Reference
Frameworks: the intellectual foundation of value-driver modeling
Part of the Value Driver Modeling skill · loaded on demand from SKILL.md
Contents
- The McKinsey value-driver framework (Koller, Goedhart and Wessels)
- The key value driver formula, derived and interpreted
- Economic profit and why it is the working root metric
- Rappaport's shareholder value network (1986)
- Decomposing ROIC into operating drivers
- DuPont decomposition (Donaldson Brown, 1910s to 1920s)
- Identity versus causal link: the construction rule that keeps trees honest
- Baseline reconstruction and normalisation
- Deterministic sensitivity analysis and the tornado chart
- Scenario cases
- Back-of-the-envelope sizing
- How the pieces fit: from tree to decision
1. The McKinsey value-driver framework
"Valuation: Measuring and Managing the Value of Companies" was first published in 1990 by Tom Copeland, Tim Koller and Jack Murrin of McKinsey. Later editions are by Tim Koller, Marc Goedhart and David Wessels; the seventh edition appeared in 2020. The book's central claim has not changed across editions: a company's value is determined by its ability to earn a return on invested capital above its cost of capital, and by its ability to grow while sustaining that spread. Everything else (multiples, earnings per share, accounting profit) is a symptom or a shortcut.
The framework rests on four ideas the authors call the cornerstones of value (set out in the companion book, Koller, Dobbs and Huyett, "Value: The Four Cornerstones of Corporate Finance", 2010, and carried through "Valuation"):
- The core of value: value is created by earning a return on capital above the cost of capital. Growth creates value only when the spread is positive.
- Conservation of value: value is created by cash flows, not by rearranging claims on them. Financial engineering that does not change cash flows does not change value.
- The expectations treadmill: share-price performance reflects changes in expectations, so managing to the share price and managing to value are different jobs.
- The best owner: value depends on who owns the asset, because owners differ in what they can do with it.
For driver modelling the first cornerstone is the operative one. It gives the tree its root and its shape: the root must be a metric that charges for capital, and the branches must show both where profit comes from and where capital is consumed.
Why the framework was built: in the 1980s the standard corporate objective was earnings growth, and a generation of conglomerates grew earnings while destroying value by investing at returns below their cost of capital. The value-driver framework was a direct response. It makes the capital cost explicit, so that a manager cannot show growth without also showing the capital the growth consumed and the return it earned.
2. The key value driver formula, derived and interpreted
Start from the standard growing-perpetuity result for a business at steady state: value equals next year's free cash flow divided by the difference between the cost of capital and the growth rate.
Value = FCF / (WACC - g)
Free cash flow is NOPAT less net investment. Net investment is whatever must be reinvested to grow NOPAT at rate g. If the business earns ROIC on new capital, then to grow NOPAT by g x NOPAT it must invest (g x NOPAT) / ROIC. So:
FCF = NOPAT - NOPAT x (g / ROIC) = NOPAT x (1 - g/ROIC)
Substituting gives the key value driver formula:
Value = NOPAT x (1 - g/ROIC) / (WACC - g)
The term (1 - g/ROIC) is the fraction of profit that can be paid out after funding growth; Koller, Goedhart and Wessels call g/ROIC the investment rate. Three readings of the formula matter for modelling:
- If ROIC equals WACC, the formula collapses to NOPAT / WACC regardless of g. Growth adds nothing. This is the single most useful fact for pushing back on a growth story: ask what ROIC the growth earns.
- If ROIC is below WACC, higher g lowers value. Growth destroys value.
- If ROIC is above WACC, higher g raises value, and the effect is larger the wider the spread.
A numerical illustration using the worked example's baseline: NOPAT $15.0M, ROIC 12%, WACC 9%, g 3%. Value = 15.0 x (1 - 0.25) / 0.06 = 15.0 x 0.75 / 0.06 = $187.5M. Hold everything else and set ROIC to 9%: Value = 15.0 x (1 - 0.333) / 0.06 = $166.7M, and now changing g does nothing. Set ROIC to 18%: Value = 15.0 x (1 - 0.167) / 0.06 = $208.3M.
Limits of the formula, which you must respect in a client setting:
- It is a steady-state formula. It values a business that has already reached its long-run ROIC and growth. For a company in transition (a turnaround, a fast-growing unit), use an explicit forecast for the transition years and apply the formula only to the continuing-value period.
- It assumes ROIC on new capital equals ROIC on existing capital. If the marginal return differs (a distributor opening branches at lower returns than its mature ones), use marginal ROIC for the growth term.
- WACC must exceed g. If someone's plan has g approaching WACC, the formula explodes and so should your scepticism.
- It values steady-state cash flows. A one-off release of capital (working capital falling from 72 to 62 days) is a cash flow that happens once during the transition; add it to the steady-state value separately, and do not capitalise it as if it recurred.
3. Economic profit and why it is the working root metric
Economic profit for a single period is:
Economic profit = Invested capital x (ROIC - WACC)
which, since ROIC = NOPAT / invested capital, is identically:
Economic profit = NOPAT - Invested capital x WACC
The second form reads as profit after a charge for capital, and it is the form most useful in a tree because NOPAT and the capital charge become the two top branches. Koller, Goedhart and Wessels show (Appendix on economic profit and the key value driver formula) that the value of a company equals invested capital plus the present value of future economic profit, so the period metric and the value metric agree: maximise the present value of economic profit and you maximise value.
Why it is the working root for most focus and diagnosis questions:
- It is annual, so it reconciles to the P&L and balance sheet the client already has, unlike a DCF.
- It charges for capital, so it exposes growth that consumes capital at low returns.
- It is additive across units, products and branches, so the tree can be cut by segment and the segments sum.
- Its change year to year decomposes cleanly into a margin effect, a turnover effect, a growth effect and a cost-of-capital effect.
The metric is the same idea as Stern Stewart's EVA (Stewart, "The Quest for Value", 1991); the McKinsey term is economic profit. The arithmetic is identical; the adjustments to accounting figures differ in detail.
4. Rappaport's shareholder value network (1986)
Alfred Rappaport, "Creating Shareholder Value" (1986, revised 1998), gave the first widely used map from operating decisions to shareholder value. His network runs: corporate objective (shareholder value) at the top; valuation components (cash flow from operations, discount rate, debt) below; then a small set of value drivers; then management decisions (operating, investment, financing) at the base. The seven value drivers are: sales growth rate, operating profit margin, income tax rate, working capital investment, fixed capital investment, cost of capital, and value growth duration (the length of time the business can earn returns above its cost of capital).
Rappaport also introduced the threshold margin: the operating margin on incremental sales at which growth exactly earns its cost of capital. Below the threshold, growth destroys value; it is the same insight as the key value driver formula from a margin rather than a return perspective, and it is the origin of the discipline of asking what margin the growth carries.
What Rappaport supplies to this skill: the list of drivers is a checklist for the top two levels of any tree. If a tree does not have a node for each of the seven, something has been left out, usually working capital or the duration of advantage.
5. Decomposing ROIC into operating drivers
Koller, Goedhart and Wessels take ROIC apart in two steps and then keep going.
Step one, the DuPont-style split:
ROIC = (NOPAT / Revenue) x (Revenue / Invested capital) = NOPAT margin x capital turnover
Step two, each side into its components:
NOPAT margin = Gross margin - (operating cost / revenue) - (depreciation / revenue), all after tax Capital turnover = 1 / (working capital / revenue + fixed capital / revenue)
Step three, each component into the operating metrics a manager sees:
- Gross margin into price realisation (list price, discount, rebate), mix, and unit cost.
- Operating cost / revenue into cost per unit of activity (cost per order, per delivery, per head) times activity per unit of revenue.
- Working capital / revenue into days: receivables on revenue, inventory on cost of goods sold, payables on cost of goods sold.
- Fixed capital / revenue into capacity utilisation, capex per unit of capacity, asset life.
The authors' point is that the decomposition should stop at the level where a line manager can be held accountable: not "capital turnover" but "inventory days at branch level"; not "margin" but "average discount from list on the top 200 SKUs". They also warn against decomposing symmetrically for its own sake: go deep on the branches where the business's value is concentrated and shallow elsewhere.
6. DuPont decomposition (Donaldson Brown, 1910s to 1920s)
The DuPont identity is credited to Donaldson Brown, a DuPont engineer turned financial officer, who used it from around 1914 and carried it to General Motors in the 1920s. The original form:
ROE = (Net income / Sales) x (Sales / Total assets) x (Total assets / Equity) = Net margin x Asset turnover x Equity multiplier
The five-factor extension separates tax and interest:
ROE = (Net income / Pre-tax income) x (Pre-tax income / EBIT) x (EBIT / Sales) x (Sales / Assets) x (Assets / Equity) = Tax burden x Interest burden x Operating margin x Asset turnover x Equity multiplier
For value-driver work the operating form (ROIC = NOPAT margin x capital turnover) is the one to use, because it strips out financing and tax so that the tree describes the business rather than its balance-sheet structure.
Why DuPont matters beyond the arithmetic: it tells you what kind of business you are looking at, and therefore which branch of the tree deserves depth. A 12% ROIC can be:
| Type | NOPAT margin | Capital turnover | Where the tree should go deep |
|---|---|---|---|
| Distributor | 3.75% | 3.2x | Gross margin, working-capital days; a point of margin is a third of profit |
| Manufacturer | 8% | 1.5x | Utilisation, yield, capex per unit; fixed assets dominate |
| Utility | 24% | 0.5x | Asset base, regulated return; margin is largely given |
| Services firm | 15% | 0.8x | Utilisation and rate; receivables and WIP are the capital |
The same return, four different sets of levers. A tree that does not begin with this split will decompose the wrong branch.
7. Identity versus causal link: the construction rule that keeps trees honest
A driver tree contains two kinds of relationship, and most bad trees fail because they do not distinguish them.
An identity holds by definition. Revenue = volume x price. Gross profit = revenue - COGS. Economic profit = NOPAT - capital charge. The children reproduce the parent exactly, in every period, with no hypothesis involved. If the children do not sum, there is a unit error or a missing term.
A causal link is a claim. "Order frequency is driven by fill rate." "Discount depth is driven by sales-rep incentive design." It may be true, partly true, or false, and it holds with a lag, a coefficient and noise. It cannot be added to an identity tree as a sibling because it has no arithmetic path to the root.
The construction rule: decompose with identities until every leaf is a metric that appears in an operating report; then hang causal links below the identity leaves, labelled as causal, each with a proposed test (a correlation across branches, a controlled trial, an interview finding). This gives three guarantees:
- The tree always reconciles to the accounts, because the identity layer is arithmetic.
- Every hypothesis is visible, so
hypothesis-stress-testingcan pick it up. - Sensitivity can be run on the identity leaves without argument, because moving a leaf moves the root by construction. Sensitivity on a causal link requires the coefficient, which is a separate piece of evidence.
Common places the rule is broken: "customer satisfaction" placed beside "price" under revenue; "employee engagement" beside "cost per order"; "brand strength" beside "volume". Each is a plausible causal driver and each belongs below the leaf it drives, with a test.
8. Baseline reconstruction and normalisation
The tree is only as good as the number at its root, and the root comes from the client's accounts, which were not prepared for this purpose.
Reconstruction, in order:
- EBITA from reported operating profit, adding back amortisation of acquired intangibles (which is a purchase-accounting artefact, not an operating cost) but not depreciation (which is).
- Operating taxes at the effective cash rate applied to EBITA, not the reported tax charge, which reflects financing and one-offs.
- NOPAT = EBITA - operating taxes.
- Invested capital = operating working capital + net PP&E + capitalised operating intangibles + other operating assets - other operating liabilities. Exclude cash above operating needs, financial investments, and non-operating assets. Include operating leases capitalised if the lease accounting standard does not already do so.
- ROIC = NOPAT / invested capital, on average capital for the year where possible.
Normalisation catches the things that would make a single year unrepresentative: one-off charges and gains, partial-year acquisitions, mid-year price changes, timing of payables and stock counts around year-end, and accounting policies that shift cost between the P&L and the balance sheet. Every adjustment goes into a reconciliation table from reported to normalised, so the client's CFO can see and challenge each one. A baseline without a reconciliation table is an assertion.
9. Deterministic sensitivity analysis and the tornado chart
The tornado diagram comes from decision analysis, the discipline built at Stanford and SRI by Ronald Howard and colleagues from the 1960s (Howard, "Decision Analysis: Practice and Promise", Management Science, 1988, describes the practice). In decision analysis the deterministic sensitivity phase precedes any probabilistic work: it identifies which of the many uncertain inputs actually matter, so that effort on probability assessment goes only where it changes the decision. The tornado chart is its output.
The procedure:
- Define the output (the root metric) and the base case (all inputs at base).
- For each input, set a low and a high value. In decision analysis these are typically the 10th and 90th percentiles of the assessor's belief; in consulting practice they should be evidence-based: historical range, peer range, dispersion across segments, or a labelled judgement.
- For each input in turn, set it to low with all others at base, record the output; set it to high, record; return to base.
- Swing = |output at high - output at low|.
- Sort inputs by swing, descending.
- Plot horizontal bars from output-at-low to output-at-high, one per input, top to bottom in swing order, around a vertical line at the base output. The result is widest at the top and narrows going down, hence the name.
What it tells you: the ranking of inputs by their influence on the output, given the ranges. What it does not tell you: anything about interactions (the method is one-at-a-time), anything about probability (a wide bar with a narrow probability may matter less than a narrow bar that is certain to move), and anything about controllability (the top bar may be an external factor).
Three disciplines that separate a useful tornado from a decorative one:
- Ranges have sources. A chart built on plus-or-minus 10% everywhere is a chart of the modeller's convenience. Its ranking is meaningless because a variable with a real range of plus-or-minus 2% (tax rate) is being treated like one with plus-or-minus 25% (volume).
- Bars bracket the base. If an input's low and high outputs are on the same side of the base, the model has a sign error or a non-monotonic relationship, and the chart is wrong.
- The reading is written down. A chart without the sentence "the top three explain X% of total swing" has not been interpreted.
A spider chart (output plotted against each input varied across its range, all on one set of axes) shows the same information with the slopes visible, and reveals non-linearity that the tornado hides. Use it as a check when a relationship is not linear; present the tornado.
10. Scenario cases
Tornado analysis is one-at-a-time; the world is not. Scenario cases exist to move drivers together in coherent sets.
A scenario in this skill is a named set of assumption values on the assumptions sheet, selectable by a switch, not a copied model tab. The usual set:
- Base: the normalised baseline, carried forward with no management action beyond business as usual.
- Management cases: one per coherent programme. A commercial programme moves margin and modest volume together; a working-capital programme moves days together; a cost programme moves fixed cost and perhaps volume (if capacity is cut).
- Combined plan: the management cases together, with any interactions handled (margin applied to the higher revenue, working capital days applied to the higher COGS).
- Downside: what a sceptical board member would propose, usually a demand shock hitting volume, margin and receivables together.
For each, the output row is the same six numbers: EBITA, NOPAT, invested capital, ROIC, economic profit, value. Value at stake is the plan minus base, presented with the downside beside it. A point estimate without a range is not a finding; it is a hope with decimals.
Scenarios here are not the Shell-style scenario planning of war-gaming, which builds alternative external futures from critical uncertainties. These are internal cases on a single model. When the downside depends on what competitors do, hand that branch to war-gaming.
11. Back-of-the-envelope sizing
Enrico Fermi's estimation method, applied to value: decompose the quantity into a few factors you can estimate to within a factor of two, multiply, and see whether the answer is $5M or $50M. In value-driver work the envelope has three uses:
- Deciding whether to model at all. If the envelope shows the top lever is ten times the second, the ranking is settled and the model is a formality.
- Catching errors. A model that says a point of margin is worth $400M on a $400M-revenue distributor has a units problem, and the envelope ($4M EBITA, $3M NOPAT, roughly $40M to $50M of value) says so instantly.
- Talking to the client. A CEO remembers "a point of margin is $3M after tax and about $50M of value"; a CEO does not remember cell F47.
The envelope rule that ties to the model: because value = invested capital + present value of economic profit, and at steady state that present value is economic profit / (WACC - g), the value of a permanent change is (change in annual economic profit) / (WACC - g). Capital released or consumed during the transition is already inside that number, since the economic profit change includes the capital charge. With WACC 9% and g 3%, every $1M of permanent economic profit is worth $16.7M.
The standard envelope lines for a value tree:
- One point of gross margin = revenue x 1% x (1 - tax) of NOPAT and of economic profit, divided by (WACC - g).
- One day of working capital = (base / 365) of capital released once; the annual economic-profit gain is that amount x WACC; value is the gain divided by (WACC - g).
- One point of volume growth at constant margin = contribution margin x revenue x 1% of EBITA after tax, less WACC on the working capital and capex it consumes, divided by (WACC - g).
- One point of WACC = invested capital x 1% of annual charge, but (WACC - g) also moves, so use the model; for a WACC of 9% and g of 3% each point of WACC moves value by roughly 15%.
12. How the pieces fit: from tree to decision
The sequence the skill follows, and the reason for the order:
- Root metric first, because it determines what the tree charges for.
- Baseline second, because a tree hung on an unreconciled number cannot be defended.
- Identities third, causal links fourth, because the arithmetic must be certain before the hypotheses are added.
- Envelope fifth, because it sets the scale and decides whether the model is worth building.
- Model sixth, with hygiene, because the client's finance team will trace cells.
- Tornado seventh, because it ranks the drivers given the ranges.
- Scenarios eighth, because they add the interactions the tornado omits and produce the value-at-stake range.
- The driver note last, because everything before it exists to produce three sentences a CEO can act on.
The frameworks give the model its logic: Koller, Goedhart and Wessels supply the root and the spread; Rappaport supplies the checklist of drivers; DuPont supplies the first split and the diagnosis of business type; the identity rule supplies the honesty; decision analysis supplies the ranking; scenarios and the envelope supply the range and the sanity check.