Core principle: Decompose the decision into revenue drivers, cost drivers, and timing — then stress-test the assumptions that swing the outcome. The model isn't the answer; the sensitivity analysis is.
When to Use This Skill
- Evaluating whether to invest in a project, product, or initiative
- A pricing decision needs analytical backing
- "Is this profitable?" or "what's the ROI?" without a structured model
- A decision-synthesis analysis needs financial criteria and scores
- A fermi-estimation output needs refining into a rigorous model
- Comparing options with different cost structures or revenue timelines
Core Methodology
Step 1: Frame the Financial Question
- What's the decision? (invest/don't, A vs. B, price X vs. Y)
- Time horizon? (match to decision's natural lifecycle)
- Who bears costs, who captures value?
- Discount rate? (cost of capital, opportunity cost, or stated hurdle rate)
If inputs are rough, accept them from fermi-estimation via Contract M and flag what needs validation.
Step 2: Map Unit Economics
Decompose to the smallest repeatable economic unit — per customer, transaction, unit sold, API call.
- Revenue per unit: price × volume, or recurring revenue × retention
- Cost per unit: direct (COGS, marginal) + allocated indirect
- Contribution margin: revenue − cost per unit
- Break-even volume: fixed costs / contribution margin
If clean unit economics don't exist (e.g., platform with network effects), identify the value driver (users, transactions, data volume) and model around it.
Step 3: Build the Cost-Benefit Table
Enumerate explicitly. Categorize each:
- Direct costs: engineering time, infrastructure, licensing, materials
- Indirect costs: opportunity cost, coordination overhead, technical debt
- Direct benefits: revenue, cost savings, efficiency gains
- Indirect benefits: option value, learning, strategic positioning
Discount future to present using Step 1's rate. Compute:
- NPV: sum of discounted net cash flows
- IRR: discount rate at which NPV = 0
- Payback period: when cumulative net cash flow turns positive
Step 4: Model Scenarios
Three cases by varying key assumptions:
- Base: most likely
- Upside: optimistic but plausible (not best-case fantasy)
- Downside: pessimistic but plausible (not worst-case catastrophe)
Recompute NPV, payback, contribution margin per scenario. Wide upside-downside spread = high uncertainty, lower confidence.
Step 5: Run Sensitivity Analysis
Identify which assumption, if wrong, changes the decision.
- Vary each key assumption ±25% and ±50%
- Record the largest NPV swing
- That's the key financial risk — deserves the most validation effort
If two assumptions interact (e.g., price and volume), model the interaction explicitly.
Step 6: Produce Financial Criteria for Decision-Synthesis
Feeding decision-synthesis via Contract G, translate findings into criteria:
- Convert thresholds to must-have or want-to-have ("NPV > $0" must-have; "payback < 18 months" want-to-have)
- Score each option 1-5 on financial performance
- Flag key financial risk and scenario spread as confidence modifiers
- Suggest weights based on how financially dominant the decision is
Output Format
🎯 Decision Under Evaluation
- Decision: [what's being decided]
- Time horizon: [N years/months]
- Discount rate: [X% — with justification]
📋 Key Financial Assumptions
| Assumption | Value | Source | Confidence |
|---|---|---|---|
| [Monthly active users Y1] | [value] | [fermi/market/internal] | H/M/L |
| [Price per unit] | [value] | [source] | H/M/L |
| [Infra cost per user] | [value] | [source] | H/M/L |
📊 Unit Economics
- Revenue per unit: [value]
- Cost per unit: [value]
- Contribution margin: [value] ([%])
- Break-even volume: [value]
⚖️ Cost-Benefit Summary
| Category | Item | Y1 | Y2 | Y3 | NPV |
|---|---|---|---|---|---|
| Cost | [item] | [value] | [value] | [value] | [value] |
| Benefit | [item] | [value] | [value] | [value] | [value] |
| Net | [value] | [value] | [value] | [NPV] |
- IRR: [value]%
- Payback period: [N months/years]
📊 Scenario Analysis
| Scenario | NPV | Payback | Key Assumption Changed |
|---|---|---|---|
| Downside | [value] | [value] | [assumption at pessimistic] |
| Base | [value] | [value] | Most likely values |
| Upside | [value] | [value] | [assumption at optimistic] |
⚠️ Sensitivity Analysis
| Assumption | -50% | -25% | Base | +25% | +50% | NPV Swing |
|---|---|---|---|---|---|---|
| [key 1] | [NPV] | [NPV] | [NPV] | [NPV] | [NPV] | [max-min] |
| [key 2] | [NPV] | [NPV] | [NPV] | [NPV] | [NPV] | [max-min] |
- Key financial risk: [the assumption that, if wrong, changes the answer]
🏆 Recommendation
- Financial verdict: [invest / don't / conditional on validating X]
- Confidence: H/M/L — informed by scenario spread and sensitivity
- Decision-synthesis criteria (Contract G):
- ["NPV > $X over 3 years"] — type: must/want, weight: [N]
- ["Payback < 18 months"] — type: must/want, weight: [N]
- Financial scores per option (if comparing):
- Option A: NPV = $X, payback = Y months, score = [1-5]
- Option B: NPV = $X, payback = Y months, score = [1-5]
Common Traps
False precision: Reporting NPV to the dollar when inputs are order-of-magnitude. Match output precision to input precision.
Ignoring opportunity cost: Comparing investment to zero instead of next-best use of the same resources. The baseline is "do the next-best thing," not "do nothing."
Hockey stick projections: Revenue grows exponentially, costs stay flat. Challenge any model where margins improve dramatically without a structural reason.
Sunk cost inclusion: Already-spent costs in forward-looking analysis. Only future cash flows matter.
Single-scenario thinking: Only the base case. Without upside/downside, stakeholders can't assess risk.
Trusting LLM arithmetic: An LLM applying this skill structures the model correctly but may compute NPV/IRR/sensitivity inaccurately. Verify in a spreadsheet. Use this skill for structure, assumptions, and logic — not as a real financial model replacement.
Thinking Triggers
- "What are the unit economics at the atomic level?"
- "Which single assumption, if wrong by 2x, flips this decision?"
- "What's the opportunity cost — what else could these resources do?"
- "Does the upside-downside spread change my confidence?"
- "Am I modeling what I hope, or what's most likely?"