The Analyst Who Has Read Everything

Interrogate a Name the Way You'd Pressure-Test a Thesis

When a thesis is not yet resolved, you ask questions. Does the margin story hold up against the last six quarters of operating leverage. Is management's tone on capital allocation actually consistent with the buyback commentary from the prior call. Are competitors saying anything that would break the bull case.

AI Analyst Chat answers those questions directly from the company's filings, earnings calls, and press releases. Every answer cites the exact passage it came from, so you validate the claim in one click rather than rebuilding it yourself.

A thesis pressure-test that used to take an afternoon now takes one conversation. The work that fills the afternoon (re-reading the last four 10-Qs, scanning the call transcripts, lining up the quotes) runs in the background of a single thread.

Because primary sources are ingested within seconds of publication, answers reflect what the company has disclosed as of right now, not last night's snapshot.

Every Answer, Cited to the Filing

Chat draws exclusively from validated primary sources. No web summaries, no secondary commentary, no hallucinated numbers. Each answer ships with a direct link to the underlying passage: a section of a 10-K, a timestamp in an earnings call transcript, a paragraph in a press release.

This is the line that separates AI Analyst Chat from a general-purpose assistant. The questions sound similar. The output is built differently. An IC memo, a sell-side note, or a client write-up needs claims that are auditable on the way out the door, and that requires every number and every quote to be traceable back to its source.

Build a Thesis Across Multiple Turns

A thesis is rarely answered in one prompt. You start with a question, then follow the thread: what did they say on the prior call, how does that compare to peers, where is the operating leverage actually coming from. Each conversation preserves full context and citations, so follow-ups build on prior answers instead of starting over.

Behind the scenes, Marvin compares every new question and every new filing against the Common Ground it maintains for each company. Common Ground is a structured model of everything the company has publicly asserted, with each fact tagged to source and date. Asking how management commentary has shifted, or what is different in the latest 10-Q, becomes a query over that model rather than a fresh read of the document.

Examples of questions analysts run:

• "If management's margin expansion thesis relies on fixed-cost leverage, how consistent is that with their historical operating leverage?"
• "I'm building a bull case on market share gains. What evidence from competitor filings would contradict this view?"
• "Compare revenue growth across MSFT's business segments for the last 6 quarters."
• "How has management tone on capital allocation changed between the last two earnings calls?"

Pull Peers, Suppliers, and Competitors Into the Same Thread

Mention a different company mid-conversation and Marvin spins up a specialist sub-agent with full access to that name's filings, earnings calls, and press releases. The main agent waits, runs follow-ups across multiple turns if the first answer is thin, and integrates the result back into the thread.

You do not always need to name the peers. Ask about "this company's competitors", "supplier margins", or "who does the most business with them" and Marvin uses its company relationships map, grounded in executive commentary and filing references, to identify the relevant names and query each one in parallel. A peer comparison comes back in a single response. Read more in our cross-company queries overview.

For qualitative context a filing cannot answer (a product launch from last week, a regulatory development, a recent news cycle) Chat can also reach out to the web. Web search is scoped strictly to qualitative news. Financial numbers, guidance, and price targets stay sourced from primary documents.

Charts, Tables, and Financial Metrics, Built Into the Answer

Ask a data question and the answer comes back as an interactive chart you can hover, zoom, and download. Bar charts for peer comparisons, line charts for trends, structured tables when the data reads better as rows. Charts export as images or PDF for client decks and research notes (interactive charts and tables overview).

The same financial statements and derived metrics that sit on every company page (income statement, balance sheet, cash flow with five years of quarterly, annual, and LTM history, plus FCF, ROIC, valuation multiples, earnings and revenue surprises against consensus) are queryable directly from chat. A lock icon distinguishes as-reported figures from platform-computed metrics, and Chat carries the same distinction into answers so you know which number came from the filing and which was calculated on top of it.

In practice that means a question like "how has MSFT's free cash flow trended over the last eight quarters" returns the data, the short interpretation, and an interactive chart in one response, with links back to the filings the inputs were drawn from.

Turn the Thread Into a Deliverable

Conversations export as formatted outputs: investment memos, KPI tracking dashboards, executive summaries, plain text. Source attribution stays intact through the export, so the deliverable is audit-ready out of the box. Drop it into an internal research database, paste it into a model, or send it to a client without rebuilding the citations.

When a thread captures something you want to keep current, Deep Research Agents take the same conversation and run it on every new filing automatically. Chat handles the thinking. Agents handle the maintenance.