Implementation Business Case Generator

This generator writes the money argument for a path you have already chosen, after outcome ranges exist. It is not a second forecast, and it is not permission to mint a net present value.

If you do not yet have ranges, stop. A multi-scenario outcome modeler is the prior job: named drivers, named scenarios, a range per measure. This pack takes the preferred path from that work, plus the client's own financial parameters, and produces NPV, payback, and sensitivity. Collapse the range into one heroic NPV and you threw away the quality of the analysis.

A strategic option comparison synthesizer still owns qualitative trade-offs on client-agreed criteria. An initiative prioritization matrix generator still owns ranking and sequence. Neither artifact is a business case. The case is the cash argument a CFO can approve or reject without reverse-engineering your workbook.

Quality here means finance can see the drivers, the rate, and what flips the answer. A finished-looking number that hides those things is a different product.

Feed the engine two things only.

The recommendation inputs. Cash-flow shape for the preferred path: capex timing, run-rate cost, benefit timing, residual value if the committee uses one. Those amounts must already be ranges from the modeler, not a midpoint the partner prefers.

Client-supplied financial parameters. The discount rate they actually use (WACC or the hurdle treasury will defend), tax treatment they will own, the planning horizon they use for similar spends, and the payback convention the investment committee applies. If a parameter is missing, it stays missing.

The LLM drafts the schedule language. The rules engine applies the arithmetic. Neither may invent the rate.

Invented WACC is the first failure. A textbook industry average, a peer set the team assembled, or a model default is not the client's cost of capital. The committee will ask whose rate it is. If the answer is yours, the NPV is yours.

If treasury has not given a rate, write "rate not supplied" and do not publish NPV. You may still show undiscounted cash ranges, and payback under a horizon they did give you. You may not discount with a fill-in.

Match the horizon to the one they use. Stretching the tail so payback looks inevitable is the same failure as inventing a rate. If they plan three years for distribution-center programs, the finding is three years. Label any longer tail as a sensitivity they own, not as the case.

Get the rate, horizon, and payback convention from treasury or the controller in writing, with a date, before you run the engine. Load the preferred-path cash ranges without taking a midpoint first. Then run NPV and payback across that range, and sensitivity on the drivers the modeler already flagged plus swings in rate and horizon.

For the preferred path, the money page shows three things together:

• NPV as a range, produced by running the supplied rate across the cash-flow range you already have
• Payback as a range on that same cash-flow range
• Sensitivity: which supplied drivers, and which rate or horizon swings, move NPV across a sign change or a hurdle, or push payback past the line the committee will not accept

Do not lead with a single NPV. Name the preferred path and the decision on the title slide. Put the range and the flips on the money page.

If the NPV range includes both a case that clears the hurdle and a case that does not, that overlap is the finding. The honest sentence is "resolve this driver" or "choose knowing the money argument is not settled." A quiet headline number is a choice you made for the room.

The engine should not pick a base case. If the partner wants a base, it is a client-agreed slice of the range, labeled as such, sitting next to the full range. Payback without a discount rate is still a range. Do not report one year as if the cash were certain.

Finance reviews assumptions, then arithmetic. A pack that hides drivers in a locked file or an appendix nobody opens will fail the first challenge, or pass because nobody could find the challenge. Both are quality failures.

Keep the sheet on the money page, or on the page immediately after:

• Discount rate, source (treasury memo, named controller), date given
• Horizon and payback convention
• Each cash-flow driver that feeds the preferred path: range, source, who will defend it
• What was not supplied
• What a swing in each driver does to NPV sign or to payback relative to the committee's line

Figures from board packs and contracts can arrive through an unstructured document extraction pipeline. Someone still has to mark each figure as history or as a client-owned forecast. Extraction does not turn a page number into a financial parameter.

The model usually lives where the client already plans. Excel, Anaplan, and Pigment are that class of place: workbooks and connected plans that can hold a schedule and a rate. Treat them as the system of record for the case, not as ranked NPV products. Do not assume any of them ships a particular DCF engine. There is no magic NPV product. The work on top is the same: client rate, client horizon, ranges in, ranges out, sheet visible.

When the CFO changes the rate in the room, the range should move. A PDF of one NPV trains the room to argue with the slide.

Illustrative only. No results, no case study.

A partner at Kestrel is taking a preferred path to Meridian Grocery's investment committee: automate the existing regional DC rather than add 3PL overflow. Analyze already produced overlapping cash ranges. Labor, mix, and capex timing still flip the rank against overflow. Those ranges exist. They have not been packaged.

The first-pass generator, run before treasury has spoken, fills a WACC from a grocery peer set, collapses the cash range to a midpoint, and prints one NPV and one payback on the title slide. The assumption sheet is a hidden tab. The pack looks finished.

Refuse that pack.

What belongs in the room is the preferred path named, NPV and payback as ranges under Meridian's own rate and the horizon treasury uses for DC spends, sensitivity that still shows labor and capex timing can push payback past the committee's comfort, and a sheet that either cites treasury or says "rate not supplied." If treasury has not spoken, there is no NPV on the title slide.

A recommendation adversarial stress-tester then rehearses the first finance objection: whose rate, which cash year is a hope, what happens if ramp slips. The stress-tester does not rewrite the recommendation. It checks whether the pack can survive that question with the sheet still open.

• Invented WACC. Peer-set rates, textbook defaults, model fill-ins. No client rate, no NPV. Undiscounted cash and a labeled gap are honest. A discounted headline is not.
• One heroic NPV. The midpoint of a scenario range, or the upside, presented as the case. If the range can miss the hurdle, the title number is a choice, not a result.
• Hidden drivers. Appendix-only sheets, locked files, sensitivity that never makes the circulating pack. If finance cannot see what moves the answer, they will reject the case or accept a number they cannot defend later.

The generator is doing its job when the preferred path is packaged after ranges exist, the rate and horizon are the client's, NPV and payback are ranges with sensitivity, and the assumption sheet is in the pack. It is failing when the slide has one number, the rate is yours, or the workbook hides the drivers.