Analyzing Financial Statements & Tax Returns with AI

For commercial and small-business lenders, the financial spread is the heart of the file. Personal and business tax returns, profit-and-loss statements, balance sheets, and bank statements all have to be read, reconciled, and turned into ratios that drive the credit decision. AI won't replace your spreading software or your judgment, but it can read dense financials in seconds, flag what's unusual, and explain what a number means — so you analyze instead of transcribe.

What You'll Learn

• How to extract and summarize key figures from financials and tax returns
• How to ask AI to calculate and explain common credit ratios
• How to spot inconsistencies and red flags across documents
• The verification discipline that keeps AI-assisted analysis safe

Reading a Business Tax Return in Minutes

A business tax return (Form 1120, 1120-S, or 1065) buries the figures you need across multiple schedules. Redact the entity name and EIN, then paste the relevant text or upload the PDF to Claude (best for long documents) and ask:

This is a [BUSINESS] federal tax return for tax year [YEAR]. Extract
the lender-relevant figures into a clean table:
- Gross receipts / sales
- Cost of goods sold and gross profit
- Officer compensation
- Depreciation and amortization
- Net income (ordinary business income)
- Interest expense
- Any distributions or guaranteed payments to partners

Then compute an approximate EBITDA (net income + interest + taxes +
depreciation + amortization) and show your arithmetic line by line.
If a figure is not present, say "not found" — do not estimate.

The "show your arithmetic" instruction is essential: it lets you check the math at a glance, and the "do not estimate" rule blocks the most common failure — a confident, invented number.

Calculating and Explaining Credit Ratios

Once you have clean figures, AI can compute and, more usefully, explain the ratios that matter:

Using these figures [PASTE REDACTED FIGURES], calculate:
1. Debt-Service Coverage Ratio (DSCR) = net operating income / total
   debt service. Annual debt service is [AMOUNT].
2. Current ratio = current assets / current liabilities.
3. Debt-to-worth = total liabilities / tangible net worth.

For each ratio, show the calculation, state the result, and give a
one-sentence plain-language interpretation of what it tells a lender.
Do not apply any specific approval threshold — I'll compare to our
credit policy.

Note the last instruction: you compute the ratio with AI but apply your institution's policy thresholds yourself. The AI explains; the policy decides.

For consumer lending, the same approach works for debt-to-income:

A borrower has gross monthly income of $6,000 and these monthly debt
payments: [LIST]. Calculate the front-end and back-end DTI ratios,
show the math, and explain in one sentence what each ratio measures.

Spotting Inconsistencies Across Documents

This is where AI quietly shines. When you have several documents that should agree — a P&L, a tax return, and bank statements — AI can cross-check them:

I'm reviewing three documents for one borrower: a year-end P&L, the
business tax return, and 12 months of business bank statements
(redacted). Compare them and flag any inconsistencies:
- Does revenue on the P&L roughly match gross receipts on the return?
- Do the bank deposits roughly support the reported revenue?
- Are there large or unexplained transfers, or owner draws not
  reflected elsewhere?

List each discrepancy with the documents involved. Do not assume fraud
— just flag items that warrant a follow-up question.

The output becomes your stipulation list — the questions to ask the borrower before the file goes to committee. Reading three documents against each other by hand is slow and error-prone; AI does the first pass in seconds, and you investigate what it flags.

Summarizing a Personal Financial Statement

For guarantors and small-business owners, the personal financial statement (PFS) matters as much as the business. Ask:

Summarize this personal financial statement (redacted) for a credit
file: total assets, total liabilities, net worth, liquid assets, and
contingent liabilities. Note concentration risk if most net worth is
in a single illiquid asset like one property or the business itself.
Bullet points. Flag anything that needs documentation.

The Verification Discipline

AI-assisted financial analysis comes with one ironclad rule: AI's numbers are a draft until you reconcile them to the source. Language models are pattern matchers, not calculators — they can transpose digits, misread a schedule, or smooth over a figure. Your workflow:

1. Have AI extract and compute, showing all arithmetic.
2. Spot-check every key figure against the actual document.
3. Re-run any ratio that drives the decision in your spreading software or a calculator.
4. Apply your credit policy thresholds yourself.

Used this way, AI turns hours of transcription and tie-out into minutes of review — without ever putting an unverified number in front of a credit committee.

Key Takeaways

• AI can extract lender-relevant figures from tax returns and financials in minutes — always require it to show arithmetic and say "not found" rather than estimate.
• Use AI to calculate ratios like DSCR, current ratio, and DTI, and to explain them in plain language — but apply your own policy thresholds.
• Cross-document checks (P&L vs. tax return vs. bank statements) turn AI into a fast first-pass reviewer that builds your stipulation list.
• Treat every AI figure as a draft: reconcile to the source and re-run decision-driving ratios before relying on them.