Keeping a Transaction Decision Log That Survives Review

A transaction decision log records each decision that moved money: what was decided, the figure it turned on, who made it, the date, and what would have changed it. A log is written while the transaction runs. A log assembled afterwards records only what people remember deciding. Its value is that it makes a later review possible rather than comfortable.

What is a decision log, and what is it not?

Picture a household that spends four months choosing between two flats and finally buys one. A year later somebody asks why. The answers come back in the shape the year has given them: the schools nearby, the commute, the way the second flat felt on the one wet afternoon they saw it. Nobody can produce the sentence that was actually decisive at the time, or name the fact that would have flipped the choice. The household did not write it down. Almost nobody does.

A transaction team has the same problem at a much larger scale and with much larger consequences. Harivansh Packaging Limited settles dozens of things across a purchase, and four or five of them move real money. A minuteThe written record of a meeting: who was there, what was discussed, and that a decision was taken. A minute records the meeting, not the reasoning behind the outcome. of each meeting will exist, because minutes are routine. A minute records what was discussed and a log records what was settled and on what basis, and those are two different documents written for two different readers. One is a record of a gathering. The other is a record of a choice.

So a decision log is not a minute, and it is not a document file either. The signed agreement, the completion statement and the diligence reports all exist and are all kept, and none of them records the moment somebody chose one number over another. A file records what was agreed. A file does not record what was nearly agreed instead, or what the person agreeing believed at the time.

A log, then, is a contemporaneous recordA record written at the time the event it describes happens, rather than reconstructed later. The word carries no legal weight here; it simply means written on the day. of settled choices, one entry per choice, in a fixed shape. The log has no filing status, no external reader by right, and no formal standing. An internal working document is exactly the kind of record that gets skipped. Nothing external requires it, so nothing external notices when it is missing.

What is the procedure, in order?

Eight steps, run in this sequence from before the transaction opens to after it closes. The order matters because two of the steps have to be settled before the first decision is ever taken, and a log that starts being kept in the middle of a purchase has already lost the entries that usually matter most.

1. Name the keeper before the work startsOne person, named, before the first decision is settled. Not a role, not a rota, not the whole team.
   Checking: can the name of the person who will write every entry be stated?
2. Fix the entry shape and do not vary itFive fields, in the same order, for every entry from the first to the last.
   Checking: is the blank entry written out somewhere before anything is filled in?
3. Apply the test as each decision is settledRun the logging testThe single question that decides whether a decision belongs in the log at all. It is a filter, and its purpose is to keep the log short enough that somebody will read it. at the moment of settlement, not at the end of the week.
   Checking: was the test applied to this decision on the day, or applied to a list afterwards?
4. Write the entry on the day it is settledSame day. Not the same week, and not at the next meeting.
   Checking: does the entry date match the day the decision was actually taken?
5. Fill the fifth field in the same sittingThe four easy fields and the fifth are written together or the fifth is never written at all.
   Checking: is there any entry in the log with four fields filled and one blank?
6. Record an unconditional decision as unconditionalWhere nothing would have changed it, the entry says nothing would have changed it.
   Checking: does the fifth field say nothing, or is it simply empty?
7. Close the log at completion and hand it over unalteredNo tidying, no rewording, no removal of entries that now look awkward.
   Checking: is the version handed over the same as the version written?
8. Read it before the next transaction opensRead at the start of the next purchase, not filed and forgotten.
   Checking: has anybody read last time's log this time?

Steps one and two exist because a log with several authors becomes several logs, and a log whose shape changes halfway through cannot be read across. Everything after them is repetition: the same test, the same five fields, the same day, over and over until completion.

What are the five fields, and which one decides the entry's worth?

Every entry carries the same five fields, in the same order, every time. What was decided. One sentence naming the outcome, not the discussion. The figure it turned on. The number that moved, expressed as a number. Who decided. A name or a named body, so the entry has an author rather than a passive voice. The date. The day it was settled. What would have changed it. The fact or the assumption the choice rested on, stated so that somebody can later go and check whether it held.

Notice what the first four have in common. Every one of them can be reconstructed from documents months later. The agreement says what was decided. The completion statement carries the figure. The signature block and the approval trail carry the name. The version history carries the date. If a team lost the log entirely, a patient person with the file could rebuild those four fields for every material decision on the purchase.

The fifth field is the only one of the five that cannot be rebuilt from anything, and it is therefore the only one whose absence is permanent. No document anywhere records what Devyani Kulkarni believed would have changed her mind about the peg. Her belief existed for a few days and then dissolved. Either somebody caught it in writing at the time or it is gone, and no amount of care with the files afterwards will bring it back.

The field ordering is therefore not decorative. The four easy fields come first because they build the habit and take almost no time. The fifth comes last because it is the one that takes thought, and putting it last means it is the one an interrupted writer leaves blank. Step five of the procedure exists precisely to close that gap: the entry is written in one sitting or it is not written.

Why must the log be written while the transaction runs?

Consider how a person describes a journey taken last year. The route actually taken gets described, and it gets described as though it were obviously the route. The three wrong turns nearly made have gone. The traveller is not lying. Nobody can recover the version of the traveller who did not yet know which way the road went.

A log assembled after completion records the decisions that turned out to matter, in the form that makes sense given what happened next, and the options that were live at the time disappear because nobody remembers being uncertain. Three things happen at once, and each of them is invisible from the inside.

First, selection. Some choices only look like choices in hindsight. Suppose the net comes out small. Here it did: plus Rs 12 crore against minus Rs 15 crore, a net of minus Rs 3 crore on a headline equity value of Rs 1,140 crore, which is 0.26 per cent. The team that decided to compute the working capital adjustment and the net debt adjustment separately, rather than netting them, will not remember that as a decision at all. Written afterwards, that entry never gets made. At the time nobody knew how the two would land, so at the time the entry gets made.

Second, reshaping. The earn-out threshold of Rs 145 crore of earnings before interest, tax, depreciation and amortisation (EBITDA), set against the Rs 132 crore Sundarban Polymers Private Limited actually earned, was argued over. Afterwards it is remembered as the obvious level. The argument has been quietly removed by the fact that a number was eventually chosen.

Third, and worst, the fifth field simply cannot be answered honestly afterwards. Ask somebody a year later what would have changed their mind and they will answer with whatever did in fact change since. The two questions are not the same. An honest, careful, well-intentioned team produces the same distorted log if it writes the log late. Selection and reshaping are not dishonesty, so the rule has to be structural rather than a matter of care.

Which decisions get logged, and which ones do not?

A log that records everything is a diary, and a diary of a four-month transaction runs to hundreds of lines that nobody will ever open. A filter that depends on judgement gets applied differently by the same person on a Tuesday and a Friday. So the filter has to be a single question rather than a matter of judgement.

The question is this. Would a different answer have changed a figure that reaches the price, the mechanism or the timetable? If yes, it goes in the log. If no, it does not. The test is that short.

Run it across the six things in the figure. The completion adjustment is struck against the working capital peg, so setting the peg at Rs 96 crore reaches the price: with actual working capital at Rs 108 crore the price moved up by Rs 12 crore. It goes in. Rs 60 crore turns on the earn-out threshold, so setting it at Rs 145 crore reaches the price. It goes in. Computing both adjustments separately determines how the completion statement is built, so that decision reaches the mechanism. It goes in. Proceeding at Rs 1,140 crore reaches the price by definition. It goes in.

Moving a diligence meeting by a day changes nobody's figure and goes nowhere near the log, unless that day is the one that pushes a condition past a date somebody is counting, in which case it reaches the timetable and it does. Changing the internal reporting template changes what the team looks at and not what anybody pays. The template change stays out.

A log nobody can read is a log nobody reads, so the exclusion is the point rather than a regrettable side effect: the test excludes almost everything a transaction team does. Four entries across a four-month purchase is not a thin log. Four entries is a log somebody will actually open.

What does the would-have-changed-it field actually do?

One field carries more weight than the rest. Would-have-changed-itThe fifth field of an entry. The field names the fact or the assumption the choice rested on, written so that somebody can go and check later whether it held. is not a note of regret and it is not a list of risks. The fifth field names the fact or the assumption the decision rested on, stated in a form somebody can later go and check.

Take entry one. The peg was set at Rs 96 crore because the months used to normalise looked like ordinary trading months. The fifth field says so: what would have changed it is evidence that those months were themselves unusual. One written line makes it possible, a year afterwards, to ask a real question. Actual working capital came in at Rs 108 crore, so the price adjusted up by Rs 12 crore. Was the seller sitting on a genuinely elevated position, or was Rs 96 crore simply set too low? Without the fifth field the Rs 12 crore looks like luck, and with it the Rs 12 crore becomes a checkable claim about which months were normal.

Now take the same idea to the earn-out. Entry three records Rs 60 crore payable if EBITDA reaches Rs 145 crore. The threshold sits 9.8 per cent above the Rs 132 crore Sundarban Polymers Private Limited earned. The fifth field names a different view of how far above the last full year a threshold can sit and still be reachable, and a view of reachability is a statement somebody can test. Did the business have a plan that made 9.8 per cent look ordinary? Was it a stretch that everyone knew was a stretch?

Strip that line out and there is nothing left to test, and the figure shows what happens next. If the Rs 145 crore is reached the review says the threshold was set well. If it is not reached the review says it was set too high. The decision is identical in both branches. Only the result differs. A decision recorded without the fifth field can be judged only by its outcome, and judging a decision by its outcome is judging the wrong thing.

What does the completed log look like on this purchase?

Four entries, twenty fields, one sheet. Every figure below was already settled earlier on this purchase. The log records the decisions and assesses none of them, and that discipline is what the whole record rests on.

Entry one, the working capital peg. The decision was to set the working capital pegThe normalised level of working capital written into the agreement, against which the actual level at completion is compared so that the price can be adjusted. for Sundarban Polymers Private Limited at Rs 96 crore. The figure it turned on is that same Rs 96 crore, normalised from the trading pattern of the business being bought. The transaction team decided it and Devyani Kulkarni approved it, before signing. What would have changed it is evidence that the recent months used to normalise the level were themselves unusual trading months.

Entry two, the two completion adjustments. The decision was to compute the working capital adjustment and the net debt adjustment separately rather than netting them off against each other. The figures are plus Rs 12 crore, from actual working capital of Rs 108 crore against the Rs 96 crore peg, and minus Rs 15 crore, from actual net debt of Rs 195 crore against the Rs 180 crore the transaction assumed. The net is minus Rs 3 crore, and the equity value paid moves from Rs 1,140 crore to Rs 1,137 crore. The transaction team decided it, at completionThe moment the purchase legally takes effect and the money changes hands, as distinct from signing, which is when the agreement is entered into.. Nothing would have changed it, and the entry says so.

The net of minus Rs 3 crore is 0.26 per cent of the Rs 1,140 crore headline, and the smallness of that net is exactly why both were computed rather than a reason to skip either. A buyer who checked only the working capital side would have been wrong by Rs 12 crore. A buyer who checked only the net debt side would have been wrong by Rs 15 crore. Neither would have been wrong by Rs 3 crore.

Entry three, the earn-out threshold. The decision was to set the earn-outA further payment to the sellers that becomes due only if the business hits a stated level after completion. Nothing is paid if the level is not reached. threshold at Rs 145 crore of EBITDA, with Rs 60 crore payable if it is reached. The Rs 60 crore is 5.28 per cent of the Rs 1,137 crore paid. Ashwin Rege, who leads the transaction team, decided it during negotiation. What would have changed it is a different view of how far above the Rs 132 crore last full year a threshold could sit and still be reachable.

Entry four, the decision to proceed. The decision was to proceed at an equity value of Rs 1,140 crore. The earnings-neutral price derived on the stated test is about Rs 1,044 crore, so the gap is about Rs 96 crore. Note carefully that this Rs 96 crore is the gap over the neutral price and has nothing to do with the Rs 96 crore working capital peg in entry one: the two figures are unrelated quantities that happen to coincide. Devyani Kulkarni decided it, before signing. The fifth field names the synergy expected, and the entry states that synergy as a figure rather than describing it in words.

One caution about the Rs 1,044 crore. The neutral price is derived on the locked figure of Rs 61 crore for the target's profit after tax, and Rs 61 crore is a rounded value. The exact chain gives Rs 61.35 crore. The neutral price is therefore a derived, rounded reference and the entry records it as one. A reviewer who rebuilds the arithmetic and lands somewhere slightly different needs to know whether the difference is a rounding or a mistake, so an entry that carries a rounded input says it is rounded.

Who keeps the log, and who reads it?

The log is kept by a single keeperOne named person who writes every entry in the log. The point is consistency of shape and voice, not seniority or authority over the decisions themselves., named before the work starts. Not the chair of each meeting, not whoever happens to be in the room, and not everybody. Each author develops a different sense of what counts as an entry and a different way of filling the fifth field, so a log with several authors becomes several logs. The keeper does not need to be senior. The keeper needs to be present and consistent.

Think about a household ledger. If one person keeps it, the categories mean the same thing in March as they did in January and the year adds up. If four people write in it whenever they remember, by December nobody can tell whether groceries includes the vegetable seller or not, and the total is arithmetic without meaning.

Now the readers, and this is the part that changes how the entries should be written. There are three of them. The review that runs after completion will test each stated condition. Anybody who joins the transaction after these decisions were settled and needs to know what each one rested on. And the same team on its next transaction, looking for which of its practices worked.

None of the three is in the room when the log is being written, so an entry cannot rely on anything the writer could have said out loud. No shared context, no shorthand, no assumption that the reader knows which meeting this was. If the entry says the peg rested on the trading pattern of the business, it has to say enough that a stranger can go and look at that trading pattern.

What does a log make possible that memory does not?

Writing the log is the work. The return on that work is one specific separation, and without a log that separation cannot be made at all.

A log lets somebody separate a decision that was wrong from a decision that was right and simply landed badly, and without it those two are indistinguishable. That sounds abstract until it is crossed with the two possible outcomes and the four cells are laid out.

Where the stated condition held and the result followed, everybody agrees and nothing is at stake. Where no condition was stated and the result was poor, everybody agrees again and there is still nothing specific to learn. The two disagreeing cells are the ones that matter. A reasonable condition that then failed to hold looks, to a review working from the result, like a bad decision. And a choice with nothing recorded behind it that happened to land well looks like a good one. Both of those readings are wrong, and both are the default reading whenever the log is missing.

The default reading is outcome biasJudging the quality of a choice by how it turned out rather than by what was known when it was made. Outcome bias is a way of reading, not a fault in the person doing the reading. in its plainest form, and the word is not an accusation. Reading from the result is simply what a reader does when there is nothing else to read. Give the same reader a written condition and the reading changes. The reader stops asking whether it worked and starts asking whether the condition held, and that is a question about the decision.

The word for what a log gives a team is reviewabilityThe property of a record that lets somebody test it later against a stated condition rather than against the result. A record without a stated condition has none of it.. And the reason it matters over years rather than over one transaction is this. A team without a log cannot tell which of its practices worked and therefore cannot repeat them on purpose, so it improves only by chance. It will keep a habit that produced a good result once and drop a habit that produced a poor one once, and neither move is connected to whether the habit was any good.

What is a decision log not for?

Two things it is not, and both need saying because both are what a log quietly turns into if nobody names the boundary.

A log is not a filing. The log has no external status, no recipient by right and no prescribed form. Where a decision recorded in it would later reach a public disclosure by a listed buyer, that obligation is set by the Securities and Exchange Board of India (SEBI) and published at sebi.gov.in, and nothing about it is stated here. Where it touches an approval or a resolution under company law, that sits with the Ministry of Corporate Affairs at mca.gov.in. The log is not those things and does not substitute for them.

A log is not a defence document either, and the second boundary is the more dangerous of the two because it is more tempting. A log written to protect its author records the safe version of every decision, and the safe version is precisely the one with the fifth field empty. The drift works like this. A writer who expects the log to be read adversarially will write entry three as "the threshold was set at Rs 145 crore following discussion and analysis" rather than "9.8 per cent above the last full year was judged reachable, and a different view of reachability would have moved the threshold". The first version cannot be attacked. The safe version also cannot be tested, learned from, or used by anybody.

Which is why the boundary has to be stated at the start rather than discovered halfway. The log records and does not assess. Nothing in the four worked entries says whether the peg was well set, whether the threshold was right, or whether Rs 1,140 crore was the correct price. Merit depends on what the money would otherwise have done and on what the combined business goes on to achieve, and no figure anywhere settles either. So the arithmetic on this purchase is checkable and the merit of it is not.