---
name: ppa-financial-model
description: >
  Models financial impact of price/pack changes including cannibalization and
  competitor response. Trigger when asked to: model P&L by SKU, estimate ROI of
  a PPA initiative, build sensitivity analysis for price or cost changes, simulate
  competitor reaction, quantify cannibalization between own SKUs, or project revenue
  and margin under multiple scenarios. Also trigger for: "modelo financiero PPA",
  "P&L por SKU", "canibalización", "análisis de sensibilidad", "respuesta competitiva",
  "ROI de cambio de precio", "impacto en margen", "escenarios de precio". TomTom MCP
  activates when zone data present. Always renders complete inline HTML dashboard with
  P&L waterfall, cannibalization matrix, sensitivity tornado, and scenario comparison.
---

# ppa-financial-model

Full financial modeling of PPA initiatives: SKU-level P&L, MNL cannibalization matrix,
Monte Carlo sensitivity, and competitor response scenarios. Always dashboard-first.

---

## Plain Language

```
"¿Cuánto gana/pierde la empresa si lanzamos el 600g?"  → sku_pl_model.py
"¿El 600g nuevo canibaliza el 500g existente?"          → cannibalization_matrix.py
"¿Qué tan sensible es el resultado a la inflación?"     → sensitivity_analysis.py
"¿Qué pasa si el competidor baja precio 10%?"           → competitor_response.py
"Dame el resumen ejecutivo financiero"                   → financial_summary_report.py
```

**Core insight (EY PPA p.6 + Polestar 7-step model):**
A PPA change that looks positive in isolation often destroys value through cannibalization
of existing SKUs. The full picture requires modeling incremental revenue — not total revenue
— and net margin after accounting for substitution effects within the portfolio.

---

## Core Equations

### SKU P&L model
```python
# Revenue:
revenue = volume × price

# COGS:
cogs = volume × cost_per_unit   # cost_per_unit = cost_per_g × pack_size_g

# Gross profit:
gross_profit = revenue - cogs
gross_margin = gross_profit / revenue

# Operating profit:
operating_profit = gross_profit - marketing_spend - trade_spend

# Volume (elasticity-adjusted):
volume_new = volume_base × (1 + elasticity × pct_price_change)
```

### MNL Cannibalization matrix (Multinomial Logit)
```python
# Market share for SKU i (MNL):
share_i = exp(V_i) / Σ_j exp(V_j)
V_i = α_i + β × price_i   # utility function

# Diversion ratio from SKU i to SKU j (when SKU i removed/changed):
diversion_i→j = (∂share_j/∂price_i) / (∂share_i/∂price_i)
             = share_j × (1 - share_j) / share_i   (simplified MNL)

# Cannibalization rate: % of new SKU volume coming from own portfolio
cannib_rate = Σ_j diversion_new→j × share_j_baseline

# Net incremental volume:
vol_incremental = vol_new_sku × (1 - cannib_rate)
```

### Sensitivity analysis (Monte Carlo)
```python
# For each parameter θ (elasticity, cost, competitor_price):
# Sample θ ~ N(μ_θ, σ_θ)
# Compute profit(θ) → distribution of outcomes

# Key sensitivities:
sensitivity_i = (∂profit / ∂θ_i) × (σ_θ_i / profit_base)
```

### Competitor response model
```python
# Scenario A: No response
vol_A = vol_base × (1 + E_own × pct_own_price_change)

# Scenario B: Symmetric response (competitor matches % change)
# Own volume reverts partially — net effect ≈ 0.3–0.5 of Scenario A

# Scenario C: Asymmetric (competitor responds only in affected channel)
# Model as weighted average of A and B by channel exposure
```

---

## Workflow

### Step 1 — Data validation
```bash
# data-intake-normalizer: Required: sku_name, price, cost_per_unit, volume
# Optional: marketing_spend, trade_spend, competitor_prices, elasticity
```

### Step 2 — SKU P&L
```bash
python scripts/sku_pl_model.py \
    --skus /mnt/user-data/uploads/sku_data.xlsx \
    --scenario-prices "18.50,20.00,21.00" \
    --elasticity -1.42 \
    --output results/pl_model.json
```

### Step 3 — Cannibalization matrix
```bash
python scripts/cannibalization_matrix.py \
    --skus /mnt/user-data/uploads/sku_data.xlsx \
    --new-sku '{"name":"RefrescoCo 600ml","price":18.50,"size_g":600}' \
    --output results/cannibalization.json
```

### Step 4 — Sensitivity analysis
```bash
python scripts/sensitivity_analysis.py \
    --pl-model results/pl_model.json \
    --params '{"elasticity":{"mean":-1.42,"std":0.3},"cost_pct":{"mean":0,"std":0.10}}' \
    --n-simulations 5000 \
    --output results/sensitivity.json
```

### Step 5 — Competitor response
```bash
python scripts/competitor_response.py \
    --base-volume 1000 \
    --own-price-change 0.08 \
    --elasticity-own -1.42 \
    --elasticity-cross 0.42 \
    --output results/competitor.json
```

### Step 6 — Financial summary + dashboard
```bash
python scripts/financial_summary_report.py \
    --pl results/pl_model.json \
    --cannib results/cannibalization.json \
    --sensitivity results/sensitivity.json \
    --competitor results/competitor.json \
    --output dashboard_data.json
```

**Output sequence:**
```
1. [bash_tool] All scripts
2. [web_search] "PPA financial modeling ROI [category] CPG [year]" +
                "cannibalization rate new SKU launch FMCG benchmark [year]"
3. [TomTom MCP] If zone data → P&L by zone / region
4. [show_widget] Complete inline dashboard
5. [text] NBA + SKU/category/market context
6. [text] Caveats
```

---

## Dashboard Panels (all visible inline)

1. **KPI bar** — incremental revenue, net margin, cannibalization rate,
   worst-case downside (P10), best-case upside (P90), competitor response impact
2. **P&L waterfall** — revenue → gross profit → operating profit by scenario
3. **Cannibalization matrix** — heatmap of diversion ratios between own SKUs
4. **Sensitivity tornado** — top 5 parameters ranked by impact on profit (±1σ)
5. **Scenario comparison table** — base / optimistic / pessimistic / competitor response
6. **Monte Carlo distribution** — profit distribution with P10/P50/P90 markers
7. **TomTom map** — P&L by region/zone when geographic data present

---

## Marketer Insights Layer (MANDATORY)

### Search before benchmarking
```
web_search: "cannibalization rate new SKU FMCG [category] benchmark [year]"
web_search: "competitor response price change CPG [market] [year]"
web_search: "PPA ROI consumer goods [category] LATAM [year]"
```

### Translate to business language

| Technical | Business meaning |
|---|---|
| incremental_revenue = $42K/mo | "After cannibalization, the new SKU adds $42K net — not $80K gross" |
| cannib_rate = 38% | "38% of new SKU buyers come from existing own SKUs — only 62% are truly incremental" |
| P10 = -$12K | "In 10% of scenarios the initiative loses $12K/mo — board needs to know this" |
| diversion_600ml→500ml = 0.45 | "45% of buyers who switch to 600ml were buying our 500ml — half the volume is cannibalized" |
| competitor_response = -8% net | "If competitor matches our price cut, net benefit drops 60% — our move may not be worth it" |

### NBA
- **Incremental framing:** "Always present incremental P&L, not gross — $80K revenue looks great until you subtract $30K cannibalization"
- **Cannibalization threshold:** "If cannib rate > 40%, reconsider the launch — you're mainly redistributing existing volume, not growing the pie"
- **Sensitivity priority:** "Elasticity uncertainty drives 45% of profit variance — validate with A/B test before full rollout"
- **Competitor pre-emption:** "If competitor is likely to respond symmetrically, net benefit drops to $[X] — model this scenario explicitly for CFO"
- **SKU rationalization:** "High diversion (>0.6) from SKU A to B = A is a candidate for discontinuation — simplify portfolio"
- **Launch sequencing:** "Launch new pack in [low-cannibalization channel] first — TT before MT minimizes overlap with existing SKU distribution"

---

## Key Caveats

- **MNL independence:** MNL assumes Independence of Irrelevant Alternatives — may overestimate diversion between close substitutes
- **Cannibalization lag:** Consumer switching takes 4–12 weeks to stabilize — early data overestimates incrementality
- **Monte Carlo assumptions:** Normal distribution for parameters may understate tail risk in high-inflation environments
- **Competitor response timing:** Models assume instantaneous response — actual response lag is typically 4–16 weeks in FMCG
- **Trade spend omission:** P&L without trade spend overstates margin — always include listing fees for new SKU launches

---

## Integration with OS

| Skill | Handoff |
|---|---|
| `data-intake-normalizer` | Validate SKU / P&L input |
| `price-elasticity-modeling` | Elasticity β feeds volume projection |
| `ppa-pack-size-leverage` | Cash margin delta from pack change → P&L model |
| `price-demand-optimization` | p_opt feeds revenue line |
| `margin-simulation` | Agency margin OS → cross-reference with client P&L |
| `trade-promotion-roi` | Trade spend line feeds operating profit |
| `measurement-incrementality` | A/B test validates cannibalization rate ex-post |