Analyze budget variances, actual versus plan

Data analyst · Time: 30 min · Finance

Tools that fit this task

• ChatGPT $20/mo
• Gemini in Google Sheets $14/mo
• Copilot in Excel $20/mo
• Microsoft 365 Copilot $20/mo

Prompt

Role: you are a financial controller.
Context: a budget table for {{period}} with columns line item | division | month | plan | actual | variance | variance %. The company works in {{industry}}, seasonality: {{description of seasonality or "none"}}. Materiality threshold {{percent}}% or {{amount}}.
Task: find all variances above the threshold, rank them by amount, and for each suggest one hypothesis for the cause and one question for the owner. Mark separately the line items where the variance repeats three months in a row.
Format: a table line item | division | plan | actual | variance | % | hypothesis | question. Then a list of systematic variances and a five-sentence comment for management with [CAUSE, confirm] where the cause needs confirmation. Take numbers only from the table.

Steps

1. Prepare a table: line item, division, month, plan, actual, with headers in the first row and no merged cells
2. Add formula columns for the variance in amount and in percent so the model works with ready numbers
3. Give the assistant the range with a prompt to find variances above a threshold and hypotheses for the causes
4. Get the top variances with a hypothesis for the cause and a question to ask the owner
5. Check the two largest variances against the source data and discard false hypotheses
6. Ask for a five-sentence comment for management and replace the hypotheses with confirmed causes

How to check the result

The top 5 variances match a manual sort, every cause in the comment is confirmed by the owner, and the total variance ties to the table

Pitfalls

• The model calls a seasonal peak an anomaly and ignores systematic overspending: set thresholds and seasonal context in the prompt
• A budget with real division names and payroll lines goes into a public service anonymized; inside a corporate Workspace or Copilot you can work on real data

Data that stays out of public AI tools

• CRM and ERP exports with client and employee names: anonymize them before uploading to a public service
• Company financials before publication: into a service only inside a corporate environment or with placeholder names
• API keys to systems and databases: keep them in secrets, never in cells or prompts

Other tasks for data analysts

• Design a KPI dashboard for the business owner
• Get a spreadsheet formula from a plain-language request
• Analyze customer reviews by topic
• Build a competitor price review
• Build a 12-week cash flow forecast
• Explain the numbers to the business owner
• Summarize a long PDF
• Build a quarterly results presentation
• Run a quick counterparty check before a deal

AI tools for data analysts · Prompt: Analyze budget variances, actual versus plan