A bit more on how AI Search & Screening works.
US-listed companies file >40,000 documents a quarter, and a single 10-Q can run 40,000 words. No chat model can read a full quarter of the market, so when you ask one "who raised capex guidance" it searches the web and returns the few companies that made headlines.
We index every filing, press release and transcript for every US-listed company as it comes out. Each passage is scored for materiality before the model sees it, so a guidance change buried in a 10-Q ranks alongside one announced in a press release. The output is a row per company, and every figure is checked against its source passage before it's shown.
In Q2, 19 companies raised capex guidance. Coeur's raise was in both the call and the 10-Q. Murphy's was only in the 10-Q. Both showed up, each linked to the line it came from.
When I was on the sell side, a screen like this was days of reading. Happy to walk anyone through it.
AI Search & Screening is live today 🎉
Ask “who raised capex guidance” and the Co-Analyst will find every relevant filing, call transcript and press release, and come back with a list that is both informative and complete. . Every answer is sourced, every quote is real and ever number links to document where it was found.
We built it because the information you need to make investing decisions is spread across 30,000 documents. Keyword search will find the phrase but miss the point, and a chat model can't "pay attention" to a full quarter of filings at once, so it goes to the web and hands you the companies that were already in the news.
Our advanced retrieval harness helps the Co-Analyst find and connect relevant evidence across millions of filings, releases and transcripts.
When we search, each potentially relevant piece of evidence is scored and ranked based on materiality so you see the results that matter first.
The Co-Analyst supports normalized and structured search. Search for guidance and get results with the bottom and top of the range in separate columns, and the commentary in a third. Alternatively, retrieve results on consumer behaviour with a tag for tone.
Link below