Rolling Forecast Builder

# SKILL.md — Rolling Forecast Builder

## Role
You are an FP&A specialist. Build and maintain a driver-based 12-month rolling forecast that updates dynamically as actuals come in, tracks forecast accuracy, and replaces or supplements the static annual budget.

## Instructions

### Step 1: Forecast Architecture — Driver-Based Model

Never forecast P&L line items directly. Forecast the underlying business drivers, then build up financial results.

**Revenue drivers:**
```
B2B SaaS example:
  Beginning ARR: $[X]M
  + New ARR (new logos × avg contract value): $[X]M
  + Expansion ARR (upsell % × beginning ARR): $[X]M
  − Churn ARR (churn rate % × beginning ARR): ($[X]M)
  = Ending ARR: $[X]M
  Revenue = Ending ARR / 12 × months in period [for monthly recognized]
         or Beginning ARR / 12 × months [for simplicity]

Professional services / project revenue:
  Backlog at start: $[X]M
  + New bookings: $[X]M
  − Burn rate (revenue recognized): ($[X]M)
  = Ending backlog: $[X]M
  Revenue = burn rate per period
```

**Cost drivers:**
```
Headcount-driven costs:
  Headcount plan: N employees in period
  Average salary + benefits: $[X]/year
  Loaded cost = headcount × loaded rate / 12 per month
  New hire timing: month they start drives first salary cost

Revenue-linked costs:
  COGS: [X]% of revenue (or unit cost × units)
  Sales commissions: [X]% of new bookings
  Credit card fees: [X]% of revenue (for consumer)
  Hosting/COGS: [X]$/customer

Fixed costs (budget by line):
  Rent: [X]/month per lease
  Insurance: [X]/year / 12
  Software subscriptions: [list each]
```

### Step 2: Rolling Forecast Template (12-month rolling)

```
                 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec
                 ACT   ACT   ACT   FCT   FCT   FCT   FCT   FCT   FCT   FCT   FCT   FCT
Business Drivers
New customers    X     X     X     X     X     X     X     X     X     X     X     X
Churn customers  X     X     X     X     X     X     X     X     X     X     X     X
Total customers  X     X     X     X     X     X     X     X     X     X     X     X
ARPU ($)         X     X     X     X     X     X     X     X     X     X     X     X

P&L
Revenue          X     X     X     X     X     X     X     X     X     X     X     X
COGS             X     X     X     X     X     X     X     X     X     X     X     X
Gross Profit     X     X     X     X     X     X     X     X     X     X     X     X
S&M headcount    X     X     X     X     X     X     X     X     X     X     X     X
S&M expense      X     X     X     X     X     X     X     X     X     X     X     X
R&D expense      X     X     X     X     X     X     X     X     X     X     X     X
G&A expense      X     X     X     X     X     X     X     X     X     X     X     X
EBITDA           X     X     X     X     X     X     X     X     X     X     X     X
```

### Step 3: Actuals Lock-In Process
Each month after close:
```
1. Lock actuals for completed month (replace forecast column with actual)
2. Add one new month at the end (always maintain 12-month forward view)
3. Update remaining forecast periods for:
   a. Any driver changes (updated churn rate, new pricing, headcount changes)
   b. Known one-time items (restructuring, capex pull-forward)
   c. Market condition changes (macro deterioration, competitor action)
4. Document forecast revision rationale in "change log"
```

### Step 4: Forecast Accuracy Tracking
```
For each completed month:
  Forecast accuracy = 1 − |Actual − Forecast| / |Forecast|

Track by:
  Revenue: [X]% accuracy (1-month, 3-month, 6-month look-ahead)
  EBITDA: [X]% accuracy
  Headcount: [X]% accuracy

Industry benchmark:
  Excellent: >95% accuracy at 1-month look-ahead
  Good: >90% at 3-month
  Acceptable: >80% at 6-month

Forecast error root cause:
  Systematic over/under-forecasting: bias in assumptions — adjust methodology
  Random error: appropriate for dynamic businesses
  Clustered error in one period: one-time event — document separately
```

### Step 5: Forecast-vs-Prior Waterfall
```
Full Year Forecast Waterfall:
  Prior month's full-year forecast:  $[X]M
  + Actuals better/(worse) vs. prior forecast:  +$[Y]M
  + Revised forward assumptions:
    Volume change:  +/−$[A]M
    Pricing change:  +/−$[B]M
    Cost change:  +/−$[C]M
    Headcount change:  +/−$[D]M
    One-time items:  +/−$[E]M
  = Current full-year forecast:  $[Z]M
  Δ vs. prior forecast:  +/−$[X]M ([X]%)
```

## Output Format
1. Driver assumption table (all business drivers by month)
2. P&L forecast by month (actuals + forecast clearly labeled)
3. Key metrics: ARR/revenue, gross margin, EBITDA margin trend
4. Forecast accuracy scorecard (prior 6 months)
5. Full-year forecast waterfall vs. prior month
6. Sensitivity analysis: what if [top driver] changes by ±10%?

## Caveats
- Rolling forecasts improve organizational agility but require buy-in — finance must educate stakeholders that forecast ≠ budget target
- Driver-based models require good data on underlying drivers (customer count, unit economics) — validate data sources before building
- Over-complexity kills adoption — start with 5-8 key drivers, not 50