Claude just released an Excel plugin. And it completely changes the game for equities investors.

Here’s why.

Claude is better than most analysts I've had....

Earnings season is brutal. Every quarter you get slammed with transcripts, investor presentations, 8-Ks, and financials across your entire coverage universe. All at once.

The challenge isn’t accessing the information. It’s reconciling what management actually said against the models you’re using to make decisions.

If you rely on sell-side research, you know the problem. Analysts cover way too many names. Human error creeps in. Assumptions get stale. Price targets drift from what the company is actually guiding.

How do you audit dozens of model assumptions against raw earnings data? How do you know which lines are overstated or understated? How do you catch when the analyst missed something material from the call?

Previously, this was hours of manual work. Cross-referencing transcripts against spreadsheet cells. Line by line.

Now you can do it in minutes.

Claude for Excel lets you interrogate financial models directly in the spreadsheet. Feed it earnings documents. Ask it to flag assumptions that don’t match guidance. Get a clear audit of where projections diverge from reality.

I’m going to show you exactly how this works.

Here’s the plan:

1 prompt to get docs from earnings season
Extract all material KPIs and statements from earnings docs
Prompt Claude with RAG to audit the model against these inputs
Flag lines that are too aggressive or too conservative vs management guidance

For this example, I’m running Meta $META 4Q25 against a Goldman Sachs model. But you can apply this to any name you cover.

This is the workflow hedge funds pay analysts to do manually. The technology is finally here.

The Prompt

We're going to use prompt chaining techniques across ChatGPT and Claude for Excel plugin to do this task.

Prompt 1: Use ChatGPT to fetch the raw documents from 4Q25 earnings (Link)

Find me Meta's investor relations website to grab the links and downloads for the latest quarter. I am specifically looking for (1) earnings transcript (2) financials (3) investor presentation - please include other files that may be relevant for updating my model.

Prompt 2: Use ChatGPT to extract material statements and KPIs from the earnings season

Prompt here: Link

Result here: Link

Note: Make sure you copy the response here into a .txt or .docx file so we can RAG this into the next step.

Prompt 3: Open your sell-side model you want to audit. Then open the Claude for Excel plugin and (1) Upload the .txt file from the previous step that extracts the material numbers and statements (2) Paste the below prompt.

Prompt 3: model_audit_prompt_dave_wang.pdf

Paste in the prompt 3 here + RAG our materiality document!

The Result

Full output here: Meta_GS_Model_4Q25.xlsx

Note: Check the last 4 tabs in the model which is what Claude produced in my prompt.

Claude created four new tabs directly inside the Excel workbook. Structured audit output sitting right next to the original model.

Here’s what each tab does:

MGMT_FACTBASE: Every material KPI, guidance range, and qualitative statement from earnings docs. Organized and sourced.
MODEL_ASSUMPTION_MAP: A map of the model’s forward assumptions traced to specific cells. You can’t audit what you can’t locate.
ASSUMPTION_AUDIT: Line-by-line comparison of model assumptions vs management anchors. Each one rated: too aggressive, reasonable, or not aggressive enough.
PRIORITY_FLAGS: The top 10 assumption risks ranked by materiality. The stuff that actually moves the needle on valuation.

What It Found:

The audit flagged assumptions where Goldman diverged from management guidance in both directions.

Some lines were too aggressive (Europe revenue growth at 40% despite EU regulatory headwinds). Some were not aggressive enough (stock-based comp at 5% growth when management explicitly said comp will outpace headcount).

It also surfaced qualitative signals that weren’t fully reflected in the model. Capex timing uncertainty. Reality Labs investment mix shifting. Operating margin compression trajectory.

The kind of nuance that’s easy to miss when you’re quickly auditing a model!

Why This Matters:

The hard part of earnings season isn’t just accessing information. It’s translating what management said into what it means for the model.

That translation step is where alpha lives. And it’s where most investors cut corners because they’re out of time.

Previously, this required hours of manual cross-referencing. Now you can do it in minutes.

But the bigger unlock isn’t just speed. It’s having a sparring partner inside your spreadsheet.

You can feed in a qualitative view (“I think Europe revenue is at risk from regulation”) and ask Claude to show you exactly which cells that flows through and how it impacts EPS.

You can take any piece of news, any management signal, any thesis you’re developing, and pressure-test how it should show up in the model.

This closes the gap between reading a transcript / any news and actually updating your price target.

The Bigger Picture:

This workflow isn’t limited to auditing sell-side models. You can run it on:

Your own models after any earnings update
Competitor coverage to see where consensus might be wrong
Historical quarters to backtest how guidance translated to actuals
Any scenario where qualitative information needs to flow into quantitative projections

The technology is finally here.

We’ll be going deeper on Claude for Excel workflows in this newsletter. If there’s a specific use case you want me to cover, hit reply.

Personal:

Something that's been a total game changer for my personal productivity this year has been integrating voice to text on both desktop and mobile.

Sharing here if helpful!

Here's my setup:

Mobile: Google Pixel has voice to text embedded natively
Desktop: Wispr Flow has a desktop app

I've found both of these to have high accuracy and I'm unconstrained by speed of typing (but rather speed of thinking).

What you can do to take this to the next level is to have custom Claude Projects refactor raw thoughts into templated outputs / certain copywriting styles .... and I have personal Telegram channels that sync between desktop and mobile. You can even connect Claude Code to these Telegram channels to automate even more.

I'm trying out including a personal section every week for non-finance discoveries and random thoughts throughout the week. Hit the reply button if you want this to be a recurring section in the newsletter + any requested topics / categories!

Get one high impact AI prompt every week