Decision Discipline: Separating a Good Decision From a Good Outcome

Decision discipline is judging a decision by the quality of its process at the time it was made, not by how it turned out. Because outcomes are uncertain, a good decision can end badly and a bad one can end well, and confusing the two teaches the wrong lessons. The grid of decision against outcome has four boxes; the two off-diagonal ones are where the learning is.

Outcomes carry luck. Processes do not. Decision disciplineThe habit of grading a decision on what was known and done at the moment it was made, and keeping that grade separate from what happened afterwards. exists because of that one asymmetry: the process is the only part of a decision anyone actually controls, so grading the process is the only way to make the next decision better. The Kaveri Cold Chain loan, an invented case, has carried that question from the start. Ishaan Verma, the analyst on the file, recommended Rs 25,00,00,000 in January; twelve months on the loan was serviced on time and profit before tax came in at Rs 1,10,00,000 against a thesis of Rs 4,20,00,000. Was that a good decision, or a lucky one, and how would anyone tell? Answering that takes five steps: what the discipline protects against, why uncertainty makes the two questions separable, which box of the grid teaches most and which least, how a process is actually graded, and where Kaveri lands.

What is decision discipline, and what is it protecting against?

The clearest illustration has nothing to do with money. An uncle drives home from a wedding after three drinks and arrives without a scratch. Asked the next morning whether it was a good decision, he points at the car in the driveway. But nothing about the driveway says anything about the decision. The decision was made at the gate of the wedding hall with the information available there: three drinks, a dark road, forty minutes of driving. Judged at that moment, it was a poor decision that happened to end well. The safe arrival did not make it wise; it made it lucky. A poor decision that ended well gets repeated with more confidence. So the danger is not last night. The danger is the next wedding.

Decision discipline is the habit of grading a decision on what was known and done at the moment it was made, and keeping that grade separate from the outcomeWhat actually happened after the decision: the result, once the uncertain things had resolved one way or the other.. What it protects against is outcome worship: the reflex to call whatever worked a good decision and whatever failed a bad one. Card players have a name for that reflex, resultingA term from card play for judging the quality of a decision purely by how the hand turned out, as if the result were the only evidence., and the idea that a decision should be judged by its processThe steps taken to reach the decision: what was written down, what was checked, what was asked, and what would have changed the answer. rather than its result has a named author, Annie Duke, whose 2018 book Thinking in Bets is the standard treatment. The discipline is not a way of excusing bad results, and it is not a claim that outcomes do not matter. A lender who never gets repaid goes out of business however good the paperwork was. The outcome contains something the decision did not: everything that happened afterwards. So the outcome is the wrong evidence for grading the decision, and that is the whole of what the discipline says.

Look at the two piles of evidence in the figure below. The pile on the left is what Ishaan Verma had on his desk in January: five register rows, a log entry with three kill criteria, a question, a model. The pile on the right is what December knew: occupancy 71 per cent, a competitor open since May, power at Rs 2,30,00,000, profit before tax Rs 1,10,00,000, loan serviced. Two different questions look at two different piles. 'Was it a good decision?' can only be answered from the left. 'Was it a good outcome?' can only be answered from the right. Outcome worship answers the left-hand question with the right-hand pile.

Why can a good decision produce a bad outcome, and a bad one a good outcome?

Because between the decision and the outcome sits time, and time is where uncertainty resolves. Think of a street vendor who decides in the morning how many plates of food to prepare. She counts the office crowd, checks whether it is a working day, looks at the sky. Two hundred plates is a well-reasoned number. Then an unannounced strike empties the office, and she carries eighty plates home. Was two hundred a bad decision? Nothing she could have known at seven in the morning said so. Now flip it: the vendor next to her guesses three hundred plates on a hunch, and a wedding party turns up unannounced and clears her out. Better outcome, worse decision. The morning had a spread of possible afternoons in it, and each vendor drew one.

A decision and its outcome can be graded differently because the outcome depends on things the decision could not contain, so luckThe part of a result that came from events the decision maker could not have known or controlled at the time, good or bad. sits between them. Every decision made under uncertainty is really a bet on a range: a range of occupancies, a range of power tariffs, a range of things competitors might do. A good process narrows the range honestly, names what would fall outside it, and says what to watch. But even a perfect process cannot collapse the range to a point, and the outcome is one draw from whatever range remained. The range that survives even a good process is why 'it worked' and 'it was well decided' are two facts, not one. Skill shows up in the range being bet on; luck shows up in which draw came out.

The figure below shows the Kaveri thesis as a range rather than a line. Ishaan Verma's forecast ran from 62 to 80 per cent occupancy. The contract alone, if it delivered its 14 points and nothing else moved, would have reached 76. December's actual was 71: the contract delivered, spillover gave nothing, and a competitor took some of what was there. The shaded band is illustrative, the spread of Decembers that January could reasonably have imagined once the unsourced spillover input and the unregistered risks are allowed to move. Look at where 71 sits: inside the band, in the lower half of it. Not the disaster, not the thesis. One draw from a range that a better process would have drawn more honestly.

What is the four-box grid, and which boxes teach most?

Decision quality runs along the bottom, weak on the left and sound on the right. The outcome runs up the side, bad at the bottom and good at the top. Every decision ever made lands in one of the four boxes that result. Sound process, good outcome: earned. Weak process, bad outcome: deserved. Earned and deserved sit on the diagonal. In each of them the outcome and the process agree, so nobody has to think, and the two boxes teach almost nothing. The interesting boxes are the other two.

The two off-diagonal boxes, sound process with a bad outcome and weak process with a good outcome, are the only boxes in which the outcome lies about the process, and the lesson lives there. The unlucky box, sound process and bad outcome, teaches about the world: it shows what a good register still could not have known, and its trap is rewriting a sound process to fit the pain. The lucky box, weak process and good outcome, teaches about the process: it shows every hole that happened not to matter this time. The uncomfortable part is what follows from that. A bad outcome forces a review, so the unlucky box gets studied. A good outcome closes the file, so the lucky box does not. So the box that holds the most correctable lesson is the box that, in practice, teaches least.

How is the quality of a process actually graded?

The structures covered under evidence and judgement have already built the yardstick, so grading a process is not a matter of taste. A decision under uncertainty is well made when five things existed at the time and were used. Was there an assumption register, with a source and a date on every input, and was the weakest input visible as weak? Was there a decision log entry that said what was decided, why, and what would change the decision maker's mind, with kill criteriaThe conditions written down in advance that, if they came true, would make the decision maker reverse or reopen the decision. in place before the outcome arrived? Was the research question sharp enough to be answered by a fact rather than by a feeling? Was there a counter to confirmation bias, a person or a rule whose job was to argue the other side and to reopen the case when contrary evidence arrived? And was the counterfactual built, comparing the decision with the road not taken rather than with nothing?

The quality of a process is graded by asking whether each of the five structures existed and how well it was used, and the outcome is not on the list. That last clause is the discipline. Look at the figure below and notice the sixth row, 'was the outcome good?', written there and struck through. The moment that question enters the grading it swallows every other row. The gradingScoring each element of the process on a fixed scale, absent, present with gaps, or present and fully used, so that two reviewers would reach a similar number. itself is simple enough for two reviewers to agree on: for each structure, absent scores nothing, present but leaky scores half, present and used scores full. Five structures, so the process is marked out of ten. Notice too that a structure can exist and still be leaky: a register with an unsourced row, a log whose kill criteria miss the risk that arrived. Existence is the first question, use is the second, and the honest grade usually lives between them.

How does the discipline change behaviour over many decisions?

A single draw from a range says almost nothing about the range, so a single decision reveals little. Twenty decisions can. Suppose an analyst grades her own process after every call, whether the call worked or not, and fixes one hole each time: a source added, a kill criterion written, a counter appointed. Her process score climbs. Luck never leaves, so her outcomes do not climb in a line. They scatter, some above the process, some below. The height of the scatter is what matters. A better process bets on a better range, so as the process rises the cloud of outcomes rises with it. The promise of the discipline is a modest one: not that every outcome improves, but that the average of outcomes follows the process, and the process is the only line that can be moved.

Over many decisions, outcomes scatter around the process, and improving the process is what drifts the whole scatter upward. There is a second effect, quieter and just as important. An analyst who is graded on process stops hiding. A leaky register is what will be found, not a bad outcome, so she writes the register honestly. Filing the February trade report is a process fault whether or not the competitor ever opens, so she reports it instead. Grade people on outcomes and they learn to manage the story of the outcome. Grade them on process and they learn to build the process. Neither the household nor the credit committee can control the flood or the competitor. Both can control what was written down before it.

How does the Kaveri decision grade?

Now the case, and be fair to it. Ishaan Verma did more than many analysts do: he kept a register with sources and dates on four of five rows, he wrote a log entry in the right form with three kill criteria, and he framed a question and answered it. Grade each of the five structures from January's desk only, and resist the pull of December. The register existed, but the spillover row rested on an unsourced estimate and two risks that later hurt, the power tariff revision and competitor entry, were never rows at all: leaky, one point. The log existed, with kill criteria of a cancelled contract, occupancy under 68 at month six, and a tariff cut, and it missed the competitor and the power cost: leaky, one point. The question, 'is Kaveri a sound credit?', was asked and answered but was too blunt to say what to check and when; the sharp version was written only in review: leaky, one point. There was no bias counter, and the February report noting the competitor's plan was filed without reopening the case: absent, nothing. The counterfactual, what if the loan had been declined, was not built until the review: absent, nothing.

Kaveri grades as a middling process, three of ten: three structures present and each leaking, two missing altogether. The outcome it met was tolerable, and that combination lands the decision in the box that teaches least unless someone grades the process anyway. The outcome was tolerable rather than good: the loan was serviced on time, which is what the lender most needed, but profit before tax of Rs 1,10,00,000 was Rs 3,10,00,000 short of the Rs 4,20,00,000 thesis, and a business earning Rs 1,10,00,000 on a Rs 25,00,00,000 loan is closer to the edge than anyone intended. So the marker lands left of the process line and just above the outcome line: weak process, acceptable outcome. The finding of the discipline is not 'it worked'. The finding is 'it worked despite': despite an unsourced spillover input, despite two unregistered risks, despite kill criteria that could not have fired for the competitor that arrived, despite a report filed in February. Had the pharma client wobbled, the same process would have met a bad outcome, and every hole would have been found by a review that this outcome never triggered.

How do lenders, analysts and investors actually use decision discipline?

A lender uses it through the review calendar. A credit committee that reviews only the loans that went sour is grading by outcome, and it will never open the file that most needs opening. The disciplined version samples approvals for review whether or not they were repaid, and grades each on the sheet above: register, kill criteria, question, counter, counterfactual. When the Kaveri file comes up in that sample, 'serviced on time' is written in the outcome column and then ignored while the process is scored, and the finding, three of ten, becomes two changes to the approval template rather than a tick.

An analyst uses it on her own record. A forecast that came true is a pleasant thing, and the temptation is to remember it as evidence of judgement. Practitioners who keep a decision log can ask the sharper question: did the outcome arrive for the reason I gave, or for a reason I never wrote down? Ishaan Verma's thesis said occupancy would reach 80 through the contract and spillover; the loan was serviced because the contract delivered and the pharma client paid. Half his reason held. A log lets him see which half, and that is a better lesson than either 'I was right' or 'I was wrong'.

A household investor meets the discipline through the neighbour whose stock tip doubled. The tip was a weak process, no register, no question, no counter, and the double was one draw from a wide range. The danger is not the tip that worked; it is the second tip, taken with more money and more confidence because the first one landed in the lucky box. Wedding-scale decisions carry the same shape: a wedding planned without a budget register that came in on cost because a relative's caterer gave a discount worked despite the planning, and the next wedding in the same house will inherit the missing register unless somebody notices that the discount, not the plan, did the work.

Kaveri Cold Chain Private Limited and Ishaan Verma are invented.
Educational material. Not advice on any investment, tax, budget or market position.