what are the prepared remarks actually for?

The prepared remarks are for showing what management chose. Knowing the choice is genuinely useful, and it is a different thing from knowing what happened. The prepared section was written days before, reviewed by several people, and read aloud from a script. Everything in it survived a selection. So it is read once for content and a second time for the selection itself, and three things are taken away.

First, what came first. The opening ninety seconds carry the subject management most wants a reader to leave with, and that ordering is a decision rather than an accident. Second, the order of everything after it. A subject placed sixth in a list of six has been placed there. Third, and this is the one worth the reading, what is not there at all.

The omissions are the readable part of the prepared remarks, precisely because the inclusions were selected to be read. An item that appears in the statements, is large enough to matter, and gets no sentence in a scripted section that had room for it is a fact about the script. The missing item goes in the margin as an absence and is carried into the questions, where somebody may raise it and where the response becomes visible.

Step three: what sits in the questions that sits nowhere else?

The unscripted format itself. Read the question and answer section next, and read it as the only part of the whole exercise that nobody drafted, ordered or approved in advance. A filed document is composed. A press note is composed. A presentation is composed. The question and answer section is the one place where somebody outside the company chose the subject and somebody inside it had to respond in real time and in front of everyone else on the line.

The unscripted format, and not the seniority of anybody speaking, is where the value of the section comes from. A chief financial officer saying something in the prepared remarks and the same chief financial officer saying something under a question are not equivalent pieces of evidence, even when the sentences are identical, because only the second one had to be produced against a question the speaker did not set.

So this section is read with the numbered list beside it and a pen, and the first pass does only two things: it marks which of the numbered questions somebody asked, and it marks which subjects came up that were on nobody's statements at all. No answer is judged yet. Sorting answers is step five, and doing it while the asking is still being discovered mixes two jobs and does both of them worse.

Step four: what does it mean when the same question comes back?

Mark it. A repeated question is the strongest single signal a transcript contains. When two different participants return to the same subject, without coordinating and often several minutes apart, the plain reading is that the first answer did not land. Neither of them chose to spend one of their own turns on a subject they felt had been dealt with.

The signal in a repeated question is worth naming. The second answer is not the signal, and neither is anything about why the first one was given. The signal is the behaviour of the other people on the call, who are doing the same job, in real time, with the same document in front of them. Independent behaviour of that kind costs nothing to mark and is available nowhere else on the call.

The pattern lives entirely in the questions, so a reader who logs only the answers has thrown it away. An answer transcribed on its own carries no trace of having been the second attempt. When the question, the person who asked and whether anybody came back are all logged, the shape of the call survives into the notes. When the answers alone are logged, what survives is a tidy list of statements with the most informative property stripped off.

Consider a residents' meeting where the same neighbour raises the water pump three times. The three replies could be written down. Far more is learned by writing down that a fourth and a fifth resident then asked about the pump too. Five people asking is a fact about the pump rather than about the replies.

Step five: was that a disclosure, or a description of one?

Now the sorting. Every answer that touched one of the numbered questions goes back over, and each one into one of two piles. A disclosure is a figure, with a period attached, and enough definition to establish what was counted. A characterisation is a description of a figure without the figure: an adjective, a direction, a cause, a reassurance.

During the reading the two feel very similar. The similarity is the whole reason the sorting gets a step of its own. Both are responsive. Both are on the subject. Both are delivered in the same voice, in the same paragraph, by the same person, and a characterisation frequently sounds more helpful than a disclosure because it is written in words rather than in numbers.

A disclosure and a characterisation feel similar in the reading and are worth entirely different amounts, and the test that separates them is whether the answer could be written into a model without asking anybody anything further. The test is applied literally, by opening the file, finding the cell and trying. If the hand stops because how much, over what period, or measured against what is still unknown, what was in hand was a characterisation.

Step six: what gets written down, and in what form?

One line for each numbered question, and one of exactly four states beside it. Answered with a figure. Answered with a characterisation. Promised for a later date. Not answered. Four states, no fifth, and no free text substituting for a state. Let mostly answered onto the sheet and the sheet stops being a sheet and becomes a memory of a mood.

The first two states are settled from the transcript itself. A figure goes into the model with its period and its source beside it. A characterisation stays open and gets asked again, either at the next call or through investor relations, in the same numbered words. Repeating the words is what makes any change in the answer visible.

The last two states are the ones worth carrying forward, and a promise with a date attached becomes a check at the next release rather than a note nobody ever looks at again. If somebody says the split will be disclosed with the annual report, that is not an answer, it is a diary entry: the date, the question number, and what specifically was promised. A question that got nothing at all gets the same treatment with today's date instead.

Step seven: what does a transcript never give?

A filed number. Checking for one is the whole of step seven, and it takes one pass. Everything written into the figure column is checked, one entry at a time, for whether the figure also exists in a document the entity filed. Where it does, the figure was already in hand and the call merely repeated it. Where it does not, the label changes.

A figure mentioned on a call and absent from the filings is a claim rather than a disclosure, and recording it as a claim, with the date it was said, is the correct treatment rather than a cautious one. Speech on a call is not audited, was not prepared to the standard a filed statement is prepared to, and in most cases will never appear in that form anywhere again. None of that says anything at all about whether it is right. The absence says something about what is being held.

The practical consequence is small and worth the trouble. The model gets a column, or a colour, marking which inputs came from a filed document and which came from speech. Six months later, when a number turns out to have been wrong, that column shows in one glance whether the error was imported or invented, and those two have completely different fixes.

What happens when four written questions meet one call?

Follow Meghna Iyer through a single call on Sarvani Coatings Limited, an invented maker of decorative paints and industrial coatings. She has the year three statements in front of her and the transcript still closed, and she writes four questions.

Question one, the margin

Gross margin rose 2.0 points in the year, from 44.0 to 46.0 per cent. The cost of materials fell from 56.0 to 54.0 per cent of revenue. A question written without the one year decomposition would be answerable by almost anything, so Meghna Iyer does the decomposition first. Revenue rose 13.92 per cent, from Rs 2,120 crore to Rs 2,415 crore, and volume rose 6.0 per cent. Revenue growing faster than volume puts realisationWhat one unit fetched on average, worked out as revenue over units sold. Building a revenue line from a quantity and a price is covered separately. up 7.47 per cent for the year. The cost of materials rose 9.86 per cent, from Rs 1,187 crore to Rs 1,304 crore, and on that same volume the materials cost per unit of output was 3.64 per cent higher than a year earlier.

Every unit made cost more to make than it did a year earlier, and the margin still improved, so the whole of the gain is that what a unit fetched climbed faster than what it took to produce, rather than anything at all becoming cheaper. Multiply the two moves together and the arithmetic closes: 56.0 per cent of revenue carried by a ratio of 96.44 lands on 54.0 per cent. Her question is therefore narrow and hard to deflect. She asks what produced the 3.83 points of index by which realisation outran input cost per unit, over the year to the end of year three.

Questions two, three and four

Question two is about a boundary rather than a figure. A Rs 6 crore restructuring charge sits inside year three other expenses of Rs 460 crore. Management set out a cost expectation at the start of the year, so the question is whether the charge sat inside the guided cost baseThe cost expectation management set out at the start of a period, against which the year is later measured. How guidance is given and how it is tested is covered separately. or outside it. Note what makes this askable: the figure is already published, and what is missing is a definition.

Question three is the awkward one. The Rs 4 crore adjusted EBITDAEarnings before interest, tax, depreciation and amortisation (EBITDA), as a company presents the figure after adding back or removing items it describes as not representative of ordinary trading. Which adjustments are legitimate, and how the figure is rebuilt, is covered separately. problem, stated plainly: a Rs 4 crore provision write back reduced other expenses in year three, and the adjusted EBITDA of Rs 452 crore adds back the Rs 6 crore charge while leaving the Rs 4 crore write back where it is. Reported EBITDA is Rs 446 crore. Run both directions and what comes out is Rs 448 crore, or 18.55 per cent of year three revenue rather than 18.72 per cent. Meghna Iyer wants to know why one direction was run and the other was not.

Question four is about repetition. Other income for year three is Rs 38 crore, of which Rs 9 crore is an insurance claim. So Rs 29 crore, or 76.32 per cent of the line, is something else, and she wants to know how much of that something else comes back next year.

What the call gave back

Now the call itself. Question one reaches Ravindra Setlur, the chief financial officer, and comes back as better mix and disciplined pricing, with no split between the two and no figure attached to either. On the four tests, that is a characterisation, and Meghna Iyer marks it as one without any comment about why it was phrased that way.

Question two comes back as a disclosure. The restructuring charge sat outside the cost base guided to at the start of the year. There is no new rupee in that sentence, and it still changes what can be modelled. A published Rs 6 crore now has a boundary around it, and the year can be measured against the guide without it.

Question three is asked once, by one participant, and the answer sets out the general policy on adjustments without reaching the write back at all. Nobody comes back to it. Note both halves of that: the answer did not address the question, and no second participant returned to the subject, so the strongest signal available on a transcript is simply absent here. On this call no subject drew a second question at all. The absence of any repeat is itself worth a line in the notes.

Question four comes back as a disclosure. The Rs 9 crore claim is named as one time and the remaining Rs 29 crore is described as treasury income on the Rs 312 crore of cash and investments. Meghna Iyer takes the split and writes beside it that the balance quoted is a closing one, so the rate implied by it is not something she can compute from what has been published.

The sheet Meghna Iyer ends with

Two of four converted to a figure, one came back as a description and one got nothing. A conversion of two in four is ordinary rather than poor. The empty line is the only one that still needs doing, and the only one a reader without a written list would never have known was missing. Nothing else on the sheet is worth as much. Meghna Iyer carries it into the next release with today's date beside it, in the same words, so that whatever comes back can be compared against what was asked rather than against what she remembers asking.

What does this look like on a desk the morning after?

An analyst covering a listed maker of paints does this against the clock, usually with three or four entities reporting in the same fortnight. The sheet is not a document anybody publishes. The sheet is the working paper behind the two phone calls to investor relations worth making this week, and both come straight off it: the description that stayed open, and the question nobody answered. Everything answered with a figure needs no call at all.

An investor holding the shares for years rather than quarters uses the same sheet differently, and mostly uses its history. Four calls back, one line item was asked about three times and never split. Two calls back, the same line was promised for the annual report. Whether the promise was kept is a fact about disclosure, is checkable against the filed document, and takes ten minutes to establish. The check is available to anybody with the transcripts and a written list, and to nobody without one.

Both readers are producing the same artefact, a short list of things that are still open, and neither is producing a conclusion about the entity from a transcript. A household deciding whether to keep a fixed deposit at a particular bank does the same thing on a smaller scale. The decision does not come from the manager's tone. The household writes down what was said, notes what was not said, and next time checks the answer for whether it stayed the same.

Where is the real thing to be found?

Five addresses hold the document this procedure runs on and the rules that attach to it.

Who publishes it	Site
Securities and Exchange Board of India	sebi.gov.in
National Stock Exchange of India Limited	nseindia.com
BSE Limited	bseindia.com
Institute of Chartered Accountants of India	icai.org
Ministry of Corporate Affairs	mca.gov.in

Sarvani Coatings Limited, Thottam Chemicals Limited, Kesaria Surface Solutions Limited, Nandivarman Paints Limited, Ravindra Setlur and Meghna Iyer are invented.
Educational material. Not advice on any investment, tax, budget or market position.