Financial Modeling
AI native modeling inside the spreadsheet — build, audit, and stress test a model in hours instead of days.
Modeling is the last manual chokepoint in diligence. The deck is read, the calls are done, the memo is drafted and an analyst is still rebuilding a model from PDFs at midnight wrestling with the sparse startup data she has. The right AI native modeling stack collapses that work from days to hours and moves the analyst's job from data entry to true judgment.
The shift
The bottleneck has never been Excel itself. It's the round trip: pull a number out of a filing, type it in, link it forward, debug a broken reference, repeat a thousand times. AI native modeling tools live inside the spreadsheet, read the workbook structure, and do that round trip for you. The analyst becomes a reviewer and a thesis author, not a typist. Every model gets built faster, audited harder, and stress tested more before it ever reaches a partner.
The two real options today
Two patterns exist. The first is a general purpose AI assistant inside Excel (Claude for Excel or ChatGPT for Excel) that sits in the side pane and acts as a copilot and an agent in one. The second is a finance specific agentic add-in like Shortcut that treats modeling as a workflow to automate end to end. The honest read in 2026: Claude for Excel has effectively dominated this space and beaten the finance specific agents at their own game — most firms will be best served standardizing on Claude or ChatGPT for Excel, though it's worth running both against a live model to see which fits the team's habits.
Anthropic's native Excel add-in
Sits in the side pane of Excel (web, Windows, Mac, iPad) and reasons over the full workbook — multi-tab navigation, cell-level citations, formula-dependency-aware edits.
- Best for
- Auditing existing models, debugging #REF!/#DIV/0! chains, updating assumptions without breaking downstream formulas, asking questions about what a model actually does.
- Gotchas
- Requires a Claude Pro/Max/Team/Enterprise seat. Workbook context is sent to Anthropic — confirm posture with the firm before using on sensitive deal data.
OpenAI's Excel & Sheets add-in
Build full spreadsheets from a plain-language brief, query across tabs, and edit formulas in place. Available in Excel and Google Sheets, which matters for firms that aren't 100% Microsoft.
- Best for
- Cold-starting a model from a prompt (DCF, cap table, scenario sheet), turning a messy export into a clean tab, conversational analysis of an inbound founder model.
- Gotchas
- Same data-sharing posture as Claude — workbook context goes to OpenAI. Sheets coverage is a plus, but the Sheets add-in lags Excel in fidelity.
Agentic Excel add-in (Fundamental Research Labs)
Built for finance professionals specifically. Executes multi-step workflows autonomously: ingest a 10-K or pitch deck PDF, build the LBO/DCF/pro-forma from it, format it, and hand it back. Less chat, more “do the work.”
- Best for
- Building net-new models from filings, extracting structured financials out of a founder PDF, repetitive transformations across many companies (comp sets, returns analyses).
- Gotchas
- Separate paid add-in. Strongest on standard finance archetypes (LBO, DCF, three-statement) — gets less reliable the further you drift from those patterns. Always audit before relying on the output.
How to choose
For most firms, the recommended posture is to standardize on Claude for Excel as the default — it has matured into the strongest all-around tool for both copilot work (auditing, debugging, asking questions about inherited models, updating assumptions across linked sheets) and the agentic work that finance-specific add-ins used to own (building net-new models from a PDF, running a comp set, cold-starting an LBO). ChatGPT for Excel is the close alternative and worth testing in parallel, especially for teams already deep in the OpenAI stack or working across Google Sheets.
Pick by the frontier model the firm already standardizes on. If the team uses Claude for memo generation and ad-hoc analysis, keep them in Claude inside Excel — context and prompting habits carry over. Same logic for ChatGPT. Either way, give it a two-week bake-off on real diligence models before locking in.
Posture and guardrails
Three rules that protect against the obvious failure modes. Always audit the output. AI-built models hallucinate formulas and quietly miscite filings; trust nothing until an analyst has walked the dependency tree. Treat workbook context as outbound data. Anything sent to Claude, ChatGPT, or Shortcut leaves the firm — clear that posture before pointing the tools at sensitive founder financials or LP facing models. Keep the thesis with the human. The model is an artifact; the investment view is a judgment. The tools speed up the artifact so the analyst can spend more time on the view.
Why this matters
A firm that ships a clean sensitivity tested model in 48 hours underwrites with more conviction than one still typing in inputs the morning of IC. Done right, this module turns modeling from a scheduling constraint into a competitive edge and frees the analysts to do the work that actually compounds: spreading more deals, building more comps, and pressure testing more theses.