Turn Claude into a research analyst (3-prompt company breakdown kit)
Hey, it's Cindy ๐ฑ The kit is three prompts that each do one job: a deep dive on the numbers, a peer check on the valuation, and a bear case that tries to break the thesis.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
before you start
Before you start: get the real filing
Claude reasons best when you hand it the actual document instead of asking from memory. For any US-listed company, the primary source is free.
๐ SEC EDGAR (sec.gov/edgar/search) is the official, free database of every US company filing. Full-text search covers filings back to 2001, and the archive goes to 1993. Grab the latest 10-K (annual report) or 10-Q (quarterly) and copy the financial statements, or download the PDF.
๐ Upload it into Claude. Claude can read PDFs up to 32MB and hold roughly 500 pages of context in one session, so a full 10-K fits without chopping it up (Claude file uploads).
๐ For non-US or very recent numbers, turn on Claude's web search. It runs on all plans worldwide and returns cited sources you can click to fact-check (claude.com/blog/web-search). Still verify every figure against the primary filing.
the numbers
Prompt 1: the deep dive
This one turns a raw filing into a scored, structured breakdown of the financials instead of a vibe. Use it first, right after you paste or upload the 10-K or 10-Q.
๐ง the deep dive
You are a buy-side equity analyst. I've given you [COMPANY]'s latest filing. Do NOT use outside memory for the numbers, pull every figure from the document I gave you and cite the line item.
Produce a structured deep dive with these sections:
1. Revenue: last 3 periods, growth rate, and what's driving it.
2. Margins: gross, operating, and net, with the trend and any one-offs.
3. Cash flow: operating cash flow vs net income, capex, and free cash flow.
4. Balance sheet: cash, total debt, and whether the company could survive a bad year.
5. Growth quality: is growth coming from the core business or from acquisitions/one-offs?
Score each section from 1 to 5 with a one-line reason. End with a 3-sentence plain summary of the financial health. Flag anything you could NOT find in the filing rather than guessing.
Why the "pull from the document" instruction matters: without it, a model can fall back on stale training data and state an old number as current. Anchoring it to the filing and asking it to flag gaps keeps it honest.
the context
Prompt 2: the peer check
A number means nothing in isolation. This prompt lines the company up against its real competitors so you can see if the price is actually cheap or expensive relative to the group.
โ๏ธ the peer check
I'm valuing [COMPANY]. List its 3 to 5 closest direct competitors (same industry, similar size where possible).
Build a comparison table with one row per company and these columns: P/E ratio, price-to-sales, revenue growth %, gross margin %, and operating margin %.
Then answer in plain language:
1. On these metrics, is [COMPANY] priced above, in line with, or below its peers?
2. If it's more expensive, what would justify the premium (faster growth, higher margins)?
3. If it's cheaper, is that a bargain or a warning sign?
For any valuation multiple, tell me the date or source so I can verify it, and mark anything that may be out of date.
Heads-up: valuation multiples like P/E move every day with the share price, and a model's training has a knowledge cutoff, so treat any multiple it gives as a starting point. Turn on web search for live numbers, then confirm against a broker or the exchange. Claude's web search returns citations you can click to check (claude.com/blog/web-search).
argue against yourself
Prompt 3: the bear case
This is the one most people skip and the reason the kit exists. You ask Claude to argue against the company, so you meet the strongest counter-argument before you commit, not after.
๐ป the bear case
You are now a short-seller who thinks [COMPANY] is a bad investment. Using the filing and what you know about the industry, build the strongest honest bear case.
Cover:
1. The 3 biggest risks to the business (competition, regulation, customer concentration, debt, margin pressure).
2. What in the latest filing supports each risk (quote the line).
3. The specific things that would PROVE the bull thesis wrong, so I know my invalidation points.
4. One scenario where this company loses half its value, and what would have to happen.
Be blunt. Don't hedge. I want the argument I'd have to beat before buying.
Why this beats a generic "is this a good stock" question: forcing the model into the opposing seat surfaces the risks a bullish prompt glosses over, and the "what would prove the thesis wrong" list gives you a real exit rule instead of a gut feeling.
go further
3 bonus prompts to run next
๐งพ Earnings-call decoder: "Here is the latest earnings call transcript. Summarise what management is excited about, what they're quietly downplaying, and any change in tone versus a confident quarter. Quote the exact lines."
๐ Red-flag scan: "Scan this 10-K for accounting or disclosure red flags: rising receivables vs revenue, growing inventory, heavy stock-based comp, related-party deals, or going-concern language. List each with the line you found it on."
๐งญ One-page verdict: "Combine the deep dive, peer check, and bear case into a single one-page memo: the bull case in 3 lines, the bear case in 3 lines, the 2 numbers I should watch each quarter, and the price levels that would change my mind. Neutral tone, no recommendation."
the craft
How to get the most out of it
๐ฑ If you're new to this: start with just Prompt 1 on one company you already understand, and read Claude's output next to the actual filing line by line. That comparison teaches you where the numbers live faster than any course.
๐ If you're advanced: save the three prompts as a Claude Project so the filing and your house style persist across sessions, and if you have access to Claude for Financial Services, its live connectors (LSEG market data, Moody's ratings) pull real-time figures straight in (anthropic.com). That's an enterprise offering, not the default consumer plan.
what to watch
The honest limitation
This is a research assistant, not a crystal ball. A model has a knowledge cutoff and can get a number wrong or read a table badly, so every figure it gives you has to be checked against the primary filing on SEC EDGAR. This is not financial advice, and no prompt replaces your own judgement or a licensed adviser. The point of the kit is to make you read the filing and argue both sides before you have an opinion, that's the edge, not the prediction.
the waitlist is now open
AI System Club is launching soon
Build AI systems with me through recorded lessons you can follow at your own pace. Join the waitlist for first access and the founding price.
btw this list gets first dibs the moment doors open :)