The AI Equity Research Stack for 2026: Data, Model, Note

This post contains affiliate links. If you sign up through one, we may earn a commission at no extra cost to you. How we handle this.

A year ago, "AI for equity research" meant pasting a 10-K into a chat window and hoping the summary was right. The setup analysts are settling on now has more structure, and it splits into three layers. A data layer supplies numbers you can trace to a filing. A model layer, where Claude Code does the analysis, works from those numbers instead of from memory. A note layer of Skills turns the analysis into something a PM will actually read.

Each layer fails differently when it's missing. Without the data layer, Claude produces confident numbers that are wrong. Without the model layer, you're back to copying figures between browser tabs. Without the note layer, you have good analysis that nobody reads. Here's what goes in each one, what it costs, and where you can save money.

The Stack at a Glance

Prices were checked in September 2026 and change often. Confirm on each vendor's site before you buy.

Layer 1: Data You Can Trace

Most AI research mistakes start in this layer. A language model asked about a company's gross margin will give you a number whether or not it knows one, so the only real defense is to make sure Claude retrieves figures instead of recalling them. The three tools here split the job three ways, and each does something the other two don't.

Fiscal.ai: the numbers Claude reads directly

Fiscal.ai is the only tool in this stack built to deliver data straight into Claude. It's an official Claude connector, running a hosted MCP server that Claude Code queries for standardized and as-reported financials, ratios, company KPIs, revenue segments, ownership data, and earnings call transcripts. Every figure links back to its source filing, and that link is what lets you defend a number in front of a PM. Paid plans cover 13,000+ public companies across the US, Canada, ADRs, and UK & Europe. The free tier connects with no credit card and covers 100 large caps at 250 data calls a day. That's enough to prove out the workflow before you pay. Our Fiscal.ai MCP walkthrough covers setup in one command.

Koyfin: the universe and the filter

Koyfin is where you decide what to look at. Its screener spans 100,000+ global securities with 5,900+ criteria. It also covers ETFs, fixed income, FX, and macro, and its charting is excellent. It has no MCP server or API, and its data license blocks exporting fundamentals, so its role in an AI stack is narrower but still valuable. You run the screen in Koyfin, export the ticker list, and hand that list to Claude to enrich with Fiscal.ai data. Our Koyfin + Claude screening guide covers that handoff, and the Koyfin vs Fiscal.ai comparison covers which one to choose if you can only afford one.

Seeking Alpha: the other side of the trade

Seeking Alpha fills a gap the first two leave open: what other investors think. Premium includes Quant Ratings, which grade a stock against its own sector across 100+ underlying metrics. It also includes earnings call transcripts and a large body of contributor analysis. For an analyst the contributor bear cases are the most useful part, because they come from people with a reason to argue against your thesis. There's no Claude connector, so this layer is read, not piped in. You paste the arguments that matter into Claude and ask it to test your thesis against them.

That prompt is the stack at its best. Seeking Alpha supplies the argument, Fiscal.ai supplies the evidence, and Claude does the cross-examination you'd otherwise put off until after the stock had already moved against you.

Layer 2: Claude Code as the Model

The model layer is where the stack beats a chat window. Claude Code works inside a project folder, so your models, theses, and past notes stay there as files it can read and update. Every earnings season, the work starts from last quarter's state instead of a blank page.

If your firm already pays for institutional data, it slots into the same layer. The LSEG Equity Research Snapshot Skill pulls IBES consensus estimates, fundamentals, price history, and macro context through LSEG's data tools. That's the institutional-grade alternative to the data layer above, and the analysis and note layers stay the same either way.

Layer 3: The Note

Three Skills cover the research cycle from before a print to a full initiation, and each one does better with real data underneath it.

Before earnings: the preview

The Earnings Preview Skill builds a one-page setup covering consensus estimates, bull, base, and bear scenarios with their price implications, the five catalysts to watch, and the options-implied move. Pair it with the prior quarter's transcript from Fiscal.ai and the scenarios rest on what management actually said, not a summary of it.

Every quarter: the thesis

A thesis drifts quietly until it breaks all at once. The Thesis Tracker Skill keeps each position's thesis as a scorecard, marking every pillar on track, behind, or ahead. It logs each new data point with its impact and updates your conviction level. Fed Fiscal.ai actuals each quarter and the latest bear arguments from Seeking Alpha, it tells you when you're holding past the thesis's sell-by date.

New coverage: the initiation

The Initiating Coverage Skill is the heavy lift. It runs a five-task workflow covering company research, the financial model, valuation, charts, and report assembly, and it ends in a 30–50 page report with a DCF, comparable companies, and a recommendation. This is where the data layer earns its cost. A 40-page report built on recalled numbers fails the first time a reader checks one figure, and one built on filing-linked data holds up.

What It Costs, and Where to Save

At the time of writing, the full paid stack runs Koyfin Plus at $39 a month, Fiscal.ai Pro at $39 a month billed annually, and Seeking Alpha Premium at about $299 a year (roughly $25 a month). That's about $100 a month plus your Claude plan, a fraction of a single terminal seat. Every Skill on this site is free.

• Most important: Fiscal.ai. It's the layer that stops Claude from making numbers up, and it's the one you'll feel most if you drop it.
• Skip if your coverage is fixed: Koyfin. If you already know your names, you don't need a 100,000-security screener. Add it when you start generating new ideas.
• Optional: Seeking Alpha. It sharpens the thesis and supplies bear cases, but the model and note layers work without it.
• Free starting point: Fiscal.ai's free tier, Koyfin's free plan, and the Skills cost nothing. Use them to prove the workflow on a large-cap name before paying for anything.

Start With One Name

Don't build the whole stack in one go. Pick one stock you already cover, connect Fiscal.ai's free tier, and run the Earnings Preview Skill ahead of its next report. If the preview is better than the one you'd have written by hand, add the Thesis Tracker next. Add the screening and sentiment layers once the core of the stack is earning its keep.