How to Use Claude as a Financial Analyst: The Miles Skill Guide

Financial analysis lives or dies on two things: a model whose logic you can trust, and an explanation a non-finance decision-maker can act on. A bare chatbot prompt tends to give you neither — numbers with no visible assumptions, and commentary that reads well but cannot be audited. Using Claude for financial analysis properly means making the workings visible. A real AI financial analyst shows its assumptions before it shows a conclusion.

Miles is that analyst. He is a finance persona you load once into Claude, ChatGPT, or any AI chat, and his first move is to surface the assumptions a number rests on rather than burying them. A forecast you cannot interrogate is a guess in a suit; Miles is built so you can always ask “why this number?” and get a straight answer.

Why generic AI financial analysis is risky

An unguided model hides its assumptions inside the answer, so you cannot tell what it took as given. It reaches for benchmarks and growth rates that sound plausible and may be invented. It reports figures to a false precision that implies more certainty than exists. And it leaves no audit trail, which in finance is the difference between a model you can defend and one you simply hope is right.

What changes with the Miles skill

Miles states the assumptions up front and keeps them separate from the outputs. He structures the work properly — a P&L, a forecast, unit economics — so the logic is legible rather than tangled. He runs scenarios and sensitivities so you can see which inputs actually move the result, and he clearly marks what came from your data versus what is his estimate. The output is something you can take into a board meeting and stand behind.

What it actually produces

Model structures and the formula logic behind them, scenario and sensitivity analysis, variance commentary, board-ready summaries, and plain-English readouts that translate the numbers for people who do not live in spreadsheets. Hand him a set of figures and a question, and he builds the structure to answer it — estimates flagged as estimates.

How to get the most out of it

Give him your real numbers and your own assumptions rather than asking him to invent them. Ask for sensitivities, not just a single answer — the range matters more than the point estimate. And ask him to mark every figure he has estimated, so you know exactly what to verify before you rely on it.

Who this is for

Founders building a model for fundraising or planning, finance leads and FP&A teams, and anyone who has to read a model and decide whether to trust it. It works with Claude, ChatGPT, or any AI chat that accepts a system prompt. For more of the finance stack, the legal & finance skills collection covers budgeting, reporting, and analysis — each a focused assistant rather than a general chatbot. Its outputs explain and structure the numbers; they are not financial advice, and figures should be verified before any decision.