The Need for AI in Competitor Analysis
SEC filings (e.g. 10-K, 14A) are a great source for competitive insights. Unfortunately, analyzing those can take a soul-crushing amount of time. Fortunately, AI is perfect for comparing and summarizing dense, long documents.
Below is a short guide and case study on how to use AI for analyzing SEC filings as well as a detailed prompt that you can adapt for your work.
Key Questions that SEC Filings can (not) Answer
The first step in competitor analysis is to be clear about what questions you are answering. Here are a few examples of questions you can start to answer with SEC filings:
- Financial: how does revenue break out between lines of business and geographies?
- Operations: which products are manufactured, where?
- Outlook: how does management view the company’s operations and prospects?
- Risk: what risk factors are emerging or fading?
Notice that I said “start to answer”. That’s because while SEC filings have detailed data, they represent only one piece in the jigsaw puzzle of information that you need to develop a full picture of your competitor. You also need press releases, earnings call transcripts, customer and analyst interviews…
As a case study, I am going to focus on the “risk factors” question.
The Process of Using AI for Competitive Analysis
AI Tools
For this exercise, I used both the “Free” and “Plus” versions of ChatGPT to see if there is a material difference in Quality of Results (QoR). There are also several emerging AI tools for financial analysis:
- AlphaSense: broad market intelligence platform,
- BamSEC: SEC filing and earnings transcript analysis,
- EdmundSEC: financial research tools, and
- Fintool: equity research tools.
Note that these tools can be expensive and may not have the flexibility you need to answer specific questions or provide particular output formats.
The Prompts
I used both simple and long prompts to see if there was a material difference in the QoR.
Simple Prompt
- “Please compare the risk factors in Lattice Semiconductor’s 2024 and 2023 10-K SEC filings and note what has changed.”
Long Prompt
You can see the long prompt here in both MS Word and markdown formats.
- Why a long prompt? See this post.
- Why markdown? See this post.
The long prompt has several distinct sections:
- Your Role: helps ChatGPT understand its place in the world.
- Task Goal & Intent: the “why” of the analysis. It tells ChatGPT the goal of the prompt.
- Needed Analysis: the key questions to answer.
- Sources & Context: guardrails to ensure the use of high-quality data sources.
- Audience: provides guidance about the required level of analysis and reporting.
- Output: 1) ensures that the output is in a format which I can easily use in slides, docs, or email. 2) Also provides a “paper trail” to validate that ChatGPT is providing what I asked for. For example, I knew an early version of my prompt had problems because I limited sources to SEC EDGAR and IR websites, but there was a Reuters news story on the source list.
- Guardrails: further constraints to help ensure that ChatGPT does the analysis and reporting the way I want.
The Results
Simple vs. Long Prompts
The simple prompt had multiple errors:
- Extra Sources: instead of just comparing the specified 10-K filings, the simple prompt also brought in 10-Q filings and news sources, albeit about the target company. More disconcerting was that ChatGPT referenced SEC filings from other, irrelevant companies.
- Wrong Years: the simple prompt compared 10-Ks filed in calendar year 2023 and 2024. That was easily fixed by updating the prompt to say “fiscal year 2023 and fiscal year 2024.”
The long prompt had better QoR and fewer problems sourcing the data.
In both cases, the experience was not perfect. For example, ChatGPT offered to set up alerts about new SEC filings, but instead just gave me instructions on how to use external tools like EDGAR, BamSEC, or Google calendar.
Free vs. Paid ChatGPT
While the free version of ChatGPT had problems sourcing the correct documents, its QoR was not materially different than the paid version.
As an example of sourcing difficulty, the free version of ChatGPT required me to provide URLs or PDFs for the 10-Ks. It would not independently go to the SEC EDGAR website or company investor relations pages. In contrast, the paid version would go extract the correct, needed documents.
In Conclusion
You absolutely should use ChatGPT for comparing and analyzing long, complex documents. But…
- Take the time to develop a long prompt with a full set of specifications and guardrails. It is an investment, but over the long run it will save time and yield better QoR.
- Spot check the information that ChatGPT gives you. AI has come a long way in the last three years, but it still hallucinates. For example, when I asked ChatGPT for the URLs of the four AI enabled financial analysis tools it got three of them right and invented a fake URL for the fourth.
- Consider paying for ChatGPT Plus. $20 / month is a high ROI investment for delivering better results, faster.
Contact me if you would like me to step you through the prompt and process.
AI Content Statement
All text in this post is human generated, written by me. The image was generated by ChatGPT-5.