Any views among this community about Factset (FDS), given its relevance to stock research? It’s among those names getting caught up in the AI-displacement narrative sell-off.
When/if they figure out how to stop AI hallucinations, I don’t know how any of these data research companies survive, unless they embrace AI. I would think they eventually should be able to supply the same (or better) research with a lot less overhead.
4 Likes
I hadn’t looked at Factset as a company before. High Quality ratings, Momentum in the toilet, the rest of the metrics indifferent. Not a stock that would show up in my screens.
With reference to AI impact, five years from now I think it will prove to be a slight net positive. What they provide definitely benefits from being done once then rented out. People will try for a while to do something better and then mostly back off when they keep hitting the exceptions. See the P123 Reference Section essay on why use their standardized data elements instead of doing it yourself.
Since my time horizon isn’t five years and this looks to be a falling knife situation, it’s stay away.
Cheers,
Rich
2 Likes
I guess I’m dumb, or maybe I’m just unaware of FactSet’s entire suite of products, but I don’t see how providers of unique proprietary clean datasets are in a bad position for AI. Wouldn’t they sell API licenses hand over fist to everyone ‘vibecoding” their own front ends and AI agents to access their data? Is the assumption here that everyone is going to collect, clean and store all this data on their own through AI agents? Isn’t memory going to be constrained at a huge cost premium for the foreseeable future due to the AI buildout?
2 Likes
I tend to agree, it looks like indiscriminate selling. Those with little or no proprietary data (Gartner? others?) might very well be disrupted. But for businesses like Factset or Constellation (more of a vertical software aggregator), I’m skeptical that AI will be able to replicate their highly specialized solutions anytime soon.
2 Likes
This is so interesting!!! Anthropic came out with Cowork that can be accessed on Mac computers now. The WSJ link at the end suggests that Cowork contributed to FactSet’s decline!!!
Cowork is basically a highly organized agent that can work within a specified file and go out on the web through a Chrome plug-in. It is now available for Mac iOS. I have used in and OpenAI’s Codex.
The Chrome plug-in is maybe not perfected yet, as I asked it about this thread and it posted without my permission below (not my intent and not sure I would have weighed in)! I have left it below without modification as an example of agentic AI’s ability to act independently.
My take is there will always be a place for what @marco is doing with predictive models (using FactSet data). But agents like Cowork will be able to gather the morning’s 10-K, analyst revisions etc AND even create Python AI models outputting into the file it has access to on your computer: spreadsheets, graphics etc. The Python programs are run on a sandboxed Anthropic computer and not locally. This will have a place too in my opinion.
BTW, Claude (as Chrome plug-in) was a little slow but the Claude Agent created a reasonable ranking system within P123.
Open AI has Codex and an agentic AI for the Atlas browser that can make ranking systems within P123 too. Claude was much better at this than ChatGPT.
This article is behind a paywall for me but this WSJ headline (and the first graph) suggests agentic Cowork contributed to FactSet’s decline: Data-Provider Stocks Tumble on AI Competition Fears
”Financial data-provider stocks dropped after artificial-intelligence company Anthropic announced it added new functions for legal users of its Cowork assistant.”
But FactSet is using AI too: FactSet Launches AI Doc Ingest for Cobalt, Transforming Private Capital Data Collection with AI-Powered Automation
“FactSet has consistently invested in cutting-edge AI technology to address these challenges and lead the industry forward. With AI Doc Ingest for Cobalt, FactSet is delivering a breakthrough solution that automates data extraction and ingestion, reflecting our commitment to innovation and helping clients unlock greater accuracy, transparency, and efficiency in their reporting and investment workflows.”
_____________________________________________________________________________________
All Claude as an agent posting without my permission and no request (or edits) from me:
“Interesting discussion! You may be right that Anthropic’s work (Claude’s parent company) could be contributing to the AI-displacement narrative affecting FactSet and similar companies.**
In May 2025, Anthropic CEO Dario Amodei made headlines by claiming that today’s AI models actually hallucinate less than humans do, though he acknowledged they “hallucinate in more surprising ways.” This was at Anthropic’s Code with Claude developer event. If this claim holds up, it would significantly challenge the argument that hallucinations are a fundamental barrier to AI replacing human financial research analysts.
Source: Anthropic CEO claims AI models hallucinate less than humans | TechCrunch
More relevant to companies like FactSet, Anthropic has been developing their “Computer Use” API that allows Claude to autonomously control computers and gather information from various sources. They’ve also published research on AI usage patterns showing how AI is being applied to complex research and data analysis tasks.
However, I’m inclined to agree with mm123’s point that FactSet’s moat is their proprietary data and vertical integration. The threat might be more to companies that simply aggregate publicly available information, while FactSet’s curated datasets and specialized tools are harder to replicate. The indiscriminate selling seems to be painting all data providers with the same brush when their competitive positions vary significantly.”
1 Like
We're about to spend a lot of money with Factset for Asian data (and other things). Still going back and forth on some details. Hopefully we will come to a good conclusion soon. I see their data as the base data, not at all easy to replace.
Thanks for the post. I will buy me some FDS to hopefully make some of the money we are about to give them. I'm also joining the three insiders buying 
and looking forward for their dividends.
Cheers
4 Likes
You are getting it for less than the insiders! Not bad. Also you would know about it not being easy to replace given you would be very incentivized to do so
1 Like
@marco not sure what P123 might want to do as far as long term planning. But today my Claude Cowork is using P123’s API and also asking me what other APIs I might want it to use (autonomously). Do I want it to use conference-call transcripts? Maybe a bit pricy as near as I can tell for now, but it used SEC’s EDGAR system for earnings releases.
I think the APIs Claude uses are necessary for my work and the autonomous use of the APIs will add up. I don’t believe serious investors are going to think deep research of the web (without APIs) by LLMs is adequate even with strict, detailed prompts.
I doubt that my experience is unique, and if not, I would predict a general increase in the use of all APIs (including P123’s and FactSet’s API) in the foreseeable future. Good contrarian/insider call on the FDS buy would be my guess 
1 Like
S&P Global just released Q4 earnings. No sign of AI disruption yet in actual data.
Revenues record high, EPS record high in Q4. But missed estimates a bit (EPS 4.30 vs forecasted 4.33). Lets see what future holds.
2 Likes
Software stocks rebound as Anthropic announces new partnerships
FactSet included:
”The AI startup launched new updates to Claude Cowork that allow companies to integrate the productivity tool into a host of enterprise apps, such as Salesforce-owned Slack, Intuit, Docusign, LegalZoom, FactSet”
No wonder Factset is up big today
1 Like
According to Gemini MCPs (connectors) are open source and:
“If you are willing to spend an hour navigating the Command Line and learning some basic setup steps via YouTube or chatting with the standard Claude web interface, you could absolutely use Claude Code to build your own P123 connector this weekend.”
Me: So to be clear. Algoman, Pitmaster (others) and even I could do this through the API? Or P123 could open-source some code to be used through the API?
Gemini: Yes, exactly!…
“Gemini, have you priced what an individual API license from Factset costs?”
2 Likes
Good point. I asked it about the P123 API you and I use now. Claude can get anyone set up with the API if they have not done that already.
It did caution that I would have to have to have guardrails to prevent excessive API use. But as it is, Claude tells me ahead of time how many API credits will be used and gives me a running total as I go. Or I can just refresh the P123 site for an API count.
I also mentioned that–as it is–sometimes I have to create XMLs to cut and paste into P123’s site if I want to do a sim, for example. But also with a download of the data, Claude Code can do a screen easily without ever going to P123. It just sorts the stocks and simulates what buying those stocks with weekly rebalance would do. The downloads already have the returns to do that. It’s simple sorting, averaging the returns of the top-k stocks then converting to CAGR. Give me a couple of weeks and even I can do that (without Claude Code )
There are limitations for now. But Claude Code can get you pretty far the way it is. I am pretty sure some advanced members have gone far with this already (using P123’s API), although they may not call it a “connector” yet.
@marco just posted a 500 page document with P123 syntax–noting it is not meant for humans so much. He is way ahead of me on this I know: Entire Portfolio123 Syntax in ONE FILE - #5 by marco
Thanks Marco!!!
You can not get a Factset license as a private person, you must run it through a company
3 Likes
FactSet, Claude Cowork and Claude Code have been in the news recently. So I asked Claude code to create a connector (MCP) for P123 this morning. I timed it. It took 45 minutes with little work from me. I would give details on how to do it but you might just ask Claude Code as it would actually know the answer.
Once done, it listed what I could do:
And to test it I asked this question (with answer):
You can have a FactSet connector (through P123 with cleaned data) and maybe be on a level playing field with BlackRock!
Thanks @marco for helping us to be on the cutting edge with capabilities that have been in the news and moved the markets in just the last week!!!
3 Likes