Module 918 min read · AI in Finance

Building a Personal AI Research Workflow

You now have the pieces — the tools, the techniques, the verification discipline, the compliance awareness. This module assembles them into a repeatable personal workflow: a system for researching a company, analyzing it, and forming a view, using AI for maximum leverage while keeping the judgment and verification where they belong. This is the module you'll come back to.

Why a workflow beats ad-hoc use

Most people use AI in finance reactively — a question here, a summary there. The people who get dramatically more value have a system: a consistent sequence that combines tools for their strengths and builds in verification at the right points. A workflow turns scattered AI use into a reliable process you trust.

The complete company-analysis workflow

Here's an end-to-end workflow for going from "I want to understand this company" to "I have a defensible view." Adapt the depth to your needs and time.

Phase 1 — Orient (Perplexity, ~15 min)

Live-search overview: business model, segments, recent performance, recent news, key competitors — all with citations. Progressive narrowing to drill into the most recent quarter and any major recent developments. Output: current, cited context.

Phase 2 — Gather primary documents (~10 min)

Pull the latest 10-K, most recent 10-Q, and last earnings transcript from EDGAR and the IR page. These are your ground truth. Everything important will be verified against these.

Phase 3 — Deep document analysis (Claude/Gemini, ~30 min)

Upload the filings. Targeted extraction of risks, revenue concentration, unusual items. Risk-factor analysis. MD&A-versus-numbers reality check. Earnings-call tone analysis. Output: the real story, beyond the headline numbers.

Phase 4 — Quantitative analysis (ChatGPT ADA / spreadsheet, ~20 min)

Extract the actual financials, compute key ratios and trends with executed code, compare to peers. Real numbers, real computation. Output: verified quantitative picture.

Phase 5 — Synthesis & thesis (Claude, ~25 min)

Bring verified facts and analysis to Claude. Structure the thesis: what's the bull case, the bear case, the key risks, what would have to be true. Use Claude's willingness to push back to stress-test your view. Output: a defensible, written thesis.

Phase 6 — Verify & finalize (~15 min)

Go back through every critical fact and figure in your final output and confirm it against a primary source. Fix anything unverified. Output: work you can stand behind.

~2 hrs

For a thorough AI-assisted company analysisVersus the day or more the same depth would take manually. The verification phase is non-negotiable — it's the 15 minutes that makes the other 105 trustworthy.

The multi-tool principle in practice

Notice how the workflow deliberately moves between tools: Perplexity for current cited facts, Claude/Gemini for document depth, ChatGPT for real computation, Claude for synthesis. No single tool does it all well. The professional advantage comes from routing each task to the tool that's genuinely best at it.

Building your own variations

This full workflow is for deep analysis. Build lighter versions for different needs: a 30-minute "quick take" (Phases 1, 3, 5 only), a "monitoring" workflow for companies you already know (Phase 1 + a focused Phase 3 on what changed), and a "comparison" workflow that runs Phases 1-4 across several companies in parallel. The phases are modular — assemble them to fit the task.

Capturing and reusing your work

Use persistent workspaces to compound your effort. A Claude Project or a Perplexity Space dedicated to a company accumulates context over time — every session builds on the last. For companies you follow continuously, this turns one-off analyses into a living knowledge base that gets more valuable the longer you maintain it.

The workflow mindset

A good workflow does two things at once: it makes you faster, and it makes you more rigorous. The phases force you through verification and stress-testing you might skip if you were just chatting with an AI. The leverage isn't only speed — it's that the system catches the things careless AI use misses. Build the system once, then trust the process.

The final module steps back to the big picture: the genuine limits of AI in finance, the ethical lines, and why the human-in-the-loop isn't a temporary limitation but a permanent feature of doing this responsibly.