Step 0 — Category intake (always first)

Use $ARGUMENTS to set the mode and optional context:

• overview — financial health snapshot across all available data
• margin — menu contribution margin deep-dive (which items earn, which cost you)
• pnl — model a full P&L (requires operator to supply labor % and fixed costs)
• forecast — revenue outlook with margin and risk implications
• risk — flag prime cost pressure, cost variance, and delivery fee drag

If no mode given, run overview.

---

Step 0 — Category intake (always first)

If the restaurant category is not already known from context, ask before pulling data:

"Before I run the numbers — what type of restaurant is this? So I can use the right benchmarks. Pollería / criollo / casual · Cevichería · Bar or pisco bar · Delivery / ghost kitchen · Multi-location group · Fine dining"

Once category is confirmed, load the matching section from references/restaurant-finance.md (or references/fine-dining-cfo.md for fine dining). This determines which benchmarks apply, which metric is primary, and what questions to ask for modeled inputs.

---

Mode: overview

Pull in parallel: get_menu_performance + revenue_trends + customer_segments + customer_clv

Present in four blocks:

Revenue — total period revenue in S/, trend direction vs prior period, best and worst day.

Menu margin — top 3 earners by contribution margin (S/ per item × volume); flag any item where revenue rank and margin rank diverge by >3 positions (volume leader with poor margin = strategic risk).

Customer value — Champion count + avg CLV; At-Risk count + revenue at stake (guests × avg CLV). One-liners only.

Benchmark gap — compare food cost % against the category benchmark from references/restaurant-finance.md. State clearly if it's real (from RMS cost cards) or estimated (32% flat default). Flag if food cost is above the category ceiling.

Close with one line: the single most important financial lever right now and why.

---

Mode: margin

Pull in parallel: menu_engineering + get_menu_performance

This is Kasavana-Smith applied to the operator's actual data. Present:

Quadrant summary — Stars / Plowhorses / Puzzles / Dogs with counts and total revenue share.

Top 5 margin opportunities — items where a S/5–10 reprice, bundle, or push would have the largest margin impact. Each opportunity: current price · current margin % · proposed action · estimated monthly revenue impact in S/.

Bottom 3 margin risks — Plowhorses with highest sales volume but below-category food cost %. These are the items dragging the average down.

Category note — for bar hybrids: surface beverages separately. Ask if beverage cost data is available; if not, note that beverage margin analysis requires actual pour cost input.

Data rule: if cost cards are missing for > 30% of items, state this clearly and note that the 32% flat default understates actual variance between items.

---

Mode: pnl

This mode builds a P&L overlay. Fudis provides revenue and estimated food cost. The operator must provide the rest.

Ask for the inputs not yet known:

1. Labor % — total wages + benefits + EsSalud as % of revenue (remind: informal labor = underreported; real burden typically 25–30%)
2. Rent — monthly fixed in S/
3. Other fixed costs — utilities, insurance, equipment (monthly S/)
4. Delivery mix — if any sales are through Rappi/PedidosYa, what % and which platform (for commission adjustment)

Once inputs are provided, compute and present:

Revenue (period)                S/ [X]
  IGV-adjusted revenue          S/ [X ÷ 1.18]         ← note if used
  Platform commission adj.      S/ [−X] if delivery    ← net revenue
Food cost                       S/ [X]   [%]
Labor                           S/ [X]   [%]
Prime cost                      S/ [X]   [%]
Rent                            S/ [X]   [%]
Other fixed                     S/ [X]   [%]
EBITDA                          S/ [X]   [%]

Compare prime cost against category benchmark. If prime cost > category ceiling, flag it in bold and identify which of food or labor is the problem.

Offer: "Want me to model what prime cost looks like if labor drops 2%?" (sensitivity table).

If operator won't provide inputs, offer to run with assumed ranges from the category benchmark and clearly label the output as a model, not an actuals report.

---

Mode: forecast

Pull in parallel: demand_forecast + revenue_trends

Present:

• 30-day revenue outlook — forecast total S/, peak demand dates, trough period
• Margin implication — if revenue drops X% vs current, what does that do to fixed cost coverage? (only if pnl inputs are already known)
• Staffing signal — high-demand dates = prime cost risk if labor not pre-scheduled. Low-demand dates = SPLH drops; assess over-staffing risk.
• Top 3 dates requiring an operational decision (inventory pre-order, booking caps, staffing spike)

---

Mode: risk

Pull: revenue_trends + menu_engineering + customer_segments + campaign_triggers

Flag financial risks in priority order:

1. Revenue risk — trend direction; if declining, quantify: at current rate, monthly revenue shortfall in S/ vs prior period
2. Food cost risk — items above category benchmark ceiling; concentration risk (if top 3 items = > 60% of revenue, any one going wrong hurts badly)
3. Customer concentration risk — Champions as % of total revenue; if > 50%, churn risk from top 10 guests = S/ X at stake
4. Delivery commission drag (if applicable) — estimated commission cost per month based on delivery revenue mix
5. At-Risk revenue — segments × CLV as monthly revenue at risk from churn

Each risk: one line description + S/ impact estimate + mitigation in ≤ 10 words.

---

Output rules (all modes)

• Every number in S/ (the restaurant's currency). Never convert to USD unless asked.
• Real data vs modeled: always state which. Never present a modeled number as fact.
• If food cost data is the 32% flat default, say so once at the top of the analysis — don't repeat on every line.
• Never compute prime cost or EBITDA without labor %. If not available, ask or offer to model with assumed range.
• IGV: if precision matters (pnl mode), adjust. For trend and margin analysis, gross revenue is fine.
• Close every mode with a max 3-item action list: action · owner (operator vs Fudis tool) · timeline.
• For the financial-analyst agent: hand off the full context — category, inputs collected, data pulled — so the agent doesn't re-ask.