How to Build Business Scenarios for a Company

A business scenario is a held-structure conditional. Take the company's own published lines, hold everything, move one stated quantity, and say what the structure would then report, with the assumption named in the same breath. No reason for the move, no weight, no rank, no future period. If the sentence needs a future period to be true, it is a forecast rather than a scenario.

An analyst is handed one company's accounts and asked for three cases. What may be written?

The request arrives looking small. One set of published accounts sits on the desk, a colleague wants three cases by the end of the day, and the arithmetic involved is multiplication and subtraction that a school child could check. So the hard part cannot be the sums, and it is worth saying out loud what the hard part actually is before a single figure gets typed.

Here is what most people do, and it is worth naming. Two more sets of figures get invented, each looking like the published one; each of the three is given a name that hints at how good it is; and then a number is put beside each name so that whoever reads the note downstream can average them into one figure. Every step of that feels like diligence. Every step of it does something the accounts cannot support.

One question sorts the whole problem: which of the three sentences about to be written is true without a future period being named in it? Not which is best researched, not which is most carefully argued. Which one survives having its dates struck out.

Take it out of finance for a second. A household is asked what next month's electricity bill will be. The same household is asked what the bill would come to if the same appliances ran the same hours at a stated tariff. The second question can be answered off the meter and the wiring, by anybody standing in the kitchen, and the answer stays right whether or not anybody ever runs the appliances that long. The first question cannot be answered at all, by anybody, because it depends on weather and visitors and a holiday nobody has taken yet. Both questions produce a number. Only one of them produces a number somebody else can check.

What is the whole of this resting on?

Four things, and every later step is one of the four being applied, so all four are worth taking slowly.

A scenario is a sentence about a period nobody observed. The form of the sentence therefore matters more than the arithmetic inside it. The arithmetic in a case is usually trivial and always checkable. The claim the sentence makes is not checkable. Two sentences carrying identical figures, to the rupee, can assert completely different things. The figures are the part that agrees, so no amount of rechecking them will tell the two sentences apart.

Second, a published set of accounts is a structure, and a structure is exactly what a conditional needs. Contribution a register times the number of registers gives contribution. Contribution less the standing base gives the operating resultWhat is left from a period's trading after the costs of running the business have come out, and before interest and tax are touched. It is a subtraction, so it is never a figure anybody sets.. Each of those relations holds as published, today, and goes on holding when one input is moved. The conditional is therefore true of the structure rather than of any period. A conditional needs no future period to be true, and supplying one turns it into a different kind of claim entirely.

Third, most of what is printed in a set of accounts is derived, and moving a derived line produces a structure nobody could report. An operating result is not movable; it is what is left. An operating margin is not movable; it is a division. Move the margin while holding the contribution and the standing base, and the division no longer returns the number printed beside it. The result is a set of accounts that could not be produced by any business anywhere. The movable quantities are the ones nothing else is computed from, and on this business's published lines there turn out to be four of them.

Fourth, and this is the one that catches careful people: a weight changes nothing in the arithmetic and everything in the claim. Put a number beside a case expressing how likely it is, and every rupee in that case stays precisely where it was. The writer’s assertion is what moves. A number beside a case claims knowledge of how the world will fall out, and knowledge of that kind is a different subject with different evidence behind it. The accounts contain none of that evidence. The cost of the addition is invisible because the figures do not change, and that is the whole reason it has to be refused outright rather than merely discouraged.

What actually separates a forecast from a scenario?

The difference between the two kinds of statement is settled elsewhere, in two sentences. A forecast claims its central case is what will happen. A scenario claims only that if these assumptions hold, then this follows. The whole difference sits in what is being asserted, and not in the numbers.

The distinction is not about the quality of the work. A forecast does not use worse arithmetic, or looser inputs, or less careful people. A forecast and a scenario make different kinds of claim while looking identical on the paper. Neither can be separated from the other by inspecting the figures, and a reviewer who checks the sums has checked the only part that was never in doubt.

So here is the practical test, and it is one question long. Does the sentence need a future period to be true? A conditional does not. A conditional is a statement about a structure as published, and it stays true whether or not anything ever moves, whether the business grows or shrinks, whether it is read tomorrow or in ten years. A projection does need one. Without a period there is nothing left for a projection to be about.

A case that cannot survive having its period removed was a forecast wearing a case's clothes. The test is mechanical and can be run on somebody else’s note without knowing the business at all. Every date, every year and every phrase like the year ahead is struck out of each sentence, and what is left gets read. Whatever survives is a conditional. Whatever collapses into nonsense was making a claim about a period, and a claim about a period needs the evidence that goes with one. Evidence about a period is a different and much heavier thing to carry.

Four neighbouring subjects already build the object itself, so here is the routing. A case itself, and how a set of cases is laid out and read, is covered under Scenario Analysis: How to Build Cases and Read the Range. How the case everything else is measured against gets set, and what it must not become, is covered under Base Case: How the Central Case Is Set and What It Must Not Be. Moving one input while everything else stands still is covered under Sensitivity AnalysisMoving a single input by a stated amount and recording how far the output travels because of it. The answer comes out as a rate rather than as a set of accounts.: Testing One Input at a Time. Designing a case that is deliberately hard on a structure is covered under Stress TestingPushing an input far enough to ask whether a structure holds together at all. The answer comes out as survives or does not, rather than as a figure.: How Severe-but-Plausible Cases Are Designed.

The four subjects above teach what a case is and how a set of them is read. Building one case on a single company’s own published lines is a narrower job, and its whole difficulty is that the accounts offer three or four movable quantities and no more. Everything that follows works that sentence out.

What can actually be moved in a set of published accounts?

Work it on real lines rather than in the abstract, because the abstract version of this question always sounds richer than the real one. Anjani Stationers Private Limited, an invented stationery works that trades nowhere, has its published year taken apart elsewhere at the level of a single register. In the published year it made 2,50,000 hard-bound registers and realised Rs 108.00/- on each. Of that, Rs 59.40/- went on paper and Rs 2.40/- on carriage and packing, leaving Rs 46.20/- of contribution a registerWhat is left out of one unit's sale once the costs that rise or fall with that unit have been taken off. Costs that stand still are not touched at this line.. Every one of those per-register figures multiplies back to a published total to the paisa. The closure is the only reason a stranger should accept them at all.

Add one more published line and the structure is complete: a standing baseCosts that carry on at the same size in a period no matter how many units come off the line. A shed rent, a supervisor's pay and an insurance premium all behave this way. of Rs 74,00,000/- for the same year, published elsewhere in these notes. Every line then sorts into two heaps, and the sorting is the skill.

Movable because nothing else is computed from them: the number of registers, the realisation a registerWhat the business actually gets for one unit sold, after any discount or return has already been taken off. It is a price as received rather than a price as listed., the costs that move with each register, and the standing base. Derived, and therefore not movable at all: revenue, contribution, the operating result and every margin. Each one of those is produced by the four above it.

Four quantities is the whole inventory, and a case that moves a fifth has moved something the accounts never contained. Four is a smaller supply than anybody expects who has been asked for three rich cases by the evening, and the disappointment is worth being blunt about. Four movable quantities is thin material for anybody hoping for a story. The thinness is the finding rather than a failing of the accounts, and a note that pretends otherwise has filled the gap with the writer rather than with the business.

A stall holder outside a station knows this without any vocabulary for it. He can change what he charges for a cup. He can change what goes into each cup and what it costs him. He can change the rent by moving pitch. The day’s takings are what those three produce once the crowd has walked past, so the takings are the one thing he cannot change directly. Anybody who hands him a plan whose first line is a bigger day's takings has skipped the only part he could have acted on.

How is an assumption named so a reader can check it?

Most notes say only that something has been assumed and then stop, and stopping there is what makes a case unreadable to anybody who was not in the room. Here is the shape that works, and then the shape working.

An assumption is named when a reader can tell, without asking anybody, which quantity was held, which was moved, to what, and on whose authority. Four parts, and every one of them in the same breath as the figure rather than in a footnote far below.

First, the quantity moved is named, along with the fact that it is the only one. Second, what it was moved to is named as a figure. A direction is not a case, and up a bit cannot be checked by anyone. Third, everything held is named as figures, rather than written off as the rest is unchanged. The rest is unchanged is not a figure, so a reader cannot rebuild the structure from it, and a reader who cannot rebuild the structure has to take the writer's word for the answer. Fourth, whose assumption it is gets stated. If it belongs to the illustration rather than to the business, that is said in those words. A reader has no way of telling from the outside.

An assumption a reader cannot locate is an assertion, and an assertion inside a case is indistinguishable from a figure the business published. The damage lands somewhere specific. The trouble is not that the reader disagrees with the assumption. The trouble is that six months later nobody, including the writer, can tell which of the figures in the note came out of the accounts and which came out of the writer.

The best argument for this is not a rule but a case already written. The account that walks this business's year down to what is left for the people holding the shares runs a case of its own, and it stops to say that in its two period setting the later period's residual is an assumption of the illustration and not a fact about any business. Then it prints the rule it used rather than the result alone. Both of those cost a sentence each. Both of them are the reason the case can still be checked by a stranger years later.

What does a worked conditional look like when somebody has already built one?

Rather than a second invented example, the one already standing serves. The account of who has a claim on what a year leaves runs a conditional on this same business, and it runs it properly, so what follows is finished work whose shape can be copied rather than an exercise to solve.

The published starting point is a single figure. The amount that reached the residual claimThe claim that gets whatever is left after everybody with a fixed entitlement has been paid. It is last in the order and it is the one with no promised amount. in the published year was Rs 30,00,000/-, and of that, Rs 0/- was distributed, so the whole of it stayed inside the business. The zero matters and is easy to skim past: the control there starts at Rs 0/- because a decision was taken and its published value happened to be nothing.

Now hold that Rs 30,00,000/- still and move it between two uses. Within one period the total never moves. One setting reads Rs 30,00,000/- and so does the other, and the line there is that every rupee handed to one claim is a rupee that cannot be handed to another. There is no arrangement in which both happen. Across two periods the total does move with the setting: one setting gives Rs 60,00,000/- across both and the other gives Rs 54,00,000/-, a gap of Rs 6,00,000/-.

Here is the part that makes it a conditional rather than a claim, and it is printed in the same breath as its figures rather than anywhere else. The later period's residual is an assumption of that illustration and not a fact about any business, and the rule it used is written out: if at least Rs 10,00,000/- is kept back, the later residual is taken as Rs 30,00,000/- again, and if less is kept back it is taken as Rs 24,00,000/-.

Nothing in that paragraph names a period, a date or a year, and every sentence in it is true of the structure as published. Read it back and check that for yourself. The test is being demonstrated rather than described. Two periods appear in it, but neither of them is a period on a calendar; they are the first and second slots of a structure, and the sentence would read the same to somebody picking this up in a decade.

One warning travels with that Rs 6,00,000/- and it costs a clause to carry. The Rs 6,00,000/- here is the gap between two settings measured across two periods in that illustration. The same figure to the rupee turns up elsewhere in these notes as an estimated rise in employee benefits, out of a completely unrelated subtraction. Check which subtraction produced a figure, never which figure it is. An arithmetic coincidence is the most dangerous kind of error: it arrives already wearing a confirmation.

What does the same form look like built on the published build of one register?

The hold is stated before the move, in figures, every time. Held here: the contribution at Rs 46.20/- a register, the realisation at Rs 108.00/- a register, and the standing base at Rs 74,00,000/-. Moved here: the number of registers, and nothing else.

Running it at the published volume first, before moving anything, is the check everything else depends on. At 2,50,000 registers, Rs 108.00/- each gives revenue of Rs 2,70,00,000/-. At Rs 46.20/- each it gives contribution of Rs 1,15,50,000/-. Taking the standing base of Rs 74,00,000/- out of that leaves an operating result of Rs 41,50,000/-. Every one of those three is the published figure for the published year, so the structure demonstrably closes before a single thing has been moved.

A conditional whose default fails to return the published year is sitting on a broken structure, and checking that comes first, before any case is built at all. The check takes one multiplication. Skipping it is how a note ends up with three internally consistent cases, none of which describes the business they were supposed to be about.

Now move the one quantity and read two further settings off the same structure. At 3,00,000 registers the structure would report contribution of Rs 1,38,60,000/- and an operating result of Rs 64,60,000/-. At 1,60,000 registers it would report contribution of Rs 73,92,000/- and an operating result of minus Rs 8,000/-. The contribution would stop just short of covering a standing base that has not moved.

Every one of those rows is written in the conditional, and no reason whatever is supplied for why the volume would sit anywhere in particular. The reason is the part nobody published, and it is also the part that would convert each row from a statement about a structure into a statement about the world. A tailor can say exactly what the month comes to at forty shirts and at ninety, down to the rupee, off the price and the cloth and the rent. He cannot say how many shirts will be ordered, and neither can anybody reading his accounts.

How is this different from moving one input at a time, or from a case built to be severe?

All three are built in full elsewhere, so what matters here is the boundary rather than the lesson. All three start from the same published structure, all three move something, and all three produce numbers. The three separate on what their answer is for.

Moving one input at a time asks how far the output travels for a given move in one input, and its answer comes out as a rate of change. A case built to be severe asks whether the structure holds together when an input is pushed hard, and its answer comes out as survives or does not. The held-structure conditional asks what the structure would report under a named hold, and its answer comes out as a set of accounts somebody could read.

The three differ by what their output is for, not by how the arithmetic is done, and all three are barred from carrying a weight for exactly the same reason: the accounts they were run on contain nothing about how the world falls out. Route each one by title and stop. Moving one input at a time is Sensitivity Analysis: Testing One Input at a Time. Pushing an input hard on purpose is Stress Testing: How Severe-but-Plausible Cases Are Designed. Laying a set of cases out and reading the range across them is Scenario Analysis: How to Build Cases and Read the Range.

How many cases should there be, and does the number mean anything?

Say the true thing first. The true thing is unwelcome and short. The number of cases is not evidence of anything, and three is a habit rather than a finding. A note carrying three cases is not better informed than a note carrying one. A reader who counts cases has measured the writer's effort rather than the business, and effort is not a fact about a company.

There is a count that means something, though, and it is a different one. How many quantities did each case move, and did each case move exactly one? A case that moves three quantities at once cannot be attributed, so nobody reading it can say which assumption produced which part of the answer. If the result looks wrong, there is nowhere to go and nothing to argue with. One moved quantity with its hold written out is a case a stranger can take apart and put back together, and that is the only property that survives being read by somebody who was not there.

The refusal form worth copying sits elsewhere in these notes, where four paths are drawn and then described: none of them is likelier than any other, none is central, and none is a scenario. The refusal declines a great deal while remaining useful. The account draws the paths, names what it is not claiming, and stops there.

And then the honest closing sentence. One act is one observation, and a reader who has watched one commitment through one year has learned the shape of the question and nothing about how often the answer comes out either way. The note says exactly that, rather than inventing more material to look thorough with.