Analysis Tool · Claude for PE

There is one thing language models are famously shaky at: arithmetic done in their heads. The analysis tool fixes this. When a task involves real calculation, Claude can run the numbers properly rather than estimating them, which matters a great deal when those numbers are heading into a memo.

The problem it solves

Ask any AI to multiply two large numbers in prose and you may get a confident, wrong answer. For PE work, where a misplaced growth rate or a fumbled margin can mislead a whole screen, that is unacceptable. The analysis tool lets Claude step out of prose and actually compute, the same way a calculator gives you the real figure rather than a guess.

Estimating versus computing

Without the analysis tool	With the analysis tool
Figures reasoned in prose, sometimes approximate	Figures genuinely calculated, then reported
Hard to see how a number was reached	Working is shown and can be checked
Risk of small arithmetic slips on big sums	Reliable arithmetic on large or many numbers
Fine for rough order-of-magnitude sense	Right for figures that enter a deliverable

The analysis tool moves calculation from approximate to exact.

When it tends to kick in

You do not switch the analysis tool on with a button. Claude reaches for it when a request clearly calls for calculation, particularly across an attached spreadsheet. It is most useful for:

Growth and margin math across several years of a financial model.
Reconciling totals that should agree but do not, so you can find where they diverge.
Cohort or churn calculations from a table of raw figures.
Quick sensitivities, such as recasting revenue at a different assumed growth rate.

What it looks like in practice

Attach the figures, ask for the calculation, and Claude does the math and reports the result. Crucially, it can show its working, so you are not asked to trust a number on faith.

You

Project_Atlas_financials.xlsx 3 tabs

Calculate the year-on-year revenue growth and the gross margin for each of the last three years from the attached figures. Show the workings.

Claude

Revenue growth. Year 1 to Year 2: 27%. Year 2 to Year 3: 24%.

Gross margin. Year 1: 77%. Year 2: 78%. Year 3: 79%.

Each figure is computed from the Revenue and Cost of Sales rows on the P&L tab. The growth softening from 27% to 24% is worth a question to management.

A prompt that puts it to work

Prompt · Reconcile and recompute from a model

You are assisting a Private Equity associate. I’ve attached the target’s financial model.

Calculate year-on-year revenue growth and gross margin for each year shown. Show the workings for each.
Check whether the revenue subtotals add up to the stated total, and flag any line that does not reconcile.
Recast Year 3 revenue assuming growth holds at the Year 2 rate, and tell me the difference.

Report figures in a table. Note any number the model leaves ambiguous.

Reading and verifying the working

The analysis tool makes Claude’s arithmetic reliable, but it does not check that the source figures are right, or that you asked the right question. Two habits keep you safe.

When Claude shows its working, read it. Confirm the inputs are the right cells, that the formula matches what you intended, and that the result is the sensible order of magnitude. This is the same discipline as the rest of your work with Claude, set out in Verify Everything, and it is what lets you put a computed figure into a deliverable with confidence.

What to take from this

The analysis tool turns Claude’s weakest area, mental arithmetic, into a strength by letting it run real calculations on your numbers. Lean on it whenever figures will enter a deliverable: growth, margins, reconciliations, and quick sensitivities. Ask for the calculation explicitly, read the working it shows, and always confirm the inputs were the figures you meant. The tool guarantees the math is right; you guarantee the right question was asked. To watch it check a model mid-screen, see how it catches a reconciliation gap in the Project Atlas walkthrough.