corporate_fp_and_a v1.0.0

name: corporate_fp_and_a description: "Run the full corporate FP&A cycle: driver-based budgets, flexible budgets, Level 2/3 variance analysis (price, volume, mix, quantity), rolling 12-month forecasts, MAPE/WAPE forecast-accuracy tracking, and management reporting packs. Triggers on requests to build a budget, explain budget-vs-actual variances, set up a rolling forecast, measure forecast accuracy, or prepare a monthly finance pack." metadata: { "includeInPrompt": true }

Corporate FP&A

Purpose

Do the complete financial planning and analysis loop, not just one piece of it: set the budget from drivers, flex it at actual volume, decompose every variance into its causes, roll the forecast forward every month, track how accurate the forecasts are, and report it in a management pack that drives decisions.

Workflow

1. Scope the ask. Decide which parts of the cycle the user needs: budget build, variance teardown, reforecast, accuracy review, or the monthly pack. Say what you are skipping, do not quietly skip it.

1. Budget from drivers, not line items. Model the P&L as operating volume x unit rate x mix/timing, with a short assumption log for every driver. A budget nobody can explain driver-by-driver is a guess, and it will be treated as one at variance review. The budget stays frozen as the accountability baseline. A forecast is never a replacement for it.

1. Flex before you judge. When actuals land, build the flexible budget first: restate every variable line at the actual volume. Comparing actuals to a budget written for a different volume answers the wrong question. Fixed costs do not flex. Use bin/fpa_tools.py flex.

1. Decompose every material variance. Run Level 2 first (static-budget variance = flexible-budget variance + sales-volume variance). For revenue and any multi-product line, run Level 3: split the volume variance into quantity and mix, and split flexible variance into price and efficiency pieces on the cost side. Read references/variance-analysis.md for the formulas. One blended number is not analysis.

1. Label each variance with cause and owner. Price/volume/mix/timing/one-off. Each material variance gets: the dollar effect, which driver moved, the owner of that driver, and the action. "Gross margin was below budget" is not a finding.

1. Reforecast, do not re-budget. Roll the 12-month forecast forward: close elapsed months with actuals, bias-correct the remaining months against the last 3 closes, keep the horizon constant. Use bin/fpa_tools.py roll. If a driver has moved more than 5% off plan mid-month, update the assumption now, do not wait for month-end.

1. Track accuracy like a metric. After every close, run bin/fpa_tools.py accuracy on the forecast history. Report MAPE and WAPE together, plus bias (mean signed error). A forecast that is always 5% optimistic is a broken process even when its MAPE looks fine.

1. Write the pack to drive decisions. One page per section: headline numbers, variance waterfall, driver chart, forecast-vs-budget bridge, risks and actions. If a report would not change a decision, cut it. Read references/management-reporting.md for the pack structure.

Output Contract

• Every engagement produces: the cycle steps used and skipped, the driver assumption log, flexed budget, variance teardown with F/U labels tied to operating income, reforecast with bias note, accuracy snapshot, and the action list with owners.
• Variance sub-parts must reconcile to their parents: price + efficiency = flexible variance, quantity + mix = volume variance, flexible + volume = static. A reconciliation that does not tie out means an input is wrong, never publish it.
• Numbers come from the user's data or verified sources. Never invent drivers, rates, or history to make a template look complete.

Operating Rules

• Favorable (F) means it raises operating income, Unfavorable (U) means it lowers it. Apply this one rule everywhere; never decide sign case-by-case.
• Keep budget and forecast as separate baselines, always. The budget is the commitment; the forecast is the latest estimate.
• Rolling forecast horizon stays constant every month, typically 12 months. Do not let it collapse at year-end.
• Sample CSVs live in assets/ and demonstrate the bin/fpa_tools.py input schemas. Validate inputs before running: numeric rates and quantities, matching product/line keys across files.
• Heavy method (rolling-forecast mechanics, bias correction, scenario sets, pack layout) lives in references/; SKILL.md is the decision path only.