# How to make a financial model readable by AI without losing formulas

> Practical steps for FP&A and finance teams to make Excel models readable by AI while keeping formulas, assumptions, named ranges and periods intact.

To make a financial model readable by AI without losing formulas, separate inputs from calculations, label every assumption with its period and unit, use named ranges, and share a version where each value keeps its formula and dependencies, not a flattened PDF or a pasted table. AI tools then see how numbers are derived, not just what they are.

## Why AI struggles with spreadsheets

A model’s meaning lives in structure: which cells are inputs, which are calculated, what period each column covers, which scenario is active. Export to PDF or screenshot and all of that disappears; AI sees a grid of numbers with no logic behind them.

Even with the XLSX itself, an assistant may read cached values without the formulas, misjudge units set by number formatting (thousands, percentages), or confuse a forecast column with actuals because the label sits several rows away.

## Structure the model for a machine reader

- Keep a dedicated Assumptions sheet: one input per row, with name, value, unit, period and source.
- Use named ranges for key drivers (for example Growth_Rate_FY27, Churn_Monthly) so formulas read like sentences.
- Put period labels in a single header row and state actual vs forecast in the header itself ("FY26A", "FY27F").
- State units in headers ("USD thousands"), not only through number formats.
- Avoid hard-coded numbers inside formulas; reference an input cell instead.
- Separate scenarios clearly, and label which one is active.
- Remove dead sheets, hidden test tabs and circular references you do not need.

## What to preserve when you share

| Element | Why AI needs it |
|---|---|
| Formulas | Shows how outputs depend on inputs, so questions like "what drives EBITDA?" have real answers |
| Dependencies | Lets an agent trace a total back to its assumptions |
| Named ranges | Gives drivers human-readable names AI can quote |
| Number formats | Distinguishes 0.05 from 5%, and units from thousands |
| Period and status | Separates actuals from forecasts and budgets |
| Cell reference | Allows any figure to be checked against the model |

## Formats compared

- PDF export: readable by people, loses formulas and often units; AI sees only values.
- CSV: clean values, but one sheet at a time and no formulas, formats or names.
- Raw XLSX upload: complete, but each tool interprets it its own way and readers may get an outdated copy.
- Verified AI version: values, formulas, dependencies and named ranges extracted once, checked, and published at one link.

## How The Papers handles Excel models

The Papers reads cell values with their number formats, formulas and their dependencies, and named ranges. During fidelity check, formulas are preserved with dependencies resolved, every number, currency and period must match the source, and qualifiers such as forecast or unaudited stay attached. Each value keeps a reference to its source cell.

If a label is ambiguous, for example whether a line is Revenue or GMV, the question goes to you and your answer is recorded as owner-confirmed. The model is published as an AI Version with Markdown and JSON (metrics with period, currency and classification) at a permanent link; when you update the model, the link stays and the version history grows.

## A pre-share checklist for CFOs and FP&A

- Inputs separated and named.
- Every column labeled with period and A/F/B status.
- Units written in headers.
- No hard-coded overrides inside formulas.
- Active scenario stated.
- One link shared, not a file attached to five threads.

## FAQ

### Will AI see my formulas if I upload the XLSX directly?

Sometimes, depending on the tool and how it processes the file. A published version that includes formulas and dependencies removes that uncertainty.

### Should I share the full model or a summary?

Share what the reader needs. Visibility in The Papers is per document, so you can publish a summary publicly and keep the full model private.

### Does The Papers keep my original spreadsheet?

No. The original upload is deleted after processing; the fidelity-checked AI Version remains.

### Can it check that my assumptions are reasonable?

No. Fidelity check checks that the AI version is faithful to your model, not that the forecast is right.

Clean up the Assumptions sheet, then publish the model and ask an assistant what drives your key outputs. → https://thepapers.link/

## Related

- [How to prepare a board pack that directors’ AI tools read correctly](https://thepapers.link/blog/board-packs-for-ai.md)
- [How to make an earnings report or investor update readable by AI](https://thepapers.link/blog/earnings-reports-and-investor-updates.md)
- [Markdown vs PDF vs JSON: which format AI reads best](https://thepapers.link/blog/markdown-vs-pdf-vs-json-for-ai.md)
- [The Papers: AI-readable data room for startup founders sharing investor decks](https://thepapers.link/blog/best-ai-readable-data-room-for-startup.md)

Updated 2026-10-27. Canonical: https://thepapers.link/blog/financial-models-readable-by-ai
