Module 620 min read · AI in Finance

Earnings, Filings & Document Analysis

If there's one place AI saves finance professionals the most raw time, it's here: reading long documents. A 10-K can run 200 pages. An earnings call transcript is an hour of dense talk. A credit agreement is impenetrable legalese. AI can read all of it in seconds and surface exactly what matters. This module is a deep, practical guide to document analysis — the techniques, the prompts, and the verification discipline that keeps it reliable.

The documents that matter and what they contain

The 10-K (annual report)
The comprehensive annual filing. Business description, risk factors, management discussion and analysis (MD&A), audited financials, and footnotes. The footnotes and risk factors are where the real information often hides — and where AI excels at surfacing what you'd miss.
The 10-Q (quarterly report)
The quarterly update — lighter than the 10-K, unaudited, but timely. Good for tracking how things are trending between annual reports.
The 8-K (material events)
Filed when something significant happens — executive changes, major deals, restructuring. The early-warning filing.
Earnings call transcripts
Management's prepared remarks plus the analyst Q&A. The Q&A is gold — it's where management is pushed off-script and where tone and evasion reveal things the numbers don't.
The proxy statement (DEF 14A)
Executive compensation, board structure, governance. Essential for understanding incentives and governance quality.
Where to get primary documents

Always work from primary sources. SEC filings are free at SEC EDGAR. Earnings transcripts and investor presentations are on the company's investor relations page. Download the real document and upload it to the AI — don't rely on the AI's training-data memory of what a filing "probably" says.

The tool choice for long documents

Document length determines the tool. Both context window size and reasoning quality matter.

Document situationBest toolWhy
Standard filing or transcriptClaude200K context handles most filings; best analytical nuance
Very large document or many at onceGemini1M token context holds enormous or multiple documents
Need to run calculations on extracted dataChatGPT (ADA)Extract figures then compute them with real code
Need current context around the filingPerplexityLive search for what happened since and market reaction

High-value document analysis techniques

The targeted extraction

Instead of "summarize this 10-K" (which gives a generic summary), be specific: "From this 10-K, extract everything related to (1) revenue concentration and customer dependence, (2) changes in risk factors versus what you'd expect, and (3) any unusual items in the footnotes. Quote the relevant passages." Specificity transforms output quality.

The risk-factor deep read

Risk factor sections are long and often boilerplate — but changes year over year are meaningful. "Compare the risk factors in this year's 10-K to standard industry risk factors. Which risks are unusual, newly added, or more strongly emphasized than typical?" This surfaces what management is genuinely worried about.

The earnings-call tone analysis

"Analyze the Q&A portion of this earnings call. Where did management give direct, confident answers, and where were they evasive or vague? What topics did analysts press on most?" AI is genuinely good at detecting hedging and deflection in management language.

The MD&A reality check

"In the MD&A, how does management explain the change in revenue and margins? Does their explanation match what the actual numbers show, or are they spinning?" Comparing management narrative to the hard numbers is exactly the kind of cross-referencing AI accelerates.

The verification discipline for documents

When AI quotes or cites a specific figure or passage from a document you uploaded, it's far more reliable than recalling from training data — but still not infallible. For any figure or quote that will drive a decision or appear in your own work, locate it in the actual document and confirm. AI can misattribute, blend passages, or subtly misstate. Trust but verify, especially for direct quotes and specific numbers.

The multi-document synthesis

The real power move: analyzing several documents together. Upload three years of 10-Ks and ask how the risk factors and strategy language evolved. Upload a company's filing alongside two competitors' and ask for a structured comparison. This cross-document synthesis is where AI does in minutes what would take a junior analyst days.

The time math that makes this worth mastering

A thorough manual read of a 10-K with notes might take three to four hours. A skilled AI-assisted read — targeted extraction, risk analysis, cross-referencing against the numbers, then verifying the critical findings — can deliver deeper insight in 30 to 45 minutes. That's not a marginal improvement; it's a different way of working. The analysts who master this read ten companies in the time others read two.

Next

Module 7 brings the analytical pieces together into valuation and scenario analysis — using AI to think through what a company or asset is actually worth, and the intellectual honesty that valuation demands.