I’ve been toying with this idea for a few months now, for background I’m a buyside analyst. I spend most of my day reading other people’s ideas, building back of the envelope models and hoping my top picks don’t profit warn on me before bonus discussions.
Much of my job is commoditised I reckon:
- scraping historic data (I’m not talking the stuff BBG give you, I mean segment organic, gaap to non-gaap breakdowns etc.)
- getting management commentary on a topic
These are also some of the most boring parts of my Job, I can’t tell you the number of times it’s been 10 hrs at my desk and all I’ve achieved is extracting segment Info for the last decade. That or I outsource it to sell side, and hope the analysts checked the numbers beforehand (I don’t recommend)
Around Feb I started work on a chrome extension that should allow me to extract data from similar tables across multiple years. Getting it to work for both html and pdf has been a nightmare. But it’s taken a process that would’ve been a few days to maybe 20min max for what I consider a full model.
I then added a notes section that uses a chatGPT plugin for helping me search for a topic across multiple bookmarks, I simply set up my research environment for a stock (save the pdf links as bookmarks). Then type a topic for chatGPT. Only issue here is I reckon I’ll struggle to get my firm to let me use this, even though all the inference is done locally. I’m not sending any files out
I don’t know if I’m exaggerating the issue here, but if this is a wider problem analysts face I’m thinking of releasing a test product for a select group. What d’you guys think?
Alphasense, Factset, and Bloomberg are all doing/working towards very similar ideas - AI based search of transcripts (and probably eventually documents/filings). Alphasense has very good cross doc searching but it does not quite summarise it for you like an AI would.
I think analysts though still value seeing *how* management says something rather than leaving up to an AI to over generalise. An AI might gloss over the fact that management is dissatisfied with a divisions performance and hints they're looking to sell a division in the near future. Basically AI right now struggles to read between lines which is a lot of what commentary is about imo.
Yh we’ve just finished a trial with alphasense, the search function is great, but quite similar to what we get through SIGS on refinitiv. I dunno, I think this is probably a bigger market than people think. These AI models are largely open sourced, transcripts are pretty much commoditised (the APIs are pretty cheap for them). There’s probably space for a smaller player doing the same thing. I suspect alphasense will be hiking the price on BAMsec now they’ve acquired tegus too
What is SIGS? I've used eikon and never come across any function on there similar to Alphasense.
Maybe there's a space for a cheaper, less full-product version of the aforementioned for firms that don't want to splurge. But my point is that most firms already have bbg or similar. If these guys are attacking this with a lot of money, you're creating a product that might soon just be a free addition to something most people already have.
It's similar to the Google problem in tech.
If it’s scraping tables then Bamsec/Edmundsec give you similar tables that you can mash together and download. It’s not the cleanest output (get 3 month data next to 6 or 9 month) but doesn’t take long to clean either. Then I just spot check random line items and make sure they match.
Finchat.io does a decent job of making AI summaries. Edmundsec has Alphasense like AI summaries (I found them to be a little better than AS tbh) while Finchat will dig deeper/much more of a cgpt for finance vs a transcript llm
I should really do more research on this stuff, could’ve saved myself 5 months of coding haha. Although BAMsec is only US right, the problem with European companies is everything gets published on pdf which is a nightmare to scrape
BamSEC/Tegus also own Canalyst which has models with every disclosed metric for a large range of US and non-U.S. companies and updates quarterly.
I build my own models but use canalyst and bamsec to spread historical KPIs and financials. They save me a lot of time.
EdmundSEC is pretty dope, never heard of it.
Quill AI is a pretty good tool for this. For tables already in filings, it gives you numbers + hyperlinks for every cell. For stuff embedded in the text of management/commentary, there are tools for pulling out custom defined KPIs/metrics into a table.
Thanks for the suggestions everyone -- super helpful apps/platforms. I've been using this beta tool called TinyDash -- you can essentially just upload loads of docs (think form 10-Ks, reports, excels) and ask it to build anything from tables to memos. Happy to share it but I've been using it to build memos and just do basic data dashboard stuff that I can easily share to anyone.
Magni sapiente ipsam veniam beatae quibusdam et vitae. Quis inventore officiis voluptatem rem voluptatibus vel. Ut porro asperiores consequuntur doloribus laborum. A fugit minima cum tenetur. Enim aut beatae velit deleniti.
Ipsa debitis a consequatur minus aut. Rerum ullam voluptatem vitae facilis sunt delectus.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...