The work you do
Pitch a name to the PM. Build the model. Read the next four quarters of transcripts. Sit on expert calls. Stress-test the thesis. Update the IC memo every quarter. Track the position. Most of the value is in the judgement; most of the time goes into the construction.
Where Smalt AI fits
Smalt AI is the construction layer. You stay on the judgement; the AI handles the building. Three-statement model with formulas live, sensitivity tables wired, scenario INDEX framework set up. Sourced research with paragraph-level citations to filings, transcripts, and analyst notes. Bull and bear cases built side-by-side, with the open questions called out as a list.
What you'd actually ask it
- "Build a DCF for [name], 5-year horizon, three scenarios." Get back a 14-sheet workbook — assumptions through sensitivity, mid-year discounting, three 5×5 tables. Open it in Excel; change a driver; everything recalculates because formulas are live.
- "Summarise the last four earnings calls — what changed?" A diff across quarters with the exact paragraphs that shifted, called out and cited. Faster than re-reading the transcripts; structured for memo prep.
- "Stress-test my thesis on [name]." A counter-argument across filings, news, and analyst notes — with quotes attached. Not a sycophantic re-summary; a real bear case.
- "Pull the contact details for the last 10 buy-side firms that owned [name]." A contact sheet with sources and emails drafted, queued for your review (not auto-sent).
Where it's particularly differentiated
Long-context filings analysis. Smalt AI runs Claude with up to 1M-token context. That means you can drop a full 10-K plus four quarters of transcripts plus the last initiating analyst notes into one conversation. Long documents stop getting chunked; cross-reference works the way it should.
Real Excel models, not screenshots. Other AI tools hand you a markdown table. Smalt AI hands you back an .xlsx file with =NPV, =IRR, =INDEX/MATCH live. Change WACC; the whole valuation recomputes. That's the difference between a model and a description of one.
Citations, not vibes. Every claim in research output has a paragraph-level source. You can verify before you bring it to the PM.
What it doesn't replace
The judgement. The cross-asset insight. The expert call you'll sit on. The pattern recognition that comes from running the strategy for years. Smalt AI compresses the construction time so you spend more time on what only you can do.
Related
Financial modeling · Investment research · Due diligence · DCF · Sensitivity analysis