Forecasts, allocation keys and board packs in Excel, answered against the ledger they depend on. The original file stays untouched, every version is kept and every figure traces back to its source.

Book a walkthrough How it compares

One conversation for your Excel file and your books

The cells you care about, the ledger behind them and the workbook you take away, held together by one version history.

Chat with Data with an Excel forecast workbook open in the preview panel. Row 3, COGS, is selected and attached to the next message.

An uploaded workbook, the question “Forecast my EBIT for 2026 for AT company”, and an answer with a summary table that has an Excel download button.

File and ledger, one answer

Plan, forecast and actuals in one answer. The figures in your file meet the figures in your ledger in the same table, ready to take into Excel.

The workbook preview with the COGS row selected and the notice that the range was attached to the next message.

Point at a range. Get a cited answer.

Every figure in the answer comes back with its address. Select a range, a row or a column, and the reply cites exactly the cells it used.

The message box with the attached reference pnl-forecast-2026-10-02.xlsx, version 1, Forecast!A3.

A chart of booked and forecast monthly results, followed by a data-context line and a table of the same figures.

Booked and forecast, never confused

Actuals come from the ledger, forecast months are marked as forecast, and every chart states its basis, period and scope.

The context panel with one output, pnl-forecast-2026-10-02.xlsx version 1, and one source, EBIT 2025 file.xlsx version 1.

Every output knows its sources

Your upload is never touched. Every generated file records the versions it was built from, so where a number came from always has an answer.

One reconciled ledger. Every source. Your Excel workbooks on top.

SAP, DATEV, your banks and your CRM reconcile into one canonical ledger with master records. Every workbook you bring is analysed against it, alone or several together, with no export and no VLOOKUP in between.

Connected sources

ERP and accounting systems such as SAP and DATEV
Bank accounts at more than 170 banks
CRM and sales systems such as Salesforce
Databases and file uploads

Canonical ledger and master records

One double-entry ledger, reconciled across sources
One record each for customers, suppliers, products and employees
Every figure traceable to its source record

Your workbooks

An uploaded file becomes a table beside the ledger
One workbook against the books, or several workbooks together
Answers name the sheet and the cells they used

One question. Your workbooks. Every connected source.

Built for the Excel files finance actually keeps

Formulas kept as formulas. Every version kept. Limits stated, not discovered.

Editing: Values, formulas, formats, rows, columns and sheets, changed in one batch and saved as one version with a note. Edit by asking, or type, paste and fill straight into the preview.
Formulas: Formulas stay formulas. 43 functions are recalculated and verified against Excel before a version is saved. Anything else is marked pending and left for Excel to calculate on open, never replaced by a number.
Versions: Nothing is overwritten. Every version can be opened, restored or downloaded, and an edit made on an older version waits until you apply it to the current one.
Exports: An export carries its own audit trail: a contents sheet, one sheet per result, the source documents behind each figure with their reconciliation, and the data sources used. A partial list says it is partial.
CSV files: CSV stays CSV, byte for byte, with the delimiter detected and UTF-8 encoding. A formula written into a CSV is saved as its result, and the assistant says so.
Limits: 30 MB per file, 1,000,000 cells and 256 sheets per workbook, 200,000 cells per saved edit. Charts, pivot tables and every part an edit does not touch come back exactly as they were. Legacy .xls files need saving as .xlsx first.
Data handling: EU-hosted and isolated per tenant. Never used to train models. Every computation on your data is logged, and cell contents, file names and comments are read as data, never as instructions.

Where general-purpose AI for Excel stops

Claude, ChatGPT and Copilot now edit Excel well. None of them connects a file to a reconciled ledger, traces an output to its sources, or ships an export with its reconciliation.

Claude, ChatGPT and Microsoft Copilot all open Excel files, cite cells and apply edits described in plain language. Claude for Excel and ChatGPT for Excel run as add-ins inside Excel; Claude's chat offers a file download, and ChatGPT previews files in its desktop app. Copilot works inside Excel and records who changed what in its Change History.^{[1, 3, 7, 8, 11, 13]}

None of the three documents a link from a generated file back to the files it was made from, grounding in a reconciled ledger or bank data, or an export that carries source documents and a reconciliation. Their connectors cover market data and Microsoft 365 content.^{[1, 6, 7, 14]}

Claude offers EU hosting only through cloud marketplaces and without its connectors, and the add-in sits outside Enterprise audit logs. ChatGPT offers EU residency on enterprise plans. Copilot works within the EU Data Boundary, from which models provided by Anthropic are excluded.^{[4, 5, 10, 14]}

Based on each vendor's public documentation as of October 2026. Gemini in Sheets follows the same pattern and works best with native Sheets files. Product names are trademarks of their respective owners.^[15]

Sources

Bring a workbook

A walkthrough on your own file, with your own data.

Book a walkthrough