AI automates earnings model extraction

This reveals that Canalyst’s moat was never just the spreadsheet, it was the speed and labor system behind turning fresh company disclosures into usable models before the market fully digested them. Canalyst sold near real time updates to earnings models, and that required a large analyst workflow built around primary sources, manual checks, and very fast turnaround. Tegus saw AI as a way to replace the most repetitive step, pulling numbers from calls and transcripts into models fast enough for investors to act on them.

• Canalyst’s product was built around human verified model maintenance. The company describes 4,000 plus models updated in near real time for earnings and major events, built by a 100 plus person research team with double entry and review. That operating model created trust, but it also made every fast update expensive.
  2 canalyst 3 canalyst 4 canalyst
• Tegus bought Canalyst because expert calls, filings, and models fit into one investor workflow. A fund analyst could read an interview transcript, jump into SEC filings in BamSEC, then test the implication in a Canalyst model. The transcript library generated the research question, and the model turned it into an investable view.
  1 sacra 5 sacra 6 sacra
• The hard part for AI was not summarizing a call, it was extracting the exact number that belongs in the correct model line fast enough to matter. A slightly wrong summary is tolerable when it links back to the source. A slightly wrong revenue driver can break an earnings model and damage trust with paying hedge fund users.
  1 sacra 4 canalyst

This category is heading toward thinner human workflows and deeper product embedding. As model extraction gets automated, the winning research platforms will be the ones that can push trusted numbers, transcript evidence, and scenario tools directly into investor workflows through search, APIs, and live model updates, without losing the audit trail that made the data valuable in the first place.