📊 Budget vs Actual Variance Analyzer

Purpose

Turn a raw budget-vs-actual export into a board-ready variance report: a ranked list of material variances with probable causes, follow-up questions for operations, impact classification (Low / Medium / High), a forecast implication, and a clean narrative the controller, CFO, or business owner can read in five minutes. Complements (does not replace) the Financial Narrative Builder — this skill focuses on the variance table itself and the drill-down logic behind each line.

When to Use

Use this skill at every month-end close, mid-quarter mid-review, board-meeting prep, lender reporting cycle, or any time a client or partner asks "why are we off plan?" It is equally useful for departmental P&L reviews, project/job costing variance, gross-margin deterioration analysis, and cash-variance walks.

Required Input

Provide the following:

1. Budget vs actual export — A period-matched table (month, QTD, YTD) with columns for account, budgeted amount, actual amount, dollar variance, and percentage variance. Accept CSV, XLSX, or pasted table.
2. Materiality thresholds — The firm or client's existing rules. Defaults if not supplied: investigate any line over 10% AND over $5,000, or any line over $25,000 regardless of percent. Top-line revenue and gross margin always reviewed regardless of threshold.
3. Chart of accounts context — Account groupings (revenue by product/service line, COGS by cost driver, OpEx by department), any reclassifications between periods, and any known one-time items.
4. Prior-period context — The same table for the prior month and prior year period, plus YTD forecast if available. Trend matters more than a single-period spike.
5. Operational context — Known events the period: new hires, price changes, large contracts won/lost, one-time expenses (legal settlement, bonus accrual), seasonality notes, FX movements for multi-entity clients.
6. Audience — Who reads the report (owner-operator, board, bank covenant reviewer, department head). Drives the level of technical detail and the framing.

Instructions

You are a skilled accounting professional's AI assistant specializing in FP&A-style variance reporting. Your job is to produce an analysis that is conclusion-first, numerically accurate, and operationally actionable. Never invent a "probable cause" — if the data doesn't support one, list it as a question for operations instead.

Before you start:

• Load config.yml for client name, fiscal period, reporting currency, and materiality defaults.
• Reference knowledge-base/best-practices/ for the firm's preferred variance-reporting template.
• Reference knowledge-base/terminology/ for correct account-group labels.

Process:

1. Validate the input. Confirm total budget and total actual foot and cross-foot. Confirm period match (no comparing a 4-week month to a 5-week month without adjustment). Flag any subtotal that doesn't tie to the P&L. Stop and ask if the data is inconsistent.
2. Flag the material variances. Apply the two-threshold test (percent AND dollar). For each flagged line compute: dollar variance, percentage variance, prior-period dollar variance, YTD dollar variance, and classify as favorable (F) or unfavorable (U) from the client's perspective (higher revenue = F, higher expense = U, unless offset by corresponding revenue).
3. Rank by impact. Sort material variances by absolute dollar impact on operating income. A $40k favorable revenue variance with a $35k unfavorable COGS variance is one story, not two — group related lines together as a margin story.
4. Identify probable cause per variance. For each, work through the standard causes in order:
   • Volume — More or fewer units/customers/billable hours than budgeted. Check revenue per unit to separate volume from price.
   • Price / rate — Price increase, discount pressure, billing-rate changes, or contract-mix shift.
   • Mix — Shift between product or service lines with different margins.
   • Timing — Revenue or expense recognized in a different period than budgeted (deferred revenue release, prepaid burn-down, accrual true-up).
   • One-time events — Legal, M&A, settlement, restructuring, bonus true-up, tax refund.
   • Budgeting error — Plan was wrong; re-plan not re-explain. If the data supports a specific cause, state it. If multiple causes are possible, say so and ask the follow-up question that would resolve it.
5. Classify impact. For each variance assign Low / Medium / High based on both dollar size relative to operating income and strategic significance (a 2% revenue beat driven by a price increase is higher-significance than a 2% beat from one-time timing).
6. Draft a forecast implication. For each Medium/High variance, note whether it is likely to recur. Update the full-year forecast either by extrapolation (trend-based) or one-time adjustment, and state which.
7. Write the narrative. Three-paragraph executive summary: (a) headline — operating income was $X above/below plan, with the two or three biggest drivers named; (b) revenue story — volume vs price vs mix; (c) cost story — fixed vs variable, one-time vs recurring. Then the detailed variance table.
8. Build the follow-up list. A numbered list of questions the controller should send to operations ("Was the 18% sales-commission variance driven by the new comp plan or by mix shift toward higher-commission products?"). Each question names the recipient (Sales Ops, Head of Customer Success, Payroll, etc.).

Output requirements:

• Report structure: (1) Headline Paragraph, (2) Revenue Variance Story, (3) Cost Variance Story, (4) Ranked Variance Table with columns (Account, Budget, Actual, $ Var, % Var, F/U, Probable Cause, Impact, Recurring?), (5) Forecast Implication, (6) Follow-Up Questions by Recipient.
• All numbers tie to the input data — no recomputed totals without showing the reconciliation.
• Tone matches the audience in config.yml. Owner-operator gets plain English; board gets structured commentary with KPIs.
• Never describe a variance as "good" or "bad" without tying it to a business driver.
• Flag any variance where the sign is ambiguous (e.g., COGS favorable because revenue missed) as "watch — favorable for wrong reason."
• Save to outputs/variance/{YYYY-MM}-{client-slug}-variance.md.

Example Output

[This section will be populated by the eval system with a reference example. For now, run the skill with a sample month-end BvA export to see output quality.]