ChatGPT for financial statement analysis: ratios, trends and what to check

Split the work: the spreadsheet for numbers, AI for reading

The most reliable way to use AI here is to keep the arithmetic out of the chat.

• In the spreadsheet: ratios, growth rates, common-size statements, and anything else that's a calculation. Formulas can be checked and rerun next period.
• With AI: reading the notes and the management commentary for possible causes, spotting which changes matter, suggesting questions for management, and drafting commentary for you to check.

AI can add up wrongly, take a figure from the wrong period, or use the wrong line. In a spreadsheet those mistakes are visible. In a fluent paragraph they aren't.

The ratios to work out yourself

Definitions vary between analysts. The SEC's Beginners' Guide to Financial Statements, for example, divides total liabilities by shareholders' equity for debt-to-equity, while some analysts use interest-bearing debt only. Pick one definition, write it down, and use it the same way every period.

Two formulas worth keeping. A common-size income statement shows each line as a share of revenue. With revenue in row 2:

And a growth rate that still makes sense when last year's figure was negative:

Give the AI the right material

• the statements for at least two periods, so it has something to compare;
• your ratio table, with a note that the figures are already calculated and shouldn't be recalculated;
• the notes to the financial statements and any management commentary. For a public company these are in its filings, which you can find with SEC EDGAR full-text search;
• anything you already know, such as a one-off sale or a change in the business.

A private company's figures are usually confidential. Check what your firm or company allows before they go into an AI tool: see the client data guide.

Attached: income statements and balance sheets for [years], the notes, and my ratio table (already calculated in Excel; don't recalculate it).

1. List the five most significant changes in the ratios, largest first, with the figures from my table.
2. For each, say what in the statements or notes might explain it, quoting the note or line you rely on. If nothing in the documents explains it, write "not explained by the documents".
3. List the questions I should ask management about each change.
4. Flag anything that affects comparability: one-off items, changes in accounting policy, reclassifications between periods, acquisitions or disposals.

Check the story

1. The figures

Every figure in the AI's text should match your table and the statements. Watch for figures taken from the wrong year or the wrong line, and for percentages it has worked out itself.

2. The explanations

Open the note it quotes. A note can mention something without explaining the movement: "consistent with" isn't "caused by". The checks in the flux analysis guide apply here too.

3. Comparability

Before trusting a trend, check that the periods are like for like. A reclassification, a policy change, an acquisition or a restated prior year can make a ratio move with no change in the business. Look in the notes for each of these, whether or not the AI flagged them.

4. Direction and common sense

Check that each change is described the right way round: a higher days sales outstanding means customers are paying more slowly. And check why a ratio moved. A current ratio can improve simply because debt was reclassified from current to long-term.

Maybe. It could also be a change in product mix, lower input costs, or costs moved out of cost of sales. If the notes show a reclassification, the improvement may not be an improvement at all. The commentary should say only what the documents support, and list the rest as questions.

Benchmarks: don't take the AI's word for them

Asked how a ratio compares with the industry, an AI will often give a typical range from memory. It may sound authoritative, but you can't cite it or check where it came from, and it may be out of date.

Get benchmarks from a source you can name: the filings of comparable public companies (EDGAR has them), an industry body's published survey, or your firm's own data across clients. Then compare like with like, using the same ratio definitions. The AI can help you read peers' filings once you've chosen them.

Turning the analysis into commentary

For a management report or a client meeting, a simple order works: the figure, then the cause the documents support, then the open questions. Keep "not yet explained" visible rather than filling the gap with a plausible reason. A reader can act on an honest open question. A confident wrong reason does damage.

For more prompts in the same style, each with what to check, see ChatGPT prompts for accountants.