fm-1-scope

Phase 1: Scope 🔍

"The biggest modelling errors happen before a single formula is written."

Mission

Conduct a structured, multi-round conversation to deeply understand the user's business context, model purpose, and decision needs. Not a checklist — a collaborative discovery process.

Important References

Before starting the conversation, read:

1. references/industry_knowledge.md — Subject matter guidance for 9 industries (mining, energy, infra, real estate, tech, retail, government, project finance, general corporate) with revenue drivers, cost drivers, CapEx considerations, key metrics, and common pitfalls
2. references/estimation_benchmarks.md — Hour estimates per section, complexity multipliers, typical model sizes, and fee range calculation (for the Effort Estimation sub-skill)

Core Principles

• Ask, don't assume. Even if you know the "standard" answer, confirm it.
• Suggest, don't dictate. Offer industry knowledge the user may not have considered.
• Iterate, don't rush. Better to ask 3 rounds of questions than miss a key driver.
• Document exclusions. What's OUT of scope is as important as what's IN.
• Pace the conversation. Ask 3–5 questions per round, not 30 at once.

---

Process

Round 1: UNDERSTAND THE BUSINESS (Open-ended)

Start broad. Let the user talk. Listen for clues about complexity.

Ask these together — user answers what they can:

1. What is this model for? What decision does it support? (investment appraisal, budget, project finance, valuation, feasibility, board paper, lender model?)
2. Who is the audience? (yourself, your manager, a board, investors, a lender, a regulator?)
3. What industry/sector? (mining, energy, infrastructure, real estate, tech, retail, government?)
4. Is there an existing model we're replacing, extending, or starting fresh?
5. When do you need it?
6. Do you have reference documents? (business case, term sheets, data packs, prior models?)

THEN — provide subject matter guidance based on what you learned. Read the matched industry's section in references/industry_knowledge.md (or the deeper references/industries/<industry>.md file) and proactively suggest the standard considerations for that sector.

Frame it as: "For [industry] models, teams typically also consider [X, Y, Z]. Would any of these be relevant here?"

If the user seems unsure about structure, offer 2–3 example model archetypes:

• "A simple version would cover [A, B, C]. A more detailed version would also include [D, E, F]. Which feels closer to what you need?"

Round 2: ROUTE TO THE RIGHT QUESTIONS

Most SumProduct engagements are NOT 3-statement integrated models. The question library covers 13 model types — budget, commission, cost allocation, cashflow forecast, dashboard/reporting, M&A/LBO, model optimisation, operational, valuation/DCF, project finance, and standard 3-statement.

references/question_bank.md is the router. Open it and:

1. Look at the "Model Type" table (Step 1 in the question bank). Match the user's Round 1 answers to one or more model types.
2. Always ask the universal questions (timeline, outputs/deliverables) from the question bank's Step 2 — these apply to every model.
3. Load ONLY the question file(s) matched in step 1. Many engagements combine types (e.g., project finance + operational) — load multiple as needed.
4. Walk the user through one group at a time. Do not force 3-statement questions on a commission, dashboard, or optimisation engagement.

The 3-statement question set lives in references/questions/financial_statements.md — it covers revenue/cost/CapEx/funding/working capital/tax. Use it ONLY when the matched model type is 3-statement, project finance, M&A/LBO, feasibility/investment, or strategic/scenario. For other types, the right question file is much shorter and asks model-specific things (e.g., commission asks about tiers and clawback; dashboard asks about KPIs and refresh frequency).

Round 3: CLARIFY & CHALLENGE

Review what you've gathered and push back constructively:

• Challenge vague inputs: "You mentioned 'some debt'. Even a placeholder (e.g., $10M at 6% over 10 years) helps me structure the model correctly. Can you estimate?"
• Identify gaps: "I notice we haven't discussed working capital. For [industry], this typically matters because [reason]. Should we include it?"
• Flag complexity vs value: "Monthly seasonality adds significant complexity. Is it critical for your decision, or would annual averages work?"
• Validate against audience: "You said this goes to lenders. They'll typically expect DSCR, LLCR, and a cash waterfall. Should we build those in?"
• Confirm exclusions: "Just to confirm — you don't need [X]? I want to be sure we don't have to restructure later."
• Resolve open questions: If the user can't answer something now, log it as an open item and flag whether it blocks progress or can be resolved later.

Round 4: DOCUMENT

Produce scope.md from references/scope_template.md — it holds the universal core (identity, timeline, outputs, exclusions, open questions) plus the complete per-type section blocks for all 13 model types. Use the block matching the type identified in Round 2; combined engagements take the union. Type definitions and complexity scaling: ../_fm-shared/references/model_types.md.

Do not force 3-statement sections onto non-3-statement models — that produces a scope document the user does not recognise as their project.

Round 5: CONFIRM & HANDOFF

Present scope to user:

• "Here's the scope I've documented. Please review — anything missing or incorrect?"
• Resolve any remaining open questions.
• "Once confirmed, I'll move to Phase 2 (Plan) to design the model structure."

Do not proceed to Phase 2 until the user confirms the scope.

---

Defaults (when user doesn't specify)

Read references/defaults.md — periodicity, periods, currency, scale apply to all models; the Statements/Opening BS/Working capital/Tax defaults apply to 3-statement / project finance only. Always tell the user which defaults you applied and mark them [Assumed — confirm with client].

---

Sub-Skill: Effort Estimation & Quoting 💰

When to use: User asks "how long will this take?", "can you quote this?", "estimate hours", "what would this cost?". Not always needed — skip if the user just wants the model built.

Read references/effort_estimation.md for the full process (map scope to sections → apply complexity multipliers → produce the estimate document) and references/estimation_benchmarks.md for the hour ranges and multiplier data tables.