Assumption Register: How to Record Every Input a Model Depends On

An assumption register is the list of every input a conclusion depends on, each with its value, where it came from, the date it was true, who is responsible for it, and how much the conclusion moves if it is wrong. The register exists because a forecast without its inputs written down cannot be reviewed, updated or defended. The register is what turns a number into a reviewable argument.

Underneath the answer sits one plain fact: a conclusion is a function of its inputsThe numbers and statements a calculation starts from: a price, a rate, a volume, a date. Change an input and everything downstream of it changes., and inputs age. A household deciding in January to buy a scooter on a two-year loan faces the same shape of problem. The decision rests on a salary, a fuel price, a monthly instalment and a guess about how often the scooter replaces bus fares. Six months later the fuel price has moved and the salary has not. Nobody in that house wrote the four numbers down, so nobody can say which one turned the comfortable plan into a tight one; they only know it feels tight. Writing inputs down with dates is the only way to know later which one moved. An assumption registerA table, usually one row per input, kept alongside a forecast or model, that records what was assumed, where it came from, when, and who is answerable for it. is the discipline of writing them down: one row for every input, six columns, a named owner, and a rule for deciding which inputs earn a row and which do not.

What problem does an assumption register solve?

The finance version is the same shape at a larger scale, so the scooter is worth returning to. Every plan that reaches a number rests on things assumed to be true. Some are documents that can be pointed at, the salary slip, the loan schedule. Some are guesses, how many bus fares the scooter saves. When the plan works nobody asks which was which. When it does not, the difference is everything. A plan that failed on the guess and a plan that failed on the salary slip need completely different repairs. The register exists for that later moment, and it has to be built at the earlier one.

An unrecorded input is an invisible bet: the conclusion leans on it exactly as hard as on the inputs that were checked, but nobody can see it is there. Ishaan Verma, an invented analyst at an invented investment team, recommends in January that the team lend Rs 25,00,00,000 to Kaveri Cold Chain Private Limited, an invented refrigerated warehousing business. His case is that a signed pharma contract lifts occupancy from 62 per cent to 80 per cent within a year, moving the business from break-even to a profit before tax of Rs 4,20,00,000. The Rs 4,20,00,000 forecast rests on five things: occupancy today at 62 per cent, a contract worth 14 points of capacity, 4 more points of spillover business, a tariff of Rs 1,150 per pallet per month held flat, and a power bill of Rs 1,80,00,000 held flat. Look at the picture below. Four of the five supports carry a label saying where they came from. The third carries none. The forecast does not know the difference; the reviewer, twelve months on, has to.

What are the columns of an assumption register, and why each?

The register's real reader is the person who opens it twelve months later. Not Ishaan Verma, who remembers, but a lender's monitoring officer, or a new analyst, or the same team after two of its members have left. A reader arriving that late has six questions, and each column exists to answer exactly one of them. What was assumed? The input column names it plainly: occupancy today, contract share, spillover, tariff, power. How big was it? The value column carries the number with its unit: 62 per cent, 14 points, Rs 1,150 per pallet-month. Where did it come from? The sourceThe document, person or method a number was taken from: an audited report, a signed contract, a rate card, or somebody's estimate. The source tells the reader how much weight the number can bear. column answers that, and the source is what separates a document from a guess. When was it true? The as-of dateThe date on which a figure was correct. A 62 per cent occupancy as of January says nothing about October; the date tells the reader how stale the number may be. answers that. Who checks it now? The ownerThe named person responsible for keeping a row current: noticing when the value moves, updating it, and raising a hand when it breaks the case. column names one person. And does it matter if wrong? The sensitivityHow much the conclusion moves when one input moves. A high-sensitivity input can wreck the answer with a small error; a low-sensitivity one barely shifts it. rank says how much.

Each column exists because a reviewer will one day ask the question it answers, and a register missing a column leaves that question unanswerable. Drop the source column and audited figures and guesses look identical. Drop the as-of date and nobody knows whether 62 per cent was January's number or last year's. Drop the owner and every row belongs to everybody. A row that belongs to everybody belongs to nobody. Drop the sensitivity rank and the reviewer spends the afternoon on the row that could never have mattered. The figure lays the six questions against the six columns; the table beneath it gives each column its job in one line.

Which inputs go in the register, and which stay out?

A tempting mistake is to register everything, and a register with sixty rows is a register nobody reads. Back to the scooter. The household writes down the salary, the fuel price, the instalment and the bus-fare guess. The household does not write down that instalments are paid monthly, or that twelve months make a year, or that the total is the sum of the parts. Monthly instalments and twelve-month years are how the plan is worked out, not what it depends on. The same line runs through Ishaan Verma's model. Pallets multiplied by tariff multiplied by twelve is arithmetic, and it belongs in the model. Occupancy at 80 per cent is a belief about the world, and it belongs in the register.

An input belongs in the register if the conclusion would change were it wrong; mechanics, formulas and layout stay out. That is the whole test, and it has one trap worth naming: a number that is written down as a fact is still an assumption when it is carried forward. Power cost of Rs 1,80,00,000 is last year's actual, so it feels like data. Held flat for next year it is a bet that the electricity tariff does not move, and that bet belongs in the register with its own row. The decision map below runs the test on the pieces of the Kaveri case.

How does the register connect to the base case and to sensitivity?

The register does not stand alone; it sits between two things the reader has already met. The base case is the set of values that together make the central forecast, and the register is where those values are written with their provenanceThe origin and history of an item of information: who produced it, from what, and how it reached the person using it. Provenance is what lets a reader judge trust.. Sensitivity asks which input matters most, and the answer fills the last column. So the base case fills the rows and sensitivity ranks them. Neither replaces the register; the register is what makes both of them checkable by somebody else.

Sensitivity ranks the rows so that the reviewer's attention goes first to the input that can do the most damage. How sensitivities are computed is set out under sensitivity analysis. One illustrative rule, stated on screen in the simulation below, is that each point of occupancy is worth about Rs 23,50,000 of profit before tax at Kaveri Cold Chain. Under that rule a plausible miss on each row produces a swing that can be compared. Spillover coming in at 0 rather than 4 costs Rs 94,00,000. A tariff 4 per cent softer than the rate card costs Rs 75,20,000. A power tariff revision of Rs 50,00,000 costs exactly that. The contract ramping 2 points slower costs Rs 47,00,000, and the opening occupancy being 1 point wrong costs Rs 23,50,000. In the bars, the row with the biggest swing is the same row that had no source. The biggest swing landing on the row with no source is the whole reason the source column and the sensitivity column sit side by side.

How is the register kept alive after the decision?

Here is where most registers die. Registers are written for the committee meeting, filed with the approval, and never opened until something has gone wrong. A register written once is a photograph; a register kept alive is a diary. The difference is the as-of date, read the other way round: every row that was true on a date will need checking on a later date, and the register should say when. Occupancy from Kaveri Cold Chain's own report gets rechecked every quarter when the next report lands. The contract row gets rechecked when the pharma client actually moves in. The tariff row gets rechecked when the rate card renews. Each of those is a date the owner can put in a calendar in January.

A register stays alive when every row carries a revisit date and a named owner, and when new rows are born the moment new information arrives. That second half is the harder one. In February a research report crossed Ishaan Verma's desk noting that a competitor planned a cold store nearby. Nothing in the register mentioned a competitor, so there was no row to update, and the report was filed. A living register treats that report as the birth of a row: competitor entry, source the February report, as-of February, owner Ishaan Verma, and the spillover row reopened because a competitor is exactly what spillover depends on. Follow the timeline below and notice the two rows that had no revisit date and the one row that never got born.

What did Ishaan Verma's register for Kaveri contain, and what did it miss?

Now read the actual document, drawn below as Ishaan Verma filed it in January. Five rows, six columns. Occupancy 62 per cent, source Kaveri Cold Chain's own report, as of January, ranked fifth for sensitivity. Contract 14 points of capacity, source the signed pharma contract, January, ranked fourth. Spillover 4 points, source Ishaan's estimate with nothing behind it, January, ranked first. Tariff Rs 1,150 per pallet-month held flat, source the rate card, January, ranked second. Power Rs 1,80,00,000 held flat, source last year's actual, January, ranked third. The owner column names Ishaan Verma for the rows he built and Devika Rao, the invented monitoring associate, for the two rows the monitoring desk would watch after drawdown. As a document it is better than most: every row has a source and a date, and the ranking is honest.

The rows that hurt Kaveri Cold Chain were the unsourced estimate and the two rows that did not exist. Twelve months on, occupancy reached 71 per cent rather than 80: the contract delivered its 14 points, spillover delivered 0 rather than 4, and a competitor opened nearby. Power rose to Rs 2,30,00,000 on a tariff revision nobody had recorded as a risk. Profit before tax came in at Rs 1,10,00,000 against the forecast Rs 4,20,00,000, and the loan was serviced on time. Look at the two ghost rows under the table. A power tariff revision and a competitor's entry were never registered, not because Ishaan hid them, but because neither was thought of as an assumption. Held flat felt like a fact, and no competitor felt like the absence of a fact. Both were bets, and the register that could have shown them was five rows long when it should have been seven.

How do lenders, analysts and investors actually use an assumption register?

A lender uses the register as the monitoring plan. When the invented investment team's loan to Kaveri Cold Chain drew down, the useful question was not whether the forecast was right but which rows to watch, and how often. A register with owners and revisit dates converts straight into a monitoring calendar: occupancy checked against each quarterly report, the contract row closed out when the pharma client moved in, the tariff row reopened when the rate card renewed. A lender who receives a forecast without a register has to invent that calendar, and usually watches the wrong things.

An analyst uses it as the honest boundary of a view. When Ishaan Verma presents a profit before tax of Rs 4,20,00,000, the register lets him say in the same breath: 76 of the 80 points rest on documents, 4 rest on my judgement, and here is the row I am least sure of. Naming the four points that rest on judgement is what separates a defensible view from a confident one. The register also protects the analyst later. A review that finds the spillover row marked estimate in January cannot say the risk was hidden. A forecast that cannot be reviewed cannot be trusted even when it happens to come true, so practitioners keep the register less to be right than to be reviewable.

A household uses it without the name. The scooter plan on the fridge door with four numbers, the date each was written and a note saying which one was a guess is an assumption register. When the fuel price moves the household conversation is short. The row that moved is the row that was marked as the risk. Kaveri Cold Chain's loan was serviced on time and its profit was Rs 1,10,00,000 rather than Rs 4,20,00,000; whether that was a good decision or a lucky one is a question that can only begin to be answered because the January register exists.

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