AI turned my Report into a Video
How AI is improving my research process
Last week, Hidden Gems Investing published a 30-page Special Report on Douglas Dynamics (PLOW), a snowplow manufacturer with over 50% market share and a clear catalyst for earnings to grow 54% this year. I see 104% upside over three years in my Base case valuation.
I used Google’s NotebookLM to turn the report into an 8-minute video podcast (see above).
I was amazed by the result.
The video includes two AI presenters summarizing the key points from the report to provide an overview of PLOW, its moat, cyclical rebound, the replacement cycle and pent-up demand, valuation, and catalyst path.
Check the video out above if you want to see an example of what AI can now do or simply prefer the video format to reading a 30-page report.
Here’s how I did it.
NotebookLM is a free tool from Google where you can upload sources to a notebook and query them and/or produce other outputs. Because NotebookLM only uses the sources you upload, it has far fewer hallucinations than other LLMs.
To create this video, I simply:
Uploaded the Hidden Gems Investing report.
Clicked “Video Overview”.
Clicked “Explainer”, then “Generate”.
That was it.
NotebookLM allows you to tell the AI what to focus on as well. The video you see at the top of this post included some additional instructions, but the first attempt with no prompt using just the three steps above was good enough that I could have published it.
The video had only one error - it states that the report includes the full transcripts from 13 interviews with industry sources. That’s not true; I conducted 13 interviews as part of my research which informed my analysis, but they are not reproduced.
Of course, this process is still garbage in, garbage out. That is why I believe in a world of AI hallucinations and shortcuts, feeding genuine primary research into your investment process is more important than ever.
I believe the video is high quality because it draws entirely upon the 30-page Special Report, which was the result of months of research, 13 long-form interviews with industry sources, and dozens of short interviews with dealers. I believe PLOW will grow EBIT 54% this year, 32ppts above consensus estimates, and see 104% upside over three years with a superior IRR because strong results this year should act as the catalyst for most of that upside. Don’t forget to read the Special Report. If you are a free subscriber, take a look at the video summary above.
Here are a couple more examples of how AI is helping my investment process.
If instead of clicking “Video Overview” in NotebookLM you select “Audio Overview” then “Debate”, NotebookLM will produce a 20-minute podcast between two AI hosts debating the key points of the uploaded documents and stress-testing its analysis.
Selecting “Critique” instead of “Debate” takes this further by creating a review to find holes in your research.
Another tool I find increasingly useful is the ability to create Custom GPTs in ChatGPT.
By going to “Explore GPTs” then “Create”, I have created a suite of Custom GPTs that are able to automate parts of my investment process. For example, every week I read all the earnings transcripts from companies in the same industries as stocks I am invested in, looking for any relevant read-across.
Instead of reading every transcript, I now read the most important ones in their entirety but upload ones where there may only be an occasional read-across into ChatGPT.
If you do that in ChatGPT’s standard chatbot, the quality and approach of the answers will vary each time, and many answers will not be precise and try to give you what it thinks you want.
I created a CustomGPT called “Earnings analysis” to solve this problem.
The GPT has instructions on what to look for when deciding whether a transcript or results statement has a genuine read-across to a company I’m invested in, to extract the key read-across, quote important passages in full and verbatim so I don’t have to find them, to ignore generic / imprecise commentary, double-check its work, and conclude whether I should read the transcript in full.
No coding was needed as instructions can be directly written into the “Instructions” box.
There is also the ability to upload model answers and other relevant information as “Knowledge” files, which it can refer to on an ongoing basis.
By using a CustomGPT, I am able to ensure answers are consistent and high quality. That allows me to follow more companies, extract more insights, and do so quickly.
Let me know in the comments how AI is improving your research process.
If you enjoyed this article, click “like” or “share” below to help it gain more visibility!
This article is for informational purposes only and is not investment advice. Plural Investing, LLC currently holds a position in Douglas Dynamics. Read important disclosures here.