Driver-Based Forecasting: Assumptions You Can Actually See

Driver-based forecasting builds every line from a quantity with a unit attached: units sold, rupees of realisation per unit, rupees of input cost per unit, headcount, a rate. A top-down forecast instead starts from a market total and a share assumption. Both are legitimate; only the first states what it assumes, and the two are reconciled against each other rather than chosen between.

Underneath that sits one idea established earlier. Building the revenue line takes revenue apart into volume multiplied by realisationThe average number of rupees a company gets for one unit it sells. Revenue divided by units. Realisation moves when prices move and also when the mix of what is sold moves., and showed that a single growth rate hides two separate assumptions inside itself. Driver-based forecasting generalises that discipline to every line, not just the top one, and adds a second route that starts from the size of the whole field rather than from the company. Everything below runs on Sarvani Coatings Limited, an invented paint maker, whose published year three ladder and the field it sells into are the only numbers used.

Where a research process should begin, at the field or at the company, is a different question from how a revenue number gets built. The sector material settled the first one. Once the decision to produce a revenue number has been taken, that number can be CONSTRUCTED in two ways, and the useful part is what happens when it is built both ways and the answers are put side by side.

What makes something a driver rather than a result?

Consider a tea stall outside a bus stand. Asked how business is, the owner might give two very different kinds of answer. One is: about two hundred and forty cups a day, at twelve rupees a cup. The other is: gas and milk take about a fifth of what comes in. Neither sentence is false. Only the first one, though, describes something that could be verified by standing at the corner for an hour with a counter.

A driver has to pass three tests at the same time. A driver carries a unit, so litres or cups or rupees per litre rather than a bare percentage. A driver is falsifiableCapable of being shown wrong. A statement that no possible observation could contradict is not falsifiable, however confident it sounds., meaning there is some observation that would show it was wrong. And in principle somebody standing outside the company could measure it, even roughly, without being let into the accounts.

A percentage of revenue satisfies none of the three tests, so it is a result rather than a driver. Sarvani Coatings in year three: other expenses were Rs 460 crore against revenue of Rs 2,415 crore, so 19.0 per cent. Writing 19.0 per cent of revenue into next year writes down nothing. The ratio carries no unit and cannot be measured from the street. No observation could contradict it either. The number was defined as whatever the ratio turns out to be. The figure is not an assumption. The figure is a placeholder wearing an assumption costume, and it will hold whatever value the revenue line hands it.

Top-Down Forecast: what is it, and when is it the only route available?

Now the tea stall becomes a wedding caterer. Asked how many functions she will do next season, she says there are about four thousand weddings in this town in a year and her firm gets about five out of every hundred. The caterer has just given a top-down forecast in one sentence. A total for the whole field, multiplied by a share of it.

The finance version is identical in shape. Take the field Sarvani Coatings sells into: Rs 48,300 crore at the end of year three, assumed to grow 11.0 per cent into year four. Growing at that rate gives Rs 53,613 crore. Sarvani Coatings did Rs 2,415 crore of revenue in year three, so its share of that field was exactly 5.00 per cent. Hold the share flat and next year is Rs 53,613 crore multiplied by 5.00 per cent, or Rs 2,680.7 crore. Two multiplications, one line of a spreadsheet, thirty seconds.

The route is quick, it is often the only one available, and its entire assumption is compressed into a single share number where nobody is obliged to look at it. Quick is obvious. Only available matters more than people admit: plenty of listed makers publish revenue and nothing underneath it, no litres, no tonnes, no realisation, no order book. Where no unit can be obtained, no build from units is possible, and a share based route is not a shortcut but the only door in the building. And the third part is the trap. A driver build wears its assumptions on the outside. A share assumption looks like a fact because it is one number, printed once, sitting in a cell that says 5.00 per cent.

What does building from drivers actually cost?

A driver build costs arguments. A driver build for one revenue line needs a volume assumption, a realisation assumption, and usually a segmentA line of business a company reports separately, such as decorative paint against industrial coatings. The split is disclosed in the accounts. mix assumption on top of both. Further down the ladder come input cost per unit, wages, advertising, freight, depreciation on a line not yet commissioned, and a tax rate. A dozen assumptions where the top-down route had two. Every one of them is a place where somebody can object: no, that one is wrong.

The extra arguments look like a bad trade until the gain on the other side becomes clear. The trade is more assumptions in exchange for correctable ones, and a model with three assumptions is not simpler than one with twelve if the three cannot be shown wrong. Twelve arguable numbers can each be attacked, defended, and revised one at a time when the next quarter arrives. Three unfalsifiable ones have nothing inside them to correct, so they can only be replaced wholesale. When somebody calls a model elegantly simple, the useful question is not how many assumptions it has. The useful question is how many of them could be shown wrong by something anybody could go and look at.

What happens when the two routes are run against each other?

Almost nobody does this, and it is the step that matters most. A driver build has produced a revenue number. Dividing that number by the size of the field gives a share. Not a share anybody chose. While the analyst was busy thinking about litres, the arithmetic chose one on its own.

The caterer again. She works out her season from her own diary: seventy functions last year, four more staff who can be taken on, so call it eighty five. Sensible, grounded, built from things she can count. But eighty five out of four thousand weddings is 2.13 per cent of the town against last year's 1.75 per cent, and she has just assumed she will take business from every other caterer in the district without once thinking about them. Nothing in her arithmetic mentioned competitors. The assumption arrived anyway.

A driver build converted back into an implied share will usually reveal a competitive claim that nobody made deliberately, and catching that claim is the entire purpose of the reconciliation. The gap between the two routes is not an error in either of them. Both can be internally perfect. The gap IS the assumption, sitting in the space between what was built and what the field would have allowed, and it is invisible until the two numbers are put next to each other.

So what does Sarvani Coatings year four look like down each route?

Work it properly, both ways, with every step visible. Start top-down. The field stands at Rs 48,300 crore at the end of year three and is assumed to grow 11.0 per cent, giving Rs 53,613 crore for year four. Hold Sarvani Coatings share of it flat at the 5.00 per cent it achieved in year three and revenue comes to Rs 2,680.7 crore. Against year three revenue of Rs 2,415 crore that is growth of 11.0 per cent. The match is arithmetically inevitable: hold share constant and the company grows exactly as fast as the field does.

Now nudge the share. Assume another gain of 0.13 percentage pointsThe plain difference between two percentages. Moving from 5.00 per cent to 5.13 per cent is a move of 0.13 percentage points. A move of 0.13 per cent is something else entirely., to 5.13 per cent, and revenue becomes Rs 2,750.3 crore, or 13.9 per cent growth. A 0.13 point share assumption reads as a rounding difference and is worth Rs 69.7 crore of revenue, or 2.9 per cent of the published year three. Thirteen thousandths of the field. Seventy crore of revenue. Nobody reading the model would stop at that cell.

Now build the same year from the drivers, and do it the honest way: just repeat what actually happened in year three. Year two revenue of Rs 2,120 crore became Rs 2,415 crore one year later. Across that single year volume grew 6.0 per cent and realisation grew about 7.47 per cent. Together those two growth rates are the whole of the published 13.9 per cent revenue growth. Apply both factors to Rs 2,415 crore and year four revenue is Rs 2,751.0 crore. No share was mentioned. No field was mentioned. Two quantities with units, both taken from the company's own most recent year.

Here is the reconciliation, and it is worth reading twice. Rs 2,751.0 crore of company revenue against a field of Rs 53,613 crore is a share of 5.1313 per cent. Sarvani Coatings held 5.00 per cent in year three. So repeating last year's drivers IS assuming another share gain of about 0.13 points, almost exactly the size of the gain just achieved, and the driver build never said so at any point. The difference between the two routes, Rs 70.4 crore, is not a mistake in either. The gap is a competitive assumption, and it only becomes visible when the two routes are set side by side.

One more thing worth labelling. Saying the gain just achieved was 0.13 points requires knowing the field a year earlier, and the record here gives the field only at year three. Holding the field's growth rate at the same 11.0 per cent for the previous year puts the field at Rs 43,513.5 crore, so Sarvani Coatings held 4.87 per cent then and 5.00 per cent now: a gain of 0.13 points. Holding the field's growth rate constant is an assumption, it is doing real work in the sentence above, and labelling it rather than leaving it buried is exactly the discipline set out below.

How is an assumption made visible?

Visible is not the same as present. An assumption is present in every model ever built. Visible means a reader can find it, understand what it is, and know what would change it, without having to ask the person who built the model.

Three things do that, and all three are cheap. First, a unit: 6.0 per cent volume growth, not just 6.0. Second, a label saying where the number came from, of which there are exactly three honest kinds. Disclosed means the company published it. Derived means the analyst calculated it from things the company published. Assumed means the analyst chose it. Third, and this is the one everybody skips, one sentence saying what would change it.

The sentence naming what would change the number is what turns a forecast into something that can be updated rather than replaced. Recording that the 6.0 per cent volume assumption would change if a quarter came in with volumes falling means that, when that quarter arrives, precisely one cell needs opening. Leaving it out means the next quarter simply produces unease about the whole model. Unease usually ends in a rebuild from scratch, and a rebuild is how views quietly become opinions that never lose.

Which assumptions are worth the effort?

Effort belongs where the answer is most sensitive to being wrong. The rule sounds obvious, and almost every model gets it backwards, for a completely understandable reason: effort tends to flow to whatever there is good published material about, and good published material is almost never available for the things that matter most.

Measure it on Sarvani Coatings and the pattern is stark. Get the volume growth assumption wrong by two points and year four revenue moves by Rs 51.9 crore. Get realisation growth wrong by two points and it moves Rs 51.2 crore. Get the field's growth rate wrong by two points and it moves Rs 48.3 crore. Get the share assumption wrong by that famous 0.13 points and it moves Rs 69.7 crore. Now get the effective tax rateThe tax charge divided by profit before tax. The effective rate differs from the headline rate because of allowances, past losses and timing. Sarvani Coatings paid 25.1 per cent in year three. wrong by two points, when it is printed in the accounts to the rupee, and revenue moves by nothing at all.

A model can be precise everywhere it does not matter and rounded exactly where it does, and that is the normal condition rather than the exception. The tax rate gets three decimal places because the accounts hand it over. The share of the field gets a round number because nobody publishes one. Then the round number carries seventy crore of revenue and the three decimal places carry nothing.

What is a driver build checked against?

Three things, and none of them is the analyst's own spreadsheet. The field's own volume record tests a volume assumption: a model that grows company volume at 6.0 per cent while the field it sells into has never done better than four is assuming share gain again, whether or not anybody meant it to. The field's pricing evidence tests a realisation assumption in the same way. And the company's own past drivers test both for plausibility. A maker that has never put realisation up more than eight per cent in a year is unlikely to manage fifteen next year merely because a model needs it to.

A forecast that nobody outside can check is a preference with arithmetic attached, however carefully it is laid out. This is the real reason the driver route beats the share route where both are available. The driver route is not the more accurate of the two, and often it is the less accurate. The gain is that each of its parts is exposed to something outside the model that could contradict it.

How does this get used on a working morning?

Three people use the same conversion, for three different reasons, and it is worth knowing which of the three is in play.

An analyst puts the implied share on the front sheet of the model, above the revenue line, so that anybody opening the file sees the competitive claim before the number it produces. In an investment committee the first question about a forecast is often not what growth rate was used but what share of the field that works out to. The share question cannot be answered by pointing at spreadsheet detail. A lender doing the same thing on a borrower's projections asks it more bluntly: the plan needs revenue up forty per cent, so who exactly is losing that business, and have they been told. And a household deciding whether the shop it runs can support a second outlet is doing the identical arithmetic in its head, usually without the field number. Leaving the field out is why the answer so often turns out to have assumed the whole street stops eating elsewhere.

In every one of those rooms the conversion is doing the same job: turning a private arithmetic claim into a public competitive one that somebody else can dispute. That is also why the step gets skipped. The conversion is the step that invites the argument.

Where does a driver-based forecast stop?

The work stops early. A finished driver build produces a set of forecast lines and a labelled list of the assumptions that produced them. The lines and the list are the whole deliverable. Turning them into a statement about the shares of Sarvani Coatings takes further steps the drivers do not supply: discounting each line at a discount rateThe rate used to bring a future rupee back to what it is worth today. The discount rate belongs to valuation, a later and separate subject., or applying a multiple to one of them.

The forecast and the view are two different objects. The discipline only survives while the two are kept apart. A model that runs straight from an assumption list into a number about a security has hidden the most arguable part of itself behind the most confident-looking part, which is precisely the failure the reconciliation step exists to prevent.