Imported from NirvanaGuha/GrowthSkills (skills/roi-business-case-calculator/SKILL.md). Install upstream with npx skills add NirvanaGuha/GrowthSkills --skill roi-business-case-calculator. Copyright stays with the author.

ROI & Business Case Calculator

Give it a prospect's metrics and a set of impact levers, get a defensible ROI model and a champion-ready executive note. The model is structured around the three classic value-driver buckets — Revenue Growth, Cost Reduction, Capital Efficiency — so the math survives a CFO's first question. Every input is labeled (confirmed vs. benchmarked vs. assumed), every benchmark is sourced or flagged [verify], and the sensitivity table shows the champion exactly what to argue about.

This skill quantifies and structures. It does not invent proof, inflate benchmarks, or paper over a thin value proposition. If the inputs don't support a credible number, it says so and tells the user what data to collect.

Skills this calls

• brand-brain (required first) — loads the active brand's voice, proof points, and real customer outcomes used in the model's evidence column. Do not write any customer-facing copy before brand-brain returns. Fallback if brand-brain is absent or returns no brand: read ~/.brandbrain/brands/.active + that brand's brand.md directly; if none exists, ask the user for brand name, 2–3 confirmed customer proof points with metrics, and voice adjectives before proceeding.
• data-qa-measurement-gotcha-checker (recommended) — run on any pasted CSV, spreadsheet export, or CRM pull before arithmetic. Catches unit mismatches, sampling gaps, and attribution errors that would corrupt the model.
• proof-vault (optional) — pulls confirmed customer proof points and case-study metrics for the evidence column. Synthesize from brand.md if absent.
• board-exec-summary-writer (optional) — converts the completed model into a board-ready one-pager. Call after the model is reviewed and approved; do not call in the same pass that builds the model.
• analytics-report-reviewer (optional) — reviews the finished model for unsupported claims, weak assumptions, and missing context before it goes external.

How a run works

Step 0  Load the brand          ──► call brand-brain; get voice + confirmed proof
Step 1  Gate the inputs         ──► data-qa check on any pasted data; flag dirty inputs
Step 2  Scope the mode          ──► Quick (single lever, fast answer) | Full Model
Step 3  Build the model         ──► three-bucket value-driver structure
Step 4  Sensitivity table       ──► conservative / base / optimistic scenarios
Step 5  Write the executive note ──► champion-ready, brand-voice, proof-cited
Step 6  Save artifacts          ──► ./roi/[slug]-roi-model.md (Full Model)

Step 0 — Load the brand (always first)

Invoke the brand-brain skill (Skill tool, skill: brand-brain). It returns the active brand's digest — voice adjectives, banned words, confirmed proof points with metrics, positioning, ICP. Use the returned proof in the model's evidence column; mark anything else [verify]. Obey voice and banned words in the executive note.

Fallback if brand-brain is absent or returns no brand: read ~/.brandbrain/brands/.active + that brand's brand.md directly; if none exists, ask the user for brand name, 2–3 confirmed customer proof points with metrics, and voice adjectives before proceeding.

Step 1 — Gate the inputs

Before any arithmetic:

1. Ask for the prospect's confirmed baseline metrics (see Required inputs). If the user pastes data, invoke data-qa-measurement-gotcha-checker on it first.
2. Identify what is confirmed (prospect-provided), benchmarked (industry study or your own customer data), or assumed (modeled). Label every number in the model.
3. If key baseline metrics are missing (e.g., current revenue, headcount cost, churn rate), ask — don't guess. Name exactly what's needed and why.

Mode selection

• Quick mode (default) — one lever, one clear dollar figure. Champion-ready in under 5 minutes. Use when: single use case, early-stage deal, prospect needs a fast "is it worth exploring?"
• Full Model — multi-lever spreadsheet-ready model with payback period, 3-year NPV, and sensitivity table. Use when: late-stage deal, procurement or CFO review, multi-department impact, or the user explicitly asks.

If unclear, default to Quick and offer Full Model at the end.

The three value-driver buckets (shared by both modes)

Every lever maps to exactly one bucket. This is not decoration — it determines which executive owns the number and how hard it is to challenge.

Map each lever the user provides to its bucket before calculating. If a lever spans buckets, split it — do not double-count.

Quick mode

Input: one baseline metric + one impact lever + desired output type (annual savings / revenue uplift / time saved).

Output:

## ROI quick estimate — [what / prospect]
Lever: [bucket] — [description]
Baseline: [metric] (source: [confirmed/benchmarked/assumed])
Impact assumption: [%] (source: [proof point or [verify]])
Annual impact: $[X] or [X] hrs/yr
Confidence: [High/Medium/Low] — [one-line rationale]
⚠ Data quality: [pass | flag from data-qa]
Next: ask the champion for [the missing baseline that would make this High confidence]

Keep Quick-mode output to one screen. No table unless the user asks.

Full Model

1. Input collection

Collect (or confirm already provided):

2. Model structure

Build one table per value-driver bucket containing levers that apply. Within each lever:

Lever name | Baseline | Impact % | Annual $ impact | Source / confidence | Bucket

• Confirm, then calculate. Never back-solve from a target ROI. If the prospect says "we need 3x ROI," that is a constraint to check, not an input to reverse-engineer.
• Attribution discipline. Only claim the incremental impact attributable to the solution. If churn drops 20% and the solution drives 5 pp of that, model 5 pp. Say so.
• Conservative anchor. Default to conservative assumptions; show upside in sensitivity. Champions can defend a conservative number; they can't defend one that got challenged at the table.

3. Summary roll-up

## ROI model summary — [prospect / brand / date]
Brand: [slug, via brand-brain]
─────────────────────────────────────────────────
Revenue Growth levers:   $[X] / yr
Cost Reduction levers:   $[X] / yr
Capital Efficiency levers: $[X] / yr (or qualitative if not quantifiable)
─────────────────────────────────────────────────
Total annual value:       $[X]
Solution cost (Yr 1):     $[X]
Net annual benefit:       $[X]
Payback period:           [X] months
3-yr NPV (@ [rate]%):     $[X]
ROI (Yr 1):               [X]x
─────────────────────────────────────────────────
Confidence: [High/Medium/Low]
Key assumption to validate: [the one number that moves the model most]

4. Sensitivity table

Three scenarios across the lever(s) with the highest variance. Conservative / Base / Optimistic. Label what changes in each. Show the ROI and payback period for each.

| Scenario      | Key assumption    | Annual value | ROI   | Payback |
|---------------|-------------------|--------------|-------|---------|
| Conservative  | [what's lower]    | $[X]         | [X]x  | [X] mo  |
| Base          | [central estimate]| $[X]         | [X]x  | [X] mo  |
| Optimistic    | [what's higher]   | $[X]         | [X]x  | [X] mo  |

Tell the champion: "The conservative scenario is your floor for the CFO conversation. If you can confirm [key input], you can defend the base."

5. Champion executive note

One page, brand-voice. Structure:

1. The problem (one sentence — the prospect's current cost/risk in dollar or time terms)
2. The lever (what changes and why, tied to a confirmed proof point from brand-brain/proof-vault)
3. The number (base-case ROI and payback period, labeled conservative)
4. What we'd need to confirm (the one input that would move the model to high confidence)
5. Recommended next step (a specific ask, not a generic "let's talk")

Voice and banned words from brand-brain apply. Every proof point must be in proof-vault or brand.md, or flagged [verify]. No invented social proof.

6. Save artifacts

Full Model output saves to ./roi/[brand-slug]-[prospect-slug]-roi-model.md. Quick-mode output is inline only. Never overwrite an existing model without prompting.

Principles (Non-Negotiable)

• Brand-brain first. No customer-facing copy before brand-brain returns. Its voice and proof override everything here.
• Data gate before arithmetic. Run data-qa on pasted inputs; dirty data produces a defensible-looking but wrong number.
• Confirm, benchmark, or flag. Every input is labeled. Never pass an assumed number as confirmed.
• Conservative anchor. Default to conservative; show upside in sensitivity. A challenged number kills the deal.
• No double-counting. If a lever touches two buckets, split it explicitly.
• Attribution discipline. Only claim impact attributable to the solution. Don't credit market tailwinds.
• One model per prospect. Save with both brand slug and prospect slug so nothing gets overwritten.

What Not to Do

• Don't back-solve from a target ROI — model forward from confirmed inputs.
• Don't invent customer proof points — use only what brand-brain or proof-vault returns, or [verify].
• Don't call board-exec-summary-writer in the same pass that builds the model — review first.
• Don't skip the sensitivity table to save space — it's the table that survives CFO scrutiny.
• Don't use growth-rate assumptions without labeling their source (analyst report, your own cohort data, assumed).
• Don't combine multiple prospects' metrics in one model — build one model per deal.

Quality Checklist (self-review before presenting)

• brand-brain called and returned before any customer-facing copy was written?
• data-qa run on any pasted input data; flags addressed or disclosed?
• Every input labeled: confirmed / benchmarked / assumed?
• Every lever mapped to exactly one value-driver bucket; no double-counting?
• Conservative anchor used as the base; optimistic reserved for sensitivity table?
• Payback period and 3-yr NPV calculated (Full Model)?
• Sensitivity table covers the highest-variance lever with three scenarios?
• Champion executive note names one specific next step?
• All proof points in brand.md / proof-vault, or [verify]?
• Full Model artifact saved to ./roi/[brand-slug]-[prospect-slug]-roi-model.md?