FP&A models, budgets and forecasts with versions
Spreadsheet-based planning work: operating models, budget cycles and forecast revisions, kept with their formulas, assumptions, versions and reviewer comments. Labs use them to train spreadsheet and analyst agents on how real models are built and revised.
What a record contains
One model or planning cycle: workbook versions with formulas intact, assumption tabs, variance commentary and approval notes.
| Field | What it holds |
|---|---|
model_id, purpose | Budget, forecast, board pack or deal model |
versions[] | Date and workbook reference |
assumptions | Driver tabs as built |
variance_comments[] | Actuals versus plan, explained |
approvals[] | Who signed off, by role |
{
"model_id": "FPA-77",
"purpose": "FY25 operating budget",
"versions": [{"at": "2024-10-15", "ref": "v3"}, {"at": "2024-11-20", "ref": "v7_board"}],
"variance_comments": [
{"line": "Freight-in", "period": "2025-02",
"comment": "Over plan 12% on fuel surcharge; partially offset by lane renegotiation."}
]
}How AI labs use it
- Spreadsheet agents
- Real formulas, structures and conventions.
- Variance analysis
- Commentary explaining actuals against plan.
- Evaluation
- Ask a model to update a forecast and compare with the team’s revision.
Typical preparation requirements
Agreed with the supplier before any work begins. Typical requirements include:
- Figures scaled or masked if required, consistently across versions
- Compensation and personal data excluded
- Public companies’ current forecasts can be material non-public information, so they need legal review before licensing
Every dataset has a documented owner and confirmed licensing rights. See data governance on sourcex.si.
What makes a strong package
- Formulas intact, not pasted values
- Multiple cycles from the same company
- Commentary on variances
Compared with public datasets
Public sets such as FinQA and TAT-QA are useful references, but limited as enterprise training data. The finance and accounting category page compares them with licensed data.
Who typically holds it
- Private companies
- Private-equity portfolio companies
- Consulting firms, with clients’ permission
Know a company like this?
Introduce the company to SourceX. If its data deal closes, you can earn up to $100,000 in referral fees, paid after the buyer accepts the data and SourceX receives payment.
Refer a companyQuestions
Is the company’s real performance revealed?
Figures can be scaled consistently so the model’s structure still works. Scaling can be reversed if any true amount is known, so sensitive figures may need further masking.
Need this data for a model?
Describe what you need: domain, volume, history, format and licensing terms. SourceX looks for companies that hold it and can license it.