Incremental Cash Flow: The Only Cash Flows That Count

An incremental cash flow is one that exists only because of the decision being valued. Sankalp Industrial Systems Limited, an invented manufacturer, runs that test twice: once on what enters the numerator at all, and once on what belongs to the operating business rather than to somebody else. Interest, the working capital balance and two assets outside the forecast all fail it.

Underneath the definition sits a subtraction that is easy to say and hard to perform. Value the world with the decision, value the world without it, and the difference is the decision's worth. Everything difficult about a valuation is difficulty in constructing the second half of that subtraction. The first half is a forecast, and forecasts are what modellers are trained to build. The second half is a description of a future that nobody writes down anywhere, and it is where most of the honest disagreement in this subject lives.

Incremental to what, exactly?

Start with the question nobody asks out loud. A cash flow is never incremental in the abstract. Every cash flow is incremental to something, and until that something has a name, an argument about whether a particular rupee counts cannot be settled. The two people arguing are measuring against different worlds.

Think about a delivery rider deciding whether to buy a scooter. The scooter earns money, so the earnings look incremental. But incremental to what? To standing still and earning nothing? To carrying on with a borrowed cycle and a smaller round? To renting a scooter by the day? Each of those alternatives gives a different answer, and only one of them is the alternative the rider actually faces. Name the alternative first, and the arithmetic afterwards is usually straightforward.

A valuation runs the same discipline on a company. For Sankalp Industrial Systems Limited the alternative is not nothing. The alternative is the business as it stands: Rs 12,00,00,00,000 of revenue in the base year, earnings before interest, tax, depreciation and amortisation, or EBITDAA profit line taken above four deductions: interest, tax, depreciation and amortisation are all still waiting to come out of it., of Rs 2,88,00,00,000, and a plant that will need money spent on it every year simply to keep producing what it already produces. Against that alternative, growth is expensive and the model has to say so.

How does the with-and-without test actually run on a line of a model?

Take one line of the forecast and ask a single question of it: would this rupee still move if the decision being valued were never taken? An honest no means the decision caused the movement, so the line goes into the numerator. An honest yes means the rupee stands on both sides of the subtraction and cancels out, however large and however real it happens to be.

The test is not a tidying pass applied at the end of a build. The test decides what the model contains before a single cell is typed. Applied late, it becomes an argument about a spreadsheet somebody has already grown attached to; applied first, it settles the shape of the whole exercise in an afternoon.

Why is the second half of the subtraction the one that goes wrong?

There are two lazy versions of it, and between them they account for most of the bad valuations a reader will ever meet.

The first lazy version says nothing happens. Revenue holds, margins hold, the plant keeps producing, and no money is spent. The first lazy version is not a description of a business; it is a description of a photograph. Sankalp Industrial Systems Limited spends Rs 52,80,00,000 of capital expenditureSpending that buys or renews a long-lived asset, as against spending that is consumed inside the year it happens. in Year 1 purely against the depreciation charge, before a single rupee of growth has been paid for. A world in which the existing capacity persists for free has handed the decision a gift worth exactly that maintenance spending.

The second lazy version says the business quietly declines. Sometimes that is right, and in a business losing a contract or facing a substitute it is the only honest description available. Because a slow slide makes any decision look good by comparison, it is chosen far more often than it is argued for. If the without case is a slow slide, then almost anything beats it, and the model has produced a number that says more about the modeller's assumed decline rate than about the decision.

The discipline is unglamorous: write the second half down, in sentences, before touching the arithmetic. Name what is spent, what is kept and what is lost. Then check whether anybody outside the room would recognise it as the alternative the company actually faces.

What does the test remove when it is run on a real forecast?

Here is the whole of it, item by item, for Sankalp Industrial Systems Limited in Year 1 of its five year forecast.

Read the last row twice. Ten items went into the test and four survived, and the four that survived are not the four a reader would have picked. The items that fail are the ones people are most confident about, and no two of them fail for the same reason. The categories below matter more than the list for exactly that reason: a reader who memorises the list is helpless the first time a new item arrives, and a reader who has the four categories can place almost anything.

Notice also that there is no dial to turn anywhere in this. Every judgement here is a yes or a no, taken one line at a time, and nothing about it becomes clearer by being made continuous. A straight yes or no is unusual in valuation, where most of the difficulty is a matter of degree, and it is the reason this particular skill is learned by working through cases rather than by watching a number move.

Why is interest left out when it is plainly cash going out of the door?

Call this the first category. Missing it is the most expensive mistake available in a valuation. Interest of Rs 48,00,00,000 leaves Sankalp Industrial Systems Limited in Year 1. Nobody disputes it, no accountant estimated it, and the bank statement shows it going. Interest still has no place in this numerator.

The reason is a matching rule rather than anything about interest itself. A firm-level cash flow is built to be shared out among every funder the business has, and a lender sits squarely inside that group. The lender's return has already been priced into the weighted average cost of capitalBlend what shareholders want against what lenders charge, weight each by how much of the funding it supplies, and the result is the rate a whole-business cash flow is discounted at. of 12.00 per cent this company applies, a rate in which borrowed money makes up a quarter of the mix and is priced after tax at 6.00 per cent. Strip the interest out of the cash as well, and the lender has been settled with twice: once above the line and once below it.

One line prevents it, and the line is worth memorising: whoever gets paid in the cash flow has to be the same group getting paid in the rate. Cash for the whole business, discounted at the whole business's blended rate. Or take what remains once the lenders have had their share, and discount that at the return shareholders on their own require. Never a mixture. The matching rule disposes of interest, of loan repayments, of fresh borrowing and of dividends at a stroke. Each of those four is a transfer between the business and one class of funder rather than cash the business produced.

Why does the movement in working capital count and the balance not?

Category two catches a confusion between two different objects rather than a lapse of judgement. A stock is what is there. A movement is how much it changed. Only one of the two is a cash flow, and it is never the first.

A vegetable seller keeps Rs 5,000 in the cash box so she can pay her supplier at four in the morning before the day's takings arrive. The Rs 5,000 in the box is not a cost of trading tomorrow. The money was found once, years ago, and it sits there. But when she starts supplying a second stall and needs Rs 6,000 in the box instead, the extra Rs 1,000 has to come from somewhere tomorrow. The float is a stock; the increase in the float is the cash flow.

Sankalp Industrial Systems Limited holds Rs 1,80,00,00,000 of net working capitalMoney the operating cycle absorbs: what customers still owe plus what sits in the store, less what suppliers are still waiting to be paid., being exactly 15.0 per cent of base year revenue. The balance was funded before the forecast opened and is already part of the capital the base year earns its return on, so it never appears in the forecast. The plan changes the size of the balance. Revenue rises by Rs 1,20,00,00,000 a year, working capital is held at the same 15.0 per cent of revenue throughout, and so Rs 18,00,00,000 a year has to be found. The Rs 18,00,00,000 is the increment, and it appears in every one of the five years.

Depreciation is not cash at all, so why is it in the arithmetic twice?

Here is the third category, and of the four it is the one likeliest to make a reader dismiss the whole exercise as accounting trickery. The double treatment is nothing of the kind. A charge nobody paid cash for is sitting in the middle of a calculation whose entire job is to produce a tax bill, and cash very much does leave for that.

Follow Year 1 through. EBITDA of Rs 3,16,80,00,000, less the depreciation charge of Rs 52,80,00,000, leaves operating profit of Rs 2,64,00,00,000. Tax at an assumed 25.0 per cent, not a measured one, comes to Rs 66,00,00,000, and it is genuine money leaving a genuine bank account. NOPAT of Rs 1,98,00,00,000 emerges at the far end. Then the Rs 52,80,00,000 goes straight back in. No cheque was ever written for it.

Deduct and add back cancel each other exactly, and the tax saved by the deduction is what survives. Compare two tax bills. Had the charge never been taken, taxable profit would have stood at Rs 3,16,80,00,000 and the bill at Rs 79,20,00,000. With it, the bill is Rs 66,00,00,000. Rs 13,20,00,000 separates the two, a quarter of the charge, and that single figure is the complete cash consequence of depreciation in Year 1. The machine itself was paid for over in the capital expenditure line, where cash genuinely moved.

What happens to cash the forecast was never built to include?

Category four is where a reader most often assumes something has been thrown away. Nothing has been thrown away. Failing the incremental test decides how an item is valued, never whether it is valued.

A shopkeeper runs a hardware store and has bought the vacant plot behind it. Valuing the trade of the store values the trade of the store and nothing else. The plot has not become worthless because it produces no counter sales; it is simply a different asset and needs a different method. Forgetting it entirely is the only real error available here.

The Sankalp Industrial Systems Limited forecast was built on one number: Rs 2,88,00,00,000 of base year EBITDA, growing at a fixed 24.0 per cent margin on rising revenue. Three items produce none of that number. The surplus land at Rs 45,00,00,000 produces no trading profit. The 26.0 per cent holding in Aruna Tooling Private Limited, invented, is equity accountedA stake large enough to give influence but too small to control, so the accounts carry a share of its profit rather than a share of its sales., so its trading never entered consolidated EBITDA in the first place. And the Rs 1,20,00,00,000 of cash sits outside the working cycle, with Rs 40,00,00,000 of it needed to run the business and Rs 80,00,00,000 surplus to it.

All three leave the operating model and come back afterwards at their own values, in a walk from enterprise valueWhat the whole operating business is worth to everybody funding it, before anybody's individual claim on that value has been separated out. to what a share is worth. The land and the holding together are added at Rs 1,00,00,00,000, and the cash at its full balance.

How is a shared cost treated when only part of a business is being valued?

Two brothers run two stalls in the same market and share one delivery van. Ask what the second stall is worth on its own, and the van becomes a genuinely hard question. Close the second stall and does the van cost disappear? Some of it, probably. All of it, almost certainly not. The first stall still needs deliveries. Nobody can settle the split from the outside, and pretending otherwise by dividing the van cost in half is a decision dressed as a calculation.

Sankalp Industrial Systems Limited has the same problem at a larger scale. Consolidated EBITDA is Rs 2,88,00,00,000, and the three divisions sum to Rs 2,97,00,00,000. The Rs 9,00,00,000 difference is head office cost that has been pushed down to no division at all. Now try to value the aftermarket parts and service division on its own, at Rs 1,80,00,00,000 of revenue and Rs 54,00,00,000 of EBITDA, and the incremental question arrives directly: would any of that Rs 9,00,00,000 actually disappear if the division did?

Some head office cost is genuinely attributable and some is genuinely shared. The split cannot be settled from outside the company, and the honest treatment is to state the question and to call the answer a judgement. The bounds are the only part of it that is arithmetic, and they are worth writing down: if every rupee of the head office cost were attributable to the aftermarket division, its standalone EBITDA would read Rs 45,00,00,000, and if none of it were, the figure would stand at Rs 54,00,00,000. A valuation that allocates head office cost in proportion to revenue has picked a point on that range without telling the reader it was picking.

What is a sunk cost, and why is one line enough to dispose of it?

A sunk cost is money already spent. A sunk cost shows up at exactly the same size in both halves of the subtraction, so it cancels, and one line is all the treatment it needs.

Notice how narrow that argument is. The argument makes no claim at all on three questions people expect it to settle: whether the money mattered, whether spending it was wise, and whether anybody should reflect on it. The claim it does make is narrower and much stronger. The money has already gone in every future available, so it cannot tell one future from another. An item identical on both sides of a difference contributes nothing to the difference.

The deposit paid on a shop that turned out to be in the wrong lane is the everyday version. The deposit hurts, it was real, and it is not coming back whether the trader stays or leaves. The only question with any content is which of the two remaining futures is better from here. Including the deposit in that comparison is how a bad decision gets defended by reference to what it has already cost. Preventing exactly that damage is why the rule exists.

What does Rs 1,00,00,00,000 a year of new capital actually buy?

Everything so far has been about what to exclude. Here is the other half of the idea, and it is the half that makes a forecast worth reading: an incremental benefit has an incremental cost, and the ratio between them is the most interesting number in a model.

Sankalp Industrial Systems Limited sinks Rs 1,00,00,00,000 of new capital into the business every year for five years. Three lines build that figure, and the Year 1 build is set out below.

Why does that total refuse to move? Because the first two lines of the table are separated by a constant gap of Rs 82,00,00,000, year after year. Spending was pitched at a shrinking percentage of a growing revenue line, and a shrinking percentage of a growing line is exactly what holds a rupee gap still while both figures climb. Add the flat working capital movement to that steady gap and the Rs 1,00,00,00,000 falls out on its own.

Against that outlay, NOPAT rises by exactly Rs 18,00,00,000 every year. Divide the gain by the outlay and the return on the new money is exactly 18.00 per cent. The 18.00 per cent is an incremental ratio and nothing else: fresh profit over fresh capital, silent on how profitable the existing plant happens to be. Sankalp Industrial Systems Limited already carries Rs 12,00,00,00,000 of invested capitalAdd up the money currently tied up in the business by everybody who funded it, whether it sits in machinery or in the operating cycle. and earns Rs 1,80,00,00,000 on it. The older money therefore returns 15.00 per cent.

So this model expects fresh money to outperform old money by three points. The three point gap is an assumption rather than a measurement, and no evidence anywhere in the forecast argues for it. The claim is also sharper than it looks. Every rupee of revenue in this forecast converts to profit at the same 15.00 per cent, new plant and old plant alike, so margin cannot be what separates the two returns. Turnover is. Fresh capital of Rs 1,00,00,00,000 is expected to throw off Rs 1,20,00,00,000 of sales, or 1.20 turns; the plant already installed runs at one turn, matching its capital rupee for rupee against its revenue. The assumption is entirely about how hard new machinery is expected to work.

Koller, Goedhart and Wessels tie three quantities together in one line: how fast profit grows, what freshly sunk capital earns, and the value of the business. The Sankalp forecast is that one line running. Multiply the share of profit put back in by the return that money earns, and out comes the growth rate. Four consecutive years agree to the second decimal.

The profit the share is measured against keeps growing, so the share put back falls every year while the rupee outlay holds still. The falling share is the whole reason a forecast adding a fixed Rs 1,20,00,00,000 of revenue a year shows a growth rate that slides downward, and the slide is arithmetic rather than a view about the business.

Why is a list of projects never added to a forecast capital expenditure line?

With both figures in view, a sharp reader arrives here inside a minute. The question is a fair one, and it has a straight answer.

Sankalp Industrial Systems Limited has five numbered projects under review, and their outlays total Rs 4,15,00,00,000. The forecast, meanwhile, spends Rs 1,34,80,00,000 in Year 1. The two figures refuse to reconcile. Each of them was built on a footing the other does not share, so refusing is correct.

The forecast line is an approved run rate that already exists. The run rate keeps the plant producing, and it delivers the Rs 1,00,00,00,000 a year of new capital that the growth assumption leans on. Sitting inside the model, every rupee of it has already worked its way through the free cash flow being discounted. The project list is a different exercise altogether: five proposals, not one of them approved, each held up against a 12.00 per cent bar identical to the rate applied to the forecast. Whether approving one would raise the run rate or come out of it has never been established, and so the question stands open.

A shop owner has the same two objects. There is the money spent every year on paint, a new shutter and a replacement fridge, the plain cost of keeping the shop open. And there is the proposal to open a second branch. Both are capital spending. Adding them together to describe next year's plan would be wrong unless somebody had first established that the branch is additional to the maintenance rather than instead of it.

Is the tax consequence itself an incremental cash flow?

Tax is an incremental cash flow, and treating it as one closes a gap that catches a surprising number of otherwise careful models.

Tax is not a residue that drops out at the bottom of a calculation. Tax is a payment in cash that moves when the decision moves, and so it is every bit as incremental as the revenue behind it. Sankalp Industrial Systems Limited hands over Rs 66,00,00,000 in Year 1 on the assumed 25.0 per cent rate, and that figure shifts with every operating line above it.

The practical consequence is that a tax effect can be the whole of an item's cash relevance, even when the item itself is not cash. Depreciation is the standing example here: no cash left for the charge, and yet it changes the cash paid to the tax authority by Rs 13,20,00,000. The same logic runs the other way for the interest deduction. Borrowing reduces a company's tax bill, and that reduction is real cash. The relief is simply accounted for in the discount rate rather than in the numerator, and so this company's cost of debt is stated after tax at 6.00 per cent rather than at the blended 8.00 per cent it actually pays.

How does somebody outside the model use this test?

Three readers use it every week, and none of them builds a spreadsheet to do it.

A lending officer assessing a facility for growth asks the incremental question in its cash form. The company wants Rs 1,00,00,00,000 a year. How much does that buy, and when does the cash come back? On these numbers the answer is Rs 18,00,00,000 a year of additional operating profit, arriving from the following year onward. The officer is not valuing anything; they are checking that the increment being funded produces an increment large enough to service what is being lent against it.

An analyst reading somebody else's model runs the test as an audit rather than as a build. Four questions cover most of it. Is interest anywhere in the numerator? Is a balance sitting where a movement should be? Has any non-cash charge been treated as though cash left? And is anything that produces no operating profit still inside the forecast? Four questions, and between them they catch the great majority of first models.

A household applies the same test without ever meeting the vocabulary. Working out whether one member should take a job in another city, the honest sum is never the salary on its own. Take the salary. Deduct the rent that would not otherwise be paid. Deduct the fares home. Deduct whatever earnings are given up at this end. Then adjust for whatever the running costs at home do in response. Each of those lines exists only because of the move, and every cost that would have been paid either way stays out of the sum. Most people run this test correctly on their own lives and abandon it the moment the subject is a company.

Sankalp Industrial Systems Limited and Aruna Tooling Private Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.