The exact prompt + 30 lines of Python that pull a 10-K from EDGAR and hand you a structured IC memo before your coffee's cold.
|
|
View in browser
|
|
◆
AI FINANCE BRIEF
Issue #1 · 2026-06-17 · Weekly
How AI is moving the markets — for pros and retail traders alike.
#3 Deck & IC-memo automation · #1 AI investing workflows
|
|
|
The analyst who shows up to IC with a first draft already written gets to edit instead of starting from a blank page. Here's the workflow that writes the draft.
|
|
THE WORKFLOW
10-K → structured IC memo, in one promptA 10-K is 200 pages engineered to be exhaustive, not useful. The job on IC day isn't to read it — it's to extract the decision: business model, segment economics, capital allocation, the risks that matter, and a defensible thesis. That extraction is pattern-work, and an LLM does it in seconds. Paste the filing (or its EDGAR URL) into the prompt below. You are not asking the AI to make the investment — you're asking it to organize the document so YOU can. That distinction is the whole game. Why it matters for your desk: This is the week's AI shift framed for a desk, not a feed — the one item that removes manual work is the one to pressure-test first.
Copy-paste prompt — run it on a real ticker's latest 10-K:
You are a senior investment analyst writing an investment-committee (IC)
memo. I will paste a company's latest 10-K (or its EDGAR URL). Produce a structured
first-draft IC memo with EXACTLY these sections:
1. BUSINESS MODEL — how the company actually makes money, in 3 sentences.
2. SEGMENT ECONOMICS — revenue mix, growth, and margin by segment (table).
3. CAPITAL ALLOCATION — what they did with cash (capex / buybacks / M&A / debt) and the trend.
4. RISK FACTORS — the 5 risks that actually matter (ignore boilerplate).
5. RED FLAGS — anything in the language, accounting, or disclosures that warrants a second look.
6. THESIS — the bull case in 3 bullets.
7. VARIANT VIEW — where consensus could be wrong (your non-obvious take).
8. VALUATION FRAMING — the 2-3 metrics this name should be judged on and why.
9. DECISION — PASS / DILIGENCE / BUY-AT-PRICE, with the one question that would change your mind.
Rules: cite figures from the filing with the page/section. No generic statements. If a
number isn't in the document, write "not disclosed" — never estimate.
|
Demo: run it on AAPL's latest 10-K (CIK 0000320193). First draft lands in ~6 minutes vs the 2-3 hours a memo takes from scratch. You spend your time on the variant view — the part that's actually yours. |
|
🧰 BUILD IT
Wrap it in 30 lines so it runs itselfDon't copy-paste forever. SEC EDGAR has a free JSON API — no key required, just a User-Agent header. This script pulls a company's latest 10-K and hands the text to your LLM with the memo prompt. Schedule it to fire the morning after any portfolio name files and the draft is waiting in your inbox. Why it matters for your desk: Each move here is a candidate to wire into research or always-on monitoring — pick the one that touches a task you repeat weekly.
edgar_to_memo.py — you direct it, the AI writes the memo:
#!/usr/bin/env python3
"""edgar_to_memo.py — pull a company's latest 10-K from SEC EDGAR (free) and
draft an IC memo with one LLM call. Run it the morning after a portfolio name files."""
import sys, requests
UA = {"User-Agent": "your-name [email protected]"} # SEC requires a UA
def latest_10k_text(cik: str) -> str:
cik = cik.zfill(10)
subs = requests.get(f"https://data.sec.gov/submissions/CIK{cik}.json", headers=UA).json()
recent = subs["filings"]["recent"]
for form, acc, doc in zip(recent["form"], recent["accessionNumber"], recent["primaryDocument"]):
if form == "10-K":
acc_nodash = acc.replace("-", "")
url = f"https://www.sec.gov/Archives/edgar/data/{int(cik)}/{acc_nodash}/{doc}"
return requests.get(url, headers=UA).text
raise SystemExit("No 10-K found")
MEMO_PROMPT = open("ic_memo_prompt.txt").read() # the prompt above, saved to a file
if __name__ == "__main__":
cik = sys.argv[1] # e.g. 0000320193 for AAPL
filing = latest_10k_text(cik)[:180000] # trim to model context window
# send {MEMO_PROMPT}\n\n{filing} to Claude/GPT via your provider SDK, print the memo
print("Filing pulled:", len(filing), "chars — feed to your LLM with MEMO_PROMPT")
|
Non-engineer? Paste this whole block into Claude Code and say “set this up for ticker X and email me the memo.” You run the desk; it runs the plumbing. |
|
📊 MACRO-WITH-AI
Tone-shift detector on the MD&ASame extraction pattern, pointed at the Management Discussion & Analysis. The alpha in an MD&A isn't what it says — it's what changed versus last year. Diff this year's language against last year's and let the model flag where management got more confident, more cautious, or went quiet. Why it matters for your desk: The macro read is where a missed shift costs you — running it as a standing pass catches the silent change before the desk reacts.
Guidance-language diff prompt:
Compare the MD&A section of this 10-K to the prior year's MD&A (pasted below).
List every guidance / tone change: where did management get MORE confident, MORE cautious, or
go SILENT on something they previously emphasized? Classify each shift and quote both versions.
|
Silent omissions are the tell. The model catches them faster than re-reading. |
|
🖥 DECK & MEMO
Memo → 3-slide committee deck, one more promptThe memo is for you; the deck is for the room. One follow-up prompt turns the Markdown memo into a clean 3-slide IC deck — thesis, risks, decision ask. Why it matters for your desk: Walking into the room with a draft instead of a blank page is the difference between reacting to the meeting and steering it.
Markdown-to-deck prompt:
Convert the IC memo above into a 3-slide investment-committee deck:
Slide 1 — Thesis + the one chart that proves it. Slide 2 — Risks & variant view.
Slide 3 — Valuation framing + the decision ask. Output as Markdown I can paste into a
slides tool (use '---' between slides, one headline + 3 bullets per slide).
|
Pipe the Markdown through a slides converter and you walk in deck-ready. |
|
📈 STAY CURRENT
Be the analyst with a draft already writtenReviewers remember who makes their job easy. Showing up to IC with a structured first draft — that they edit instead of you starting blank — changes how senior people see you. You become the operator who ships, the one whose names move faster through committee. That's not about working more hours; it's about removing the blank page from the highest-stakes moment in your week. Why it matters for your desk: Walking into the room with a draft instead of a blank page is the difference between reacting to the meeting and steering it. |
|
⚡ THE STACK
Everything you need this weekWhy it matters for your desk: The right tool stack is the difference between a workflow you actually run every morning and one you abandon after a week.
| SEC EDGAR API |
Free full-text filings + submissions JSON. No key — just a User-Agent header. |
| Claude / ChatGPT |
Long-context model for the 10-K; Claude handles the full filing best. |
| Markdown → PPTX |
Any slides converter (Marp, Slidev, or paste into Gamma) for the deck. |
| A diff tool |
For the year-over-year MD&A tone-shift pass. |
|
|
|
|
|
|
Run this on one name before Friday. Reply “MEMO” and we'll send the prompt + edgar_to_memo.py as a copy-paste pack.
|
|
|
Next week → Next issue: build a portfolio-monitoring agent that emails you a materiality-ranked brief before the desk opens.
|
|
Know an analyst who'd put this to work?
Forward this to one person on your desk. Refer 3 colleagues and we’ll send you our private finance-prompt vault — the exact prompts behind these workflows.
Refer & get the prompt vault →
|
|
|
|
|
|
|
AI Finance Brief · AI mastery for finance professionals & executives.
Educational how-to and tooling — not personalized investment advice. No buy/sell calls.
Read in browser ·
Archive ·
aifinancebrief.com ·
Unsubscribe
|
|