Analyzing Public Filings and Reports
Unlocking Public Filings and Industry Reports
For founders, product managers, and marketers, public financial filings (like 10-Ks) and dense industry analyst reports are goldmines of competitive intelligence. Unlike marketing copy or press releases, SEC filings require companies to disclose actual risks, strategic shifts, and financial realities under penalty of law.
The challenge is not access; it is bandwidth. A standard 10-K easily exceeds 150 pages of dense legalese and accounting tables. Industry reports often span dozens of pages of multi-column text. Without a dedicated research team, extracting actionable insights from these documents is historically a days-long process.
Large Language Models (LLMs) with large context windows, such as Claude or ChatGPT, can process these massive documents in seconds. However, when dealing with financial and strategic data, the primary risk is hallucination (the AI inventing numbers or blending data from its training set with the document). To prevent this, we must use a technique called Anchored Extraction.
The Anchored Extraction Method
Anchored Extraction forces the LLM to act as a strict parser rather than a creative writer. It requires the AI to locate specific data points, quote them verbatim, cite the page number, and explicitly state when information is missing, rather than attempting to guess or calculate it.
When analyzing a competitor’s 10-K, you should focus your AI extraction on three specific sections:
- Item 1 (Business): How the company currently defines its core market and product lines.
- Item 1A (Risk Factors): The legal and operational threats the company acknowledges. This is where you find vulnerabilities (e.g., reliance on a single supplier, high churn rates, regulatory fears).
- Item 7 (Management’s Discussion and Analysis or MD&A): Management’s narrative on why revenue went up or down, and where they are investing next.
Example: Extracting Competitor Vulnerabilities from Item 1A
Standard, weak prompts (e.g., “Summarize the risks in this document”) yield generic summaries that strip away the nuance you need. Instead, use an Anchored Extraction prompt.
The Prompt:
> “Attached is the 2023 10-K for [Competitor Name]. Act as a forensic financial analyst. I need to understand their operational vulnerabilities.
>
> Focus exclusively on ‘Item 1A: Risk Factors’. Extract the 5 most significant risks related to product development, supply chain, or customer retention.
>
> For each risk, provide:
> 1. A 2-sentence summary of the threat.
> 2. The exact, verbatim quote from the document supporting this.
> 3. The page number where the quote is found.
>
> If the document does not mention risks in one of these specific areas, state ‘Not Disclosed’. Do not use outside knowledge.”
By demanding verbatim quotes and page numbers, you create an audit trail. You can quickly open the PDF, jump to the cited page, and verify the context yourself.
Processing Dense Industry PDF Reports
Industry analyst reports present a different challenge: complex formatting. They often feature multi-column layouts, callout boxes, and charts. When an LLM reads a PDF, it typically extracts the text linearly. If the PDF is poorly formatted, the text from column A might bleed into column B, confusing the AI.
To extract market sizing (TAM/SAM) or Compound Annual Growth Rate (CAGR) accurately from these reports, you must instruct the LLM to format its output rigidly and forbid it from doing its own math.
Example: Extracting Market Sizing Data
The Prompt:
> “Analyze the attached industry report. I am looking for explicit market sizing data.
>
> Scan the document and extract all mentions of Total Addressable Market (TAM), Serviceable Addressable Market (SAM), and Compound Annual Growth Rate (CAGR).
>
> Output this as a Markdown table with the following columns: Metric, Value, Year, and Source Page.
>
> CRITICAL: If a metric is not explicitly stated in the text, write ‘Not Disclosed’ in the Value column. Do not calculate, estimate, or infer these numbers yourself.”
Synthesizing Multiple Filings for Trend Analysis
A single 10-K tells you where a company is today. Comparing 10-Ks from the last three years tells you where they are going. LLMs excel at this cross-document synthesis.
If you upload a competitor’s 2021, 2022, and 2023 10-Ks to Claude (which handles multiple large documents exceptionally well), you can track shifting priorities. For example, you might notice that in 2021, the MD&A focused heavily on “user acquisition,” while the 2023 MD&A shifted entirely to “margin expansion and enterprise retention.” This tells you the competitor is likely cutting marketing spend and focusing on upselling existing enterprise accounts, a massive strategic signal for your own positioning.
Applied Exercise: Extracting a Competitor’s Strategic Pivot
To master this, you must apply Anchored Extraction to a real document. In this exercise, you will generate a validated, 1-page strategic brief highlighting a competitor’s operational focus and vulnerabilities.
Step 1: Obtain the Document
Identify a publicly traded competitor in your space (or an adjacent space). Go to their Investor Relations website or use the SEC EDGAR database to download their most recent Annual Report (10-K) as a PDF.
Step 2: Upload and Prompt
Upload the PDF to your LLM of choice (e.g., Claude or ChatGPT). Copy and paste the following prompt exactly:
> “You are a competitive intelligence director. Attached is the latest 10-K for my competitor. I need to understand their strategic pivot and current vulnerabilities.
>
> First, analyze ‘Item 7: Management’s Discussion and Analysis’. Identify the top 3 strategic initiatives management claims they are prioritizing for the upcoming fiscal year.
>
> Second, analyze ‘Item 1A: Risk Factors’. Identify the top 3 risks they highlight regarding their competitive landscape or technology development.
>
> Output a brief with the following structure for each point:
> – Initiative/Risk: (Brief title)
> – Summary: (1-2 sentences explaining it)
> – Verbatim Evidence: (Exact quote from the text)
> – Page Citation: (Page number)”
Step 3: Verify and Synthesize
Once the LLM generates the brief, do not take it at face value. Open your PDF, navigate to the cited page numbers, and verify that the verbatim quotes exist and are not taken out of context.
Your Deliverable:
You now have a verified, factual 1-page brief of your competitor’s strategy and weaknesses. Use this to inform your next product roadmap meeting or to help your marketing team craft positioning that directly targets the vulnerabilities your competitor just legally admitted to having.