Post-Mortem: Reviewing a Financial Decision After the Outcome Is Known

A post-mortem reviews a decision once its outcome is known, working from the record made at the time rather than from memory, to find what in the process should change. A good outcome hides a bad process just as well as a bad one exposes it, so the review is run on good outcomes as well as bad. The review is blameless about people and unsparing about process, and its product is a change, not a verdict.

A decision leaves a recordThe written material made at the time of a decision: the log entry, the assumption register, the model versions, the question as framed. Distinct from anyone's later memory of the decision.; an outcome leaves a temptation to rewrite it. Twelve months after a loan is made, everyone in the room knows how it turned out, and that knowledge quietly reaches back and recolours every line that was written before it. The post-mortemA structured review of a decision held after its outcome is known. The name is borrowed from medicine, where the examination happens after the fact to learn what happened, not to revive the patient. exists to read the record before the temptation gets to work on it. The assumption register, the audit trail, the decision log, the sharpened research question, the counterfactual, the bias check and the grade all exist so that this one reading is possible. A post-mortem is run on the loan that was repaid as readily as on the loan that soured, in an order that keeps hindsight out, and it ends in changes that carry a name and a date.

What is a post-mortem for a decision, and what is it for?

Start at a kitchen table. A wedding came in 30 per cent over its budget, and the household sits down afterwards to talk about it. There are two ways the conversation can go. In the first, someone asks who overspent, and the answers arrive fast: the caterer's late additions, the uncle who insisted on the bigger hall, the cousin who ordered flowers twice. Voices rise, and by the end everyone has a view about a person and nobody has changed anything. In the second, someone fetches the budget sheet drawn up in the planning weeks and lays it beside the bills. Nobody kept a running total after week three, so the plan and the spend parted there, and the pattern is plain on the paper. The finding is not about a person. The finding is that the process had no running total, and the change is that next time there will be one, kept weekly, by a named person. Same wedding, same overspend, two entirely different products.

A post-mortem is a review of a decision, held after its outcome is known, that reads the record made at the time in order to find what in the process should change; its product is a change, never a verdict on a person. The distinction sounds soft and is anything but. A verdict ends a conversation: the person was careless, the person was unlucky, the loan was a mistake. A change begins something: the register template now requires a source, the kill criteria now include competitor entry, a named colleague now argues the other side of every large credit. The outcome does not itself say what to change. Kaveri Cold Chain, an invented borrower, posted a profit before tax (PBT) of Rs 1,10,00,000 against a thesis of Rs 4,20,00,000. The gap says only that something went differently from the plan; it says nothing about which line of the plan was weakly built. Only the record can say that, and the record is the assumption register, the decision log, the model's audit trail and the question as it was first written. The outcome tells the review where to look. The record tells it what was there.

Why run one on good outcomes as well as bad?

Picture a scooter rider who reads a message at the wheel and gets home safely. Getting home safely is not evidence that reading at the wheel is safe; it is evidence that this time nothing was in the way. If the household only reviews the rides that end in a fall, it will review the one in a thousand and wave through the nine hundred and ninety-nine that were rehearsing it. The same is true of a food stall that runs out of change every Friday and gets away with it because regulars are patient: the process is broken every week, and the outcome hides it until the week a stranger walks off. Luck testifies for both sides. A good outcome is a poor witness to a good process.

A post-mortem is run on good outcomes as well as bad because a good outcome hides a bad process just as well as a bad outcome exposes it, and the hidden case is the one that costs the most over many decisions. Kaveri Cold Chain serviced its loan on time and reported a profit before tax of Rs 1,10,00,000, so the natural move is a tick on the file and on to the next paper. But set beside the road not taken, the Malnad Grain Storage renewal that would have earned Rs 2,37,50,000 against the Rs 3,00,00,000 the Kaveri loan earned, the outcome reads as a risky bet that landed narrowly, keeping about 26 per cent of the profit it promised. The grade the sequence reached was a middling process with a tolerable outcome: it worked despite. The box of bad or middling process with a good enough outcome is the box a team never visits unless it reviews good outcomes on purpose, and that box is where the next loss is being quietly practised. Fairness is part of the method. Nobody on the team was careless. The outcome simply gave no one a reason to look. The post-mortem is the reason.

Why blameless, and how is that kept honest?

A restaurant that fines the cook for every burnt dish gets fewer reported burnt dishes, not fewer burnt dishes. Fields where a mistake can kill, aviation and hospital medicine among them, learned this the hard way and built their incident reviews around the same idea: a review that punishes the person who reports gets no reports, and a review with no reports learns nothing. The word for the alternative is blamelessA review rule that findings are written about the process, not about the character or competence of a person. Blameless does not mean nobody is accountable. Accountability attaches to fixing the process., and blameless is widely misread as soft. The rule is the opposite of soft. Because nobody in the room needs to defend themselves, the review can be hard on the process.

A post-mortem is blameless about people so that it can be unsparing about process, and blameless is not the same as consequence-free: the consequences attach to the process gaps, each of which gets a named owner and a date. Three habits keep the rule honest. First, every finding is written as a sentence about a process element: "the register had no row for competitor entry" is a finding, "Ishaan Verma forgot about competitors" is not, even though both are true. Second, the record is read before anyone speaks. The first thing in the room is January's paper and not December's opinion. Third, the person whose decision is under review presents the record themselves. Ishaan Verma reads out his own log entry, his own register, his own version history, and the review's job is to listen for where the process left him exposed. Having the decider read out the record is what separates a blameless review from a trial with polite manners. If a person repeatedly ignores an agreed process, that is a real conversation, but it is a different conversation, held elsewhere, and it is not what a post-mortem is for. Mixing the two produces neither.

What does the review read, and in what order?

A doctor reads the notes before looking at the scan; a cook tastes the sauce before reading the review. In both cases the order is chosen so that the thing most likely to bias the reading arrives last. A post-mortem has the same problem, and the biasing thing is the outcome. Once Rs 1,10,00,000 against Rs 4,20,00,000 is in the room, every line of January's paper reads as either prophecy or blindness. So the review reads the record first, in a fixed order, and lets the outcome in only when the record has been heard.

The review reads the record in order, log, register, trail and question, then builds the counterfactual, then checks for bias, then grades the process, and only then writes findings and changes. Each stage sits where it does for a reason. The log comes first because it says what was decided, on what basis, and what would have changed the decider's mind, and that last line is the frame for everything after: the review will keep asking whether what actually happened was on the watch list. The register comes second because it lists the inputs the basis rested on and shows which had a source and which did not. The trail comes third because it shows whether the number the committee saw can be rebuilt from those inputs, or whether some of it was typed. The question comes fourth because it shows what the work was pointed at and what it was never going to catch. Only now does the review build the counterfactual, from the January facts the record has just supplied, and set the outcome beside it. Then the bias check, a stage that asks what arrived after the decision and what was done with it. Then the grade, process against outcome. Then the findings, and finally the changes. Reverse the order, start from the outcome, and every stage becomes a hunt for who missed what.

What is a finding, and how does it become a change?

Back at the kitchen table, "we spent too much on flowers" is an observation, "the cousin is extravagant" is a verdict, and "the budget sheet had no running total after week three, so nobody could see the drift" is a findingA review's statement of what in the process allowed the outcome to diverge from the plan. Names a process element, not a person, and is specific enough that a change can be written against it.. The difference is that the finding names a part of the process precise enough to be pointed at and altered. Getting to that part usually takes more than one why. Why did the spend drift? Nobody noticed. Why did nobody notice? There was no running total. Why was there none? The sheet had no column for it and no one was asked to keep it. An answer that names a person is a signal to keep asking; the questioning stops only when the answer is a process element. Stopping there is the whole of root causeThe underlying process condition that allowed a divergence, reached by asking why repeatedly until the answer is something in the process rather than someone in the room. thinking as a review needs it.

A finding becomes a change when it is specific enough to name a process element, has a named owner, and carries a date; a line missing any of the three is an observation, however true. "Be more careful about competitors" fails all three. "For any capacity business, the kill criteria must include competitor entry and input-cost risk; owner, the head of credit; in force before the next credit paper" passes all three, and it is the kind of sentence that changes what happens in the next meeting. The Kaveri review ended with three such lines. Every register row must carry a source or a flag saying it has none. Kill criteria for a capacity business must include competitor entry and input-cost risk. Every credit above Rs 10,00,00,000 gets a designated disagreer, one person whose job in the meeting is to argue the other side and to carry inconvenient reports into the room. Three lines are the review's whole product, and that is enough. A review that ends with twelve findings and no owners has produced a reading list.

What did the Kaveri post-mortem find?

Now the review itself, stage by stage, twelve months after Ishaan Verma's January recommendation. Fairness to him at every stage is what makes the findings usable. The log, read first: the entry recorded the decision to recommend Rs 25,00,00,000, the basis that a signed pharma contract would lift occupancy from 62 to 80 per cent with spillover, a confidence of reasonably high, and three things that would change his mind: the contract cancelled, occupancy below 68 at six months, or a tariff cut. Occupancy at six months was around 73, above the trigger, so nothing tripped, and yet the thesis was already 7 points short of its path. The kill criteria were built to catch collapse, not shortfall, and they had no line for competitor entry or power cost. The register, read second: five rows, four sourced, the spillover row of 4 points recorded as Ishaan Verma's own estimate with no source, and no row at all for competitor entry or for a power tariff revision. The trail, read third: three versions in twelve days, from Rs 2,88,00,000 to Rs 3,60,00,000 to Rs 4,20,00,000, the last carrying a spillover cell typed to 4 over a formula that gave 3 and a power cell typed to Rs 1,80,00,000 over a formula that carried an uplift, so Rs 60,00,000 of the thesis cannot be rebuilt from recorded steps. The question, read fourth: "is Kaveri a sound credit?" No answer could fail that question, and it never scheduled a month-six check on spillover.

Then the judging stages. The counterfactual, built from January's facts only: the Malnad Grain Storage renewal at 9.5 per cent would have earned Rs 2,37,50,000 with little exposed to luck, against the Rs 3,00,00,000 the Kaveri loan earned at 12 per cent off a business that kept 26 per cent of its promised profit. On January's facts both roads were defensible; the taken road was the risky one, and it landed. The bias check: the February trade report noting a competitor's planned facility twenty kilometres away, read, judged early stage, filed, with no kill criterion to trip, no register row to update, and no colleague whose job was to carry it into a meeting. The grade: middling process, tolerable outcome, it worked despite. The Kaveri post-mortem found that the shortfall traced to three process gaps, an unsourced register row, kill criteria that ignored competitor entry and input cost, and the absence of anyone tasked with disagreeing, and it found nothing that required Ishaan Verma to have been a worse or a better analyst than he was. Look at how the shortfall itself maps onto those gaps. Of the Rs 3,10,00,000 by which profit before tax missed the thesis, roughly Rs 1,15,00,000 sat on the four spillover points that had no source, the row a source rule would have flagged; roughly Rs 1,45,00,000 sat on the five points the competitor took, the risk a kill criterion and a disagreer would have kept in view; and Rs 50,00,000 sat on the power tariff, the input-cost risk that had no register row. Every rupee of the miss has a process address.

How do lenders, analysts, investors and households actually use a post-mortem?

A credit desk that takes the idea seriously reviews a sample of performing loans every year, not only the ones that soured, and it reads each against the paper written at sanction: the log, the register, the model as it was. The desk is looking for the top row of the grid, the loan that was repaid off a thesis that was never really tested. The top row is where the next default is being rehearsed with growing confidence. Some desks pair the post-mortem with its mirror image, the pre-mortemA technique run before a decision, in which the team imagines the decision has already failed and writes down why. Named and described by the psychologist Gary Klein., in which the team imagines before the loan is made that it has already gone wrong and writes down why; the pre-mortem is the cheapest source of kill criteria there is, and its list is exactly what the later review reads the log against.

Practitioners use the post-mortem as the closing half of a loop. The record is the opening half: the log, the register and the question are written so that a review can read them, and the review is run so that the next log, register and question are written better. A research analyst working under the conduct rules of the Securities and Exchange Board of India (SEBI) is expected to be able to show the basis of a past recommendation, and showing that basis is the record half of the loop. The analyst who then sits down twelve months later and grades the basis against the outcome, before writing the next note, has closed the loop. An investor does the same thing on a smaller sheet, for a holding bought or a holding sold: what was written down at the time, which of the reasons held, and which reason had no source behind it. A household does it after any large decision, the car, the school, the loan against the flat, and the useful form is always the same three questions: what was known then, where did the plan and the outcome part, and what one thing about how the household decides will be different next time. None of them asks: who was wrong.

Kaveri Cold Chain Private Limited, Malnad Grain Storage, Ishaan Verma, Meera Pillai, the initials in the meeting notes and the investment team are invented.
Educational material. Not advice on any investment, tax, budget or market position.