Your first finance analysis with AI

Let's do a real finance analysis the disciplined way — applying the accuracy crux end to end. The pattern works whether you're in Excel Copilot, a code-executing assistant, or your FP&A tool. Use real (or realistic) financial data — actuals vs. budget, expenses by category, revenue by segment.

A worked example — variance analysis (a top AI finance use):

Step 1 — Frame the question precisely. Finance questions must be exact: "Calculate the variance between actual and budgeted operating expenses by department for Q3, in dollars and percent, and identify the three largest unfavorable variances." Precise questions get precise, checkable answers. Give AI the data (or connect it to the source), and define your terms (what's "operating expense," which periods) — misunderstanding the data is a top source of AI finance errors.

**Step 2 — Have AI compute, not narrate.** This is the crux: ensure the numbers come from a real computation. In Excel Copilot, have it write the formulas (Excel computes). In a code assistant, have it write and run code (real Python computes). In your FP&A tool, the engine computes. Do not accept AI typing variance numbers into a sentence with no computation — force the reliable path: "Write the formula/query to calculate these variances and show it."

Step 3 — Verify the numbers. Before trusting any figure:

• Read the formula/query — did it use the right accounts, periods, and calculation? (Right variance formula? Right sign convention — favorable vs. unfavorable?)
• Reconcile to a source of truth — do the totals tie to your actuals and budget? Does the sum of department variances equal the total variance?
• Sanity-check — are the numbers plausible given what you know? A wildly off variance signals an error (wrong data, wrong formula, or a hallucinated figure).

**Step 4 — Get AI's help on the narrative, grounded in verified numbers.** Once the numbers are computed and verified, AI is genuinely great at the commentary: "Given these verified variances [the numbers], draft a variance commentary explaining the key drivers." AI writes the explanation from your verified numbers — it describes, it doesn't recompute. (Never ask it to "analyze and tell me the variances in a summary" — that reopens the calculator problem.)

Step 5 — Document (the audit trail). Keep the record: the data source, the formula/query used, the computed results, and your review. In finance, this audit trail isn't optional — it's how the analysis is auditable and reproducible (Module 3). "Here's the formula that produced this figure, tied to these source numbers" is exactly what finance and auditors need.

Step 6 — Apply judgment and own it. The variance numbers are computed; the meaning and any decision are yours. Apply your finance judgment to what the variances tell you, and own the analysis — you're accountable for it, not the AI.

What you just did: a complete finance analysis — precise question → computed (not narrated) numbers → verified → AI-drafted narrative from verified numbers → documented audit trail → your judgment and ownership. Notice the difference from casual AI use: the numbers are real (a system computed them), verified, and traceable — which is exactly what makes AI safe for finance.

The two habits from this: compute don't narrate (numbers from a real engine, never AI's prose) and verify + document (reconcile, sanity-check, keep the audit trail). Apply these to every finance task and AI dramatically speeds your work while keeping the accuracy, verification, and auditability finance demands. The rest of the course deepens each application — but the disciplined pattern is the same throughout: let AI accelerate the analysis and narrative, keep the numbers computed and verified, document everything, and own the result.