Built a static site with AI generated Concall Summaries and Analysis

Investor Toolkits
1

Hey everyone,

While conducting initial analysis on companies, I often found myself searching for earnings call summaries. While the transcripts were available easily, I did not find any good source which provides the summaries as well.
Screener’s pro version provides notes, but the free version limits access to approximately 10 notes per month.

Previously, I used ChatGPT for this purpose, but the process was cumbersome—uploading the document, waiting for the results, and then manually storing the summaries for each company. To streamline this workflow, I built a small side project that leverages GPT models to generate and serve earnings call summaries through a static site.

Web App Link: https://concalls.vercel.app/

Key Details About the App:

  1. This is primarily a makeshift solution I built for my personal analysis needs. I’m sharing it here to see if others find it valuable.
  2. The database currently includes earnings call summaries for approximately 800 companies. Most companies with a market cap above ₹2,000 Cr should be covered if their transcripts are available.
  3. The site currently provides summaries for the last four quarters only.
  4. Currently I do not have written any automation job which will update the site with the transcripts as they are available, but this would be done in the future.
  5. Prompt Used for Generating Summaries:

You are an expert financial analyst tasked with analyzing an earnings call transcript.

Please analyze the transcript and provide insights in markdown format with the following sections:

1. Summary of Quarterly Results

  • Extract and summarize key financial metrics and KPIs mentioned in the call (excluding guidance).
  • Highlight year-over-year and quarter-over-quarter changes, if available.

2. Management Guidance and Outlook

  • Summarize management’s guidance for future quarters/years.
  • Highlight any changes from previous guidance.
  • If available, assess whether management has delivered on past guidance.

3. Key Analyst Concerns

  • Identify major questions or concerns raised by analysts.
  • Summarize how management addressed these concerns.

4. Management Tone Analysis

  • Analyze the overall tone of management (confident, cautious, etc.).
  • Note any significant shifts in tone compared to previous calls.

Guidelines:

  • Only include sections with sufficient information; omit sections that cannot be adequately addressed.
  • Maintain 100% accuracy in the summary. If unsure about a detail, exclude it.
  • Use markdown tables where appropriate for clarity.
  • Format the response as clean markdown that can be directly saved as a .md file.

Current UI looks like this:

image
image3711×1899 422 KB

Looking for Feedback:

I’d love to hear your thoughts on the following:

  • Are you aware of any existing tools that serve this purpose?
  • Do you think such a tool has any value? What improvements should I prioritize?
  • How can I refine the prompt to generate better summaries?

Looking forward to your feedback and suggestions!

45 Likes
3

Hey,

First of all great job on the web app. Love the minimalistic clean UI. The information provided is also precise and adequate.

Not a required feature but I would love to see a graph representing the numbers from past 3-4 qtrs. What was the management guidance and how well they managed to meet the guidance. This could be across different terms. E.g. Revenue, Profit, EBITDA.

Appreciate your work, keep it up

4 Likes
4

Great work! It’s been in my pipeline to make this myself but would be very glad to pay for a solution instead!

Some suggestions for you to turn into a valuable product

Automated uploading and analyzes of concalls should be the topmost priority

Should implemented by the time the next earnings season starts (mid-April). This would 100% make me pay for this product, if I can read summaries of all the previous days released concalls during earnings season.

Prompt

Something should be added to the prompt to identify tailwinds as well.

This is what I personally use which generates everything that your prompt generates except tailwinds and references to page numbers (useful if you want to dig deeper and find out what exactly was said)

Attached is a concall transcript

Identify the people from the management team by their names and position and list them in a bulleted list. These shall, henceforth, be known as management when referred to

Then summarize the concall transcript under four sections, namely

  1. Forward Guidances

Extract any forward guidance given by the management team and list them in a bulleted list.

  1. Headwinds

Please identify any questions that indicate potential headwinds that the company may face in the future and summarize the management response.

  1. Tailwinds

Please identify any questions that indicate potential tailwinds that the company may face in the future and summarize the management response.

  1. Business Understanding

Please identify any questions related to understanding the business and summarize the management response.

Ignore anything mentioned in the transcript before the start of the Q&A section.

Add references to the page number for each bullet point in forward guidance section and each question in headwinds, tailwinds and business understanding section.

Custom Prompts

A POWER USER feature that could be targeted long term could be to allow users to have their own prompts and run it on a transcript or on all new concalls as they are released - POWER USER could buy credits and use them to analyze.

Minor Stuff

Some quick addons that I think could add value

  1. Dark mode (very helpful when one has to do lots of reading)
  2. Link to transcript & presentation (so that we don’t need to open BSE or Screener for it)
  3. Table of Contents sections doesn’t seem to jump to the relevant section when you click on a link, nothing happens
2 Likes
5

Thank you for sharing this amazing tool. There are few companies like carysil, td power, prataap snacks etc which are not covered in the tool. Please try to extend the tool to as many companies as possible.

Thanks a lot again!

2 Likes
6

This is an Excellent work. Kudos !!
Once Humble Suggestion - For future growth and sustainability perspective if a TAB can be added that will be great.
I have the questionnairs ready for AI driven analysis for myself using Chat GPT & Notebook LM.
I can share the prompt if it helps to create the Tab.

2 Likes
7

Commendable job! I too have been experimenting with ChatGPT prompts lately. Instead of a single prompt, have you considered a sequence of prompts? You may be able to analyze the responses and ask follow on questions conditionally.

BTW, I just happened to notice that the search keeps going irrespective of whether a new character is typed or not. Another issue I noticed is that the Markdown format produced by ChatGPT for tables seems to be broken. You may have to insert newlines at appropriate places, if you can’t get it to produce in correct format.

1 Like
8

This alone is a fantastic feature. I have done this too by feeding the old and new concalls and have gotten some useful insights and getting this automatically generated for every transcript is a great feature!

2 Likes
9

Quite a tool there, Pranshukh! Would love to know the foundations required to build something like this ‘on the go’ if u will. Keep tinkering!

1 Like
10

This is something I also need badly. I tried several techniques to capture this and transfer the context across quarters, but the system became too complex and AI started hallucinating.
So I gave up and built just a single transcript analysis instead.

Will give that approach another try and see if it can work

1 Like
11

While I have a software engineering background, this project was coded mostly using AI tools. My job was only prompting, testing and debugging.

Scraped companies and transcript urls from screener (Used python scripts)
Webapp built using next.js.
Code mostly written using Cursor

1 Like
12

I did try with multiple prompts. I tried this chain :

For each quarter, I did 5 api calls :

  1. Fed the ppt and asked AI to extract KPIs for the quarter
  2. Fed past year’s KPIs and current KPIS to ask AI to generate a table tracking the KPIs across the quarters
  3. Fed this quarters transcript and asked for analysis on guidance, management tone, summary
  4. Fed past year’s guidance and this year’s guidance and asked for comparison , analyse whether company is on track to achieve guidance, analyse whether guidance changed from last quarters or not etc etc.
  5. Collate all the information to write the Article for the web

First of all, doing this on scale was getting expensive as each quarter lot’s of tokens were required.
Secondly, the results were not very satisfactory. For some reason, AI was not able to do a very good job in detecting KPIs.
Also every business had different context, and it was messing up guidance figures also.
I wasn’t able to find too much time to experiment and make it work. So first built this basic version.

Will try to simplify this approach and try again sometime.

Regarding the unformatted markdown tables, I am aware of that issue. Need to fix.

3 Likes
13

Sure please share the prompt. I will see how that could be integrated in the current prompt.

14

Thanks for the validation and the prompt. I will try to incorporate the prompt into the currently used

15

Following are the questionnair prompt for AI -

We are conducting a deep-dive fundamental analysis of , and in this context, I need your expert assistance. Please perform a thorough analysis of the documents I have uploaded (including the industry report, last year’s annual report, concall PPTs/notes, financial figures from Screener.in, and other relevant documents).

Your analysis should provide expert insights in response to the questions below. These questions focuses on:
a) Sales growth
b) Operating leverage
c) Margin expansion
d) Debt reduction

The responses should be crisp, concise, and to the point, supported by hard evidence (numbers). The format should be bullet points under each question.

Questions:

Sales Growth - Existing

  1. What has been the growth trend of the industry over the last 10 years?
  2. Has the company’s market share been increasing or decreasing in the growing/shrinking market?
  3. Does the company command a price premium compared to its competitors?
  4. Have the company’s product/service prices been consistently increasing? If yes, by what percentage annually?
  5. Has there been a measurable increase in sales volume over the years? How has this impacted revenue growth?

Sales Growth - New
6. Have new products or services been introduced by the company in the last decade? What percentage of revenue do they now contribute?
7. Has the company expanded into new geographies? What has been the impact of geographic expansion on revenue and market share?
8. Has the company launched specialized or value-added categories of products/services? How have these performed compared to traditional offerings?

Operating Leverage - Existing
9. How effectively has the company utilized its capacity historically? What trends are visible in capacity utilization rates?
10. To what extent has the company adopted automation in its operations, and how has this impacted operational efficiency?
11. What has been the revenue per employee historically? Is there a discernible trend in this metric?
12. What has been the trend of SG&A expenses as a percentage of revenue? Are there signs of consistent reduction?
13. Has the company effectively leveraged common investments for existing products/services (e.g., shared resources, branches, or services)?
Operating Leverage - New
14. With new products/services or transformation initiatives, how effectively is the company projected to utilize its capacity?
15. What level of automation is planned for future operations, and what benefits are expected from this?
16. How is revenue per employee expected to evolve with new initiatives or transformation efforts?
17. Will SG&A expenses as a percentage of revenue reduce in the future due to planned initiatives? By how much?
18. Are there investments (e.g., shared resources, branches, services) that the company can potentially leverage for its new products/services?

Margin Expansion - Existing
19. Has the company consistently improved its margins over the last decade? What has been the CAGR of margin improvement?
20. What initiatives have historically contributed to margin expansion (e.g., cost-cutting, efficiency improvements, pricing strategies)?
21. Are there any notable external factors (e.g., market dynamics, regulatory changes) that have impacted margin trends?

Margin Expansion - New
22. Are there new products/services or initiatives that the company has introduced/planned to improve margins?
23. What specific strategies or initiatives are expected to drive margin expansion in the future?
24. How do planned margin-improvement initiatives compare with industry benchmarks or best practices?

17

I would recommend not focusing on extracting financials. It really isn’t the strength of LLMs and already there are dozens of sites that extract financials and present them in an accurate way (e.g. Screener).

If you want to integrate it into the site - you could use an third party API (I think there are a few which scrap from BSE on github + they are professional paid ones as well, lots of them pretty inexpensive). No point reinventing a (worse) wheel.

IMHO, focus on experimenting with the prompt and getting an automated workflow in place for summarizing and adding new concall transcripts as soon as they are released.

Also, using an OpenAI model for just concall transcripts might be a little bit of an overkill. You can eliminate token costs by using a SLM requiring less than 8GB of RAM on your laptop itself using tools such as Ollema which should give comparable performance.

1 Like
18

@pranshukh Can you update the webapp with the latest transcripts. e.g. I was looking for Shriram Finance Q4 summary and I could only find Q3 in the app. If you need any help with automation of the pipeline, I can help with python / spark / nifi coding.

19

Hi @myloginid
Sure I will write a cron job to perform this automation.
Will try to do it today. If not possible, then definitely will do by the weekend.

20

Updated the app with the currently available summaries. Will setup the cron jobs to update them daily as well.

I am not really getting time to work on this these days, and there are a lot of bugs to be fixed in this app. If anyone is interested to contribute, I will be happy to push this into a git repo where you can contribute.

4 Likes
21

While what Pranshu is doing is pretty impressive, I have started creating small AI generated 10-12 minutes podcasts for companies I am tracking and owning.

While I am able to include 70% of inputs I need in the short / crisp podcast, I am looking to build something more substantial which can capture much more details with low level of hallucination.

How podcasts are helping me is I create 10-15 of these over the weekend and listen to them during Gym sessions or while driving to / from office.

I have started posting them online only this weekend.

You can listen these and comment here or DM me if you want these podcasts for a specific business.

Will keep uploading about 5-10 every week.

  1. Skipper Limited - https://youtu.be/DtbVim7h-FA?si=cGoaL1pLPeN2wfnT
  2. Waaree Energies - https://youtu.be/2wQ9RBS5lSk?si=eTtep5Grkjy5tcRO

@pranshukh Apologies for posting this in your thread as I couldnt find any other thread for AI based summaries of Concalls.

1 Like
22

Hi Pranshu, thanks for making this application. Your application and the user feedback provided some inspiration on what features we needed to add.

We are currently building a similar platform and would like to get your feedback on it as well. You can check out the post here that contains a detailed description here: ArthaLens: Free Quarterly Earnings Call Transcript Summaries, Guidance & Q&A for Indian listed companies

1 Like