Valuation Model and Scenario Engine

A DCF has a small number of load-bearing parts, and getting each one internally consistent is most of the craft.

Free cash flow is the cash a business throws off after it has paid to keep running and to grow: operating profit after tax, plus non-cash charges, less reinvestment in working capital and capital expenditure. In an enterprise DCF, the convention followed in Koller, Goedhart, and Wessels is to project unlevered free cash flow, the cash available to all capital providers before financing, so that the effect of leverage lives in the discount rate rather than in the cash flows themselves.

The discount rate converts future cash into today's dollars. For unlevered free cash flow it is the weighted average cost of capital (WACC), the blended after-tax cost of debt and cost of equity, weighted by the firm's target capital structure. The CFA Institute curriculum states the consistency rule plainly: free cash flow to the firm is discounted at the WACC, and free cash flow to equity at the cost of equity. Mixing the two, discounting equity cash flows at a WACC, is one of the more common ways a model quietly overstates or understates value.

The terminal value captures everything beyond the explicit forecast. Because you cannot project year by year indefinitely, you forecast a handful of years explicitly, until the business reaches a steady state, and then capitalize the final year with a continuing value. The constant-growth (Gordon growth) form grows the final-year free cash flow one period and divides by the spread between the discount rate and the perpetual growth rate. That single figure often accounts for a large share of the total value, which is why its assumptions deserve the most scrutiny rather than the least.

The difference between a valuation a board can lean on and one that only looks precise comes down to discipline on a few points that the sources return to again and again.

• Keep the cash flows and the discount rate consistent. Nominal cash flows go with a nominal rate, a given currency with its own rate, and unlevered cash flows with the WACC. Damodaran treats this internal consistency as one of the first things to check, because a mismatch can move the answer more than any single assumption.
• Discipline the terminal value. The perpetual growth rate has to sit below the discount rate, or the Gordon denominator collapses, and Koller and colleagues caution that a company is unlikely to outgrow the broad economy indefinitely, so a sensible terminal growth rate is bounded by long-run nominal economic growth. A useful check is to see what share of enterprise value sits in the continuing value, and to look at the growth and returns that value implies.
• Ground the assumptions rather than reverse-engineer them. Growth, margin, and reinvestment should trace to the company's own history and to the economics of its industry, not be tuned backward to hit a number someone already had in mind.
• Show the range, not a false point. A DCF output is an estimate carrying real uncertainty, so sound practice pairs the base case with sensitivity and scenario analysis on the assumptions that move it most, typically the discount rate and terminal growth.
• State every input. A reviewer should be able to see each assumption and follow it into the model. An answer no one can trace is an answer no one can challenge.

Hold that standard against a live task. You are on the corporate development team at Meridian Components, a mid-market industrial parts manufacturer weighing a strategic investment. The board meets in a week and wants a valuation: what the business is worth on a discounted-cash-flow basis, what moves that number, and where the base case is exposed. The deliverable is a written pre-read a director will skim in about five minutes.

Board-ready is a standard, not a spreadsheet, and it has four properties. A board-ready valuation is traceable, so each figure ties back to a stated assumption; it recomputes, so an independent recalculation of the discounted cash flows lands on the same enterprise value; it is bounded, so a sensitivity view shows how the value flexes rather than presenting one point as certain; and it is framed, reading as a pre-read that leads with the base case and the swing factors rather than a raw model dump. Two of those properties depend on an assumptions ledger: the single record of each input, growth, margin, tax, capital intensity, discount rate, and terminal growth, that every figure in the model traces back to. Pinning that ledger down is the analyst's job, not the tool's.

The work splits in two. The middle of a DCF, projecting free cash flow, capitalizing a terminal value, discounting at the WACC, and summing to enterprise value, is a long, deterministic calculation with one right answer for a given ledger. The judgment sits at the two ends, in the assumptions that feed the model and in the narrative that frames the result. That clean split, a mechanical model in the center and judgment at the edges, is what makes this task a good candidate for AI assistance, and it is the design the next part builds on.

The most useful things to hand a model for this task are the destination shape and the ground it stands on. A prior valuation, or a house DCF template, shows the structure the board expects: how the free-cash-flow build is laid out, how the terminal value is derived, and how the pre-read reads. The three years of historical financials anchor the assumptions in something real, so a projected margin or growth rate can be sanity-checked against what the business has actually done. With both in the folder, the request shifts from "value this company" to "build this known structure from these stated inputs."

The journey is clear at both ends. You start with an assumptions ledger and three years of history; you are going to a validated enterprise value with a sensitivity narrative and a board pre-read. Anchoring the far end with a known model shape keeps the tool from inventing a structure of its own.

Projecting free cash flow and discounting it is deterministic work: for a given ledger, there is one correct revenue path, one correct free-cash-flow line, one terminal value, and one enterprise value. Asking a language model to carry that arithmetic in prose invites small, compounding slips, and in a DCF a slip early in the chain moves the answer at the end. So the workflow puts the math where it belongs. The model builds the DCF skeleton, the five-year free-cash-flow build and the Gordon-growth terminal value, as an explicit calculation from the ledger, ideally expressed as code that recomputes the same way each time it is run. Then, once you have validated that skeleton, the model performs the non-deterministic work: writing the sensitivity narrative and the board framing, where there are many acceptable phrasings and the value is in judgment, not arithmetic.

This is the largest-scale version of the deterministic split in the course. In a flux commentary the deterministic part is a handful of variances; here it is a whole model. The principle is identical. Build the calculation once, validate it once, and reserve the model's language for the parts where judgment is the work. When the tool narrates sensitivity on top of a base it recomputed in prose, two versions of the number can coexist, and neither earns trust.

The lab folder is deliberately small: the assumptions ledger, three years of historical financials, and the brief. Nothing else. A pile of market comps and broker notes would bury the inputs that actually drive the model and raise the chance the tool anchors on a number no one can cite. Minimum context is the fuel here: give the model the ledger it computes from and the history that grounds it, and keep out anything the valuation does not need.

The failure a bloated folder invites is subtle. Handed a stack of comps and broker targets, a fluent tool will sometimes reach for a market multiple in place of the ledger's own discount rate or growth path, and the resulting value looks market-grounded while it traces to an input the analyst did not choose. Fewer, load-bearing files keep each figure pointing back at an assumption someone can defend, which is the property the whole validation step later depends on.

The workflow is worth making repeatable. Capture the instructions as a reusable valuation skill, or a saved prompt: read the ledger and historicals, build the five-year free-cash-flow projection and the Gordon-growth terminal value as an explicit calculation, discount at the stated WACC, and hold the narrative until the enterprise value is validated. Written once, it runs the same way on the next target, and it encodes the split so the tool does not drift back into narrating math.

Pair it with a short pre-flight checklist the reviewer runs before anything is drafted: is every input in the ledger and sourced; does terminal growth sit below the WACC and stay bounded by long-run economic growth; does an independent recompute of the discounted cash flows tie to the model's enterprise value; does the sensitivity table bracket the base case; and does no figure in the narrative fail to trace to the ledger. The checklist is where the best practices from Part One become an operational gate rather than a hope.

Download the folder and run the build-then-narrate pattern in whatever AI you use. The folder holds Meridian's assumptions ledger (growth, margin, tax, capital intensity, discount rate, and terminal growth), three years of historical financials for context, and the brief. Let the model build the five-year free-cash-flow model and the terminal value from the ledger, validate that the enterprise value recomputes and each figure traces to an assumption, and only then have the model write the sensitivity narrative and board framing. The company and its numbers are fictional.

The heart of validation is the recompute. Take the enterprise value and rebuild it independently: discount each of the five free cash flows at the WACC, discount the terminal value, and sum. If your recomputed figure matches the model to the thousand, the base is tied; if it does not, the number goes back before a word of narrative is written. Confirm, too, that each free-cash-flow component traces to a ledger assumption, that the terminal growth rate sits below the discount rate, and that the sensitivity table moves the value in the expected directions. Work the checklist below against your model before you would ever call it board-ready.

Most of the value in a growing business sits beyond the explicit horizon, so the terminal value carries real weight. Apply the Gordon growth method to the final-year free cash flow of about $20,724 thousand. First grow it one period at 2.5%: 20,724 times 1.025 is next-year cash flow of about $21,242 thousand. Then capitalize that at the 7.5% spread between the 10% WACC and the 2.5% terminal growth: 21,242 divided by 0.075 gives a terminal value of $283,224 thousand as of the end of year five. That figure is then discounted back to the present along with the five explicit cash flows.

Discounting the five free cash flows at the 10% WACC sums to about $70,787 thousand of present value, and discounting the $283,224 thousand terminal value back five years adds about $175,860 thousand. Together they give an enterprise value of $246,647 thousand. A quick validity check sits underneath the whole model: terminal growth of 2.5% is well below the 10% WACC, so the Gordon growth denominator is a healthy positive spread rather than a near-zero one. Nudge terminal growth alone to 8% and the check earns its place. The spread collapses from 7.5% to 2%, and the terminal value balloons from $283,224 to roughly $1,119,000 thousand, close to four times the base, on a single input a reviewer could wave through. Holding terminal growth near long-run economic growth is what keeps the denominator healthy and the terminal value defensible.

This is the model's non-deterministic turn, written only after the base tied out. A sensitivity analysis flexes the two assumptions the value is most exposed to, the WACC and the terminal growth rate, and reports how enterprise value moves. Because roughly seven-tenths of the value sits in the terminal value, small changes in either input move the answer more than a change in a single year's revenue growth would, which is the swing the board needs to see.

The board pre-read that comes out of it leads with the base case and the swing factors, not the arithmetic. Something like: "On the base assumptions, the business is worth about $247 million on a discounted-cash-flow basis. The value is most sensitive to the discount rate and the terminal growth rate; a half-point change in either moves the value materially, while the year-to-year revenue growth path matters less. The base case assumes growth fading from 6.0% to 3.5% and a long-run growth rate of 2.5%, both of which sit within the range of what the business has done and what the wider economy supports." That is a paragraph a director can absorb in about a minute, and each figure in it ties back to the model.

The lab is clean by design: the ledger is complete, the historicals are consistent, and each assumption is handed to you. Real valuations are messier, and a few failure modes recur.

• The WACC is an estimate resting on other estimates. A tool will often supply a plausible-sounding discount rate assembled from a beta, an equity risk premium, and a target capital structure that no one actually sourced, and on a business this size the gap between a 9% and an 11% WACC can move enterprise value by tens of millions.
• The terminal value can carry too much of the answer. When roughly seven-tenths of value sits beyond the explicit horizon, a terminal growth rate that drifts even half a point, or an exit multiple pulled from thin air, dominates the result, so the number a board leans on rests largely on the assumption least anchored in the historicals.
• The enterprise-to-equity bridge is easy to get wrong. Directors usually want equity value, and the step from enterprise value to equity value, subtracting net debt and then adjusting for items like minority interests and options, is where a fluent tool tends to drop or double-count a line, so a model whose enterprise value ties out can still report an equity value that does not.

This does not remove the analyst's judgment about the assumptions; it clears the mechanical model-building so the judgment, and the recompute and terminal-growth validity check that guard it, have more room. A tool will sometimes produce a fluent pre-read on a model whose terminal value quietly rests on a growth rate too close to the discount rate, which is why those checks are not optional extras.

Rate your own valuation against the rubric below. Scored honestly, the rubric shows where the valuation build is dependable and where it still needs work. Your scores roll up to the workflow maturity dashboard on the course hub.