Financial Models Should Be Engineered

A financial model is the company’s DNA. It shows how one part of the company affects another and how it ends up as revenue, cost, cash flow, and value. It holds the facts, forecasts, and scenarios. Decisions are made on it.

And in almost every company, it lives in spreadsheets.

The Problem

Everyone who builds financial models in Excel knows the problems.

No version control. No collaboration. No tracking of changes - you receive Q3_Budget_v44_FINAL.xlsx - what was changed? Are the parts that were right before still right? Are the links still intact?

The numbers confirm it. 94% of spreadsheets in use contain errors. 47% of CFOs say tracking report versions, correcting errors, and chasing data impairs their ability to participate in strategic decision-making. Building a financial model from scratch takes 1–2 weeks to 6 months.

And then the best modeler on the team leaves... who knows now what =B47*C12*(1+D3) means?

Disconnected and Blind

There is a second problem: financial models are disconnected from data and incapable of advanced analytics.

46% of FP&A time goes to data collection and validation, not analysis. 89% of finance leaders make monthly decisions based on inaccurate or incomplete data. The data exists - in ERPs, warehouses, CRMs. But getting it into Excel is manual. Copy, paste, reformat, reconcile, pray nothing broke.

And even when the data arrives, Excel’s analytical ceiling is low. Monte Carlo simulation? Too advanced. Optimization - what is the optimal debt schedule? Barely possible. No time-series forecasting, no machine learning. The analytical ceiling is real.

The Attempts

The problems are well known, and the industry has tried to solve them.

One path: add engineering and leave Excel. But 89% of finance teams rely on Excel even when they have planning software. The most popular feature in every BI platform? Export to Excel.

Advanced Excel platforms aren’t worth the switch - you trade one set of limitations for almost the same ones.

The industry is pushing toward Python to break through the ceiling. CFA mandated Python starting 2025. Microsoft launched Python-in-Excel. But Python without Excel formulas is not satisfactory.

AI

Most modern AI approaches focus on making work faster inside Excel’s limitations. And AI struggles without semantics - cell addresses carry no meaning, no types, no dependency graph.

In a world where AI makes almost everyone capable of coding, using AI only inside Excel’s limitations looks like the wrong path.

The “as code” Pattern

This has happened before.

When SQL data pipelines became unmanageable, the discipline got a name: Analytics Engineering. The tool - dbt - became the standard. When infrastructure got too complex, it became Infrastructure as Code. Terraform followed.

Now look at the modern data stack. Ingestion - automated. Transformation - engineered. Storage - automated. Semantic layer - defined. Visualization - automated. The entire stack from raw data to dashboards is code-native, versioned, tested, connected.

And then, at the top of the stack, where the most important decisions happen - someone takes the data into Excel, builds a financial model, plays with the numbers, and decides. The decision layer. The most valuable layer. Completely unengineered.

Every other layer got its “as code” moment. Financial modeling hasn’t.

The Debate Is Open

The debate is already public. In early 2026, a16z’s Andrew Chen told 1.3 million people that “anything modeled as a spreadsheet is better modeled in code.” The top reply, from a finance professional: “One calculation error can cost you $1 billion. 95% right is 0% useful. Python does not allow for full verification.” Both are right.

Excel has transparency but no structure. Python has structure but no transparency. Python is the right language. But without formula parity, it’s a dead end.

Python-Excel Parity

What if Python code could generate not just values but real Excel formulas?

You write revenue = price * volume in Python. The system generates =B4*B5 in the Excel cell. Not dead values. Live formulas. The CFO clicks on the cell and sees the formula. The auditor traces the logic. The output contract is preserved.

This is Python-Excel parity - and it changes everything.

# Your model, structured as code
periods    = Variable(3)
start_year = Variable(2026)
t          = Timeline(periods, start_year)

price      = TimeVariable(t, 49)
users      = TimeVariable(t, 100, growth=.15)
revenue    = price * users
cogs       = revenue * 0.30
gross_profit = revenue - cogs

Once parity exists, every problem resolves at once:

• Version control. Models are code. Diffable, branchable, reviewable. No more _v7_FINAL_v2.xlsx.
• AI that reasons. Named variables, typed dependencies - structured code is fundamentally easier for AI to understand than flat cells. AI reads, modifies, and explains a living model.
• Live data connections. pd.read_sql("SELECT * FROM actuals", connection) replaces copy-paste from the ERP.
• Advanced analytics. numpy, scipy, statsmodels, scikit-learn - tens of thousands of functions, all auditable.
• Testability. Unit tests for financial models. CI/CD for the CFO’s office. Errors become preventable, not inevitable.

Python-Excel parity is not a feature. It is the architectural missing piece that unlocks every capability financial modeling has waited for decades.

And then, financial modeling becomes possible to engineer. And with connecting it to the entire data pipeline, it reveals enormous possibilities.

Giving AI financially engineered models connected deep to data - is what it takes to unlock AI’s full capabilities and finally bring it into operations alongside text and factual source data. The numbers, scenarios, and forecasts - that is what AI lacks to fully reason about decisions.

We’re building Modeleon - the first Financial Model Engineering platform. You write revenue = price * volume in Python. The system generates =B4*B5 in the Excel cell. Not values - live formulas. Version-controlled. Connected to live data. Human-transparent.

If this resonates - stay close. We’re building it.