of large spreadsheets contain material errors.
Panko · 2008 reviewWhy models belong in code
Code was always the better way to build financial models. The learning curve made it impossible. Until now.
AI now translates fluently between plain English and Python, in both directions. The builder's learning curve is gone. The reviewer's learning curve is gone. The foundation your model is built on can finally be code.
/ 01 · the unspoken truth
Every senior modeler has had the same 2am thought.
Forty tabs. A sign error you missed before the board meeting. A formula quietly depending on a hardcoded number four sheets away. We've all thought there must be a better way. That just changed.
JPMorgan "London Whale," traced to a copy-paste in a risk model.
Senate PSI reportTransAlta — a bad sort, no version history, real-money trades.
10-K disclosureReinhart-Rogoff. A spreadsheet range error shaped global austerity policy.
Herndon et al./ 02 · why we never crossed over
Two learning curves. Neither one resolved.
Code was always better on the things that matter for serious modeling. But getting there meant both sides of the work — the person building and the person reviewing — had to read code. Neither did. Stalemate.
The things spreadsheets can't fake.
- Automatic checks that run on every change.
- A real history that shows exactly what changed, and when.
- Reuse — last quarter's model snaps into this one.
- Reproducibility — same inputs, same outputs, forever.
- Thousands of what-if scenarios in a second.
- Models that build on other models.
Both sides had to read code. Neither could.
- Analysts don't write Python — that's a year of part-time learning, minimum.
- Reviewers — CFOs, partners, IC members — can't audit code.
- A model no senior person can audit is a model no senior person approves.
- So the work stayed where the reviewer could read it: the grid.
/ 03 · the compromise wave
Anaplan, Pigment, Causal — the best answer possible before AI.
The modern planning tools tried to solve the same problem. They built their own formula languages because they couldn't expose real code. The result inherited the spreadsheet's weaknesses and added a new language to learn.
What matters
Excel
Proprietary tool
Code + AI
Checking for errors
None built in.
Limited.
Automatic checks on every change.
Version history
Filenames (_v4_FINAL).
Snapshots only.
Full history. See every change.
What-if analysis
Slow or impossible.
Limited by the tool.
Thousands of scenarios in seconds.
Reviewer access
Familiar grid.
A new grid to learn.
Plain English on hover.
Lock-in
File-format lock-in.
Heavy.
Open, portable, yours.
/ 04 · what changed
AI is the bridge. Both learning curves vanish at once.
The builder describes the model in English. The AI writes clean, checked Python. The reviewer hovers any number and the AI explains the formula, where it came from, and the check that backs it. Both sides work in their own language.
English in.
Analysts describe the model, the scenario, or the variance question without learning Python first.
Checked, versioned Python.
Bridge Town stores and runs the code, keeps the history, and makes each change reviewable.
Plain English out.
Reviewers get explanations, assumptions, checks, and links back to the exact model version.
/ 05 · the objection
"But Excel has AI too." So why move?
Microsoft and Anthropic are putting AI inside Excel. That makes Excel a better Excel — not a better discipline. The things code gives you come from what the model is, not from the tool you type into.
A real change history
A code file shows you exactly what changed, line by line, across every version. A spreadsheet doesn't.
Reusable parts
Last year's deal model is one line away from this year's. Spreadsheets can't borrow from each other cleanly.
Checks itself automatically
Automatic checks run on every change and catch mistakes before they reach the board deck.
Built for AI assistants
AI works far better with code than with cell references. The foundation that supports AI wins.
Nothing hidden
Code is plain text. Spreadsheets hide logic behind values. When an auditor asks, you can show them.
Yours to keep
Your model is plain text in a language millions of people use. No vendor stands between you and your work.
/ 06 · what this unlocks
Things that were impossible in Excel for thirty years.
Each one is something a senior modeler has wanted for a decade and worked around for a decade. They come free the moment the model is code.
Compare two versions
See exactly what changed between the model that went to the board and the one that came back.
Set guardrails on the numbers
Require the cost of capital to stay between 4% and 15%. Catch the slip before the model ships.
Run thousands of scenarios
Real probability analysis on a laptop, in seconds. No add-in, no waiting.
Review the budget like a team
Proposals, approvals, and comments on assumptions. The audit trail builds itself.
Reopen last year, exactly
Pull up the 2024-Q4 version, run it, and get the same numbers. Always.
/ 07 · getting started
The move is a non-event.
You don't have to learn Python. Your reviewers don't either. And your counterparties never see the code at all.
1Describe it in English
Type what the model should do.
Your AI assistant writes the code, runs the checks, and shows you the familiar grid.
→ model.py · 240 lines · 18 checks pass
2Import an Excel model
Upload your existing spreadsheet.
Bridge Town turns it into a clean, checked, versioned model — line by line, with a trail you can audit.
→ 412 cells migrated · 22 checks inferred
3Export back to Excel
Counterparties get a spreadsheet.
The seller, lender, auditor, or lawyer receives an .xlsx that looks exactly like what they expect.
→ deal-model-2025-q3.xlsx · ready
/ 08 · build on the right foundation
The foundation of financial modeling is changing.
Bridge Town is a home for financial models that are code — with AI translating for everyone who shouldn't have to read it.