It happens every day. Someone opens the spreadsheet with the customer file or the planning, copies a section and pastes it into a chatbot. "What stands out?" The answer arrives within seconds, neatly phrased.
And then one question lingers: is this right?
That doubt is justified. Not because AI is not clever enough, but because an uploaded file is a poor conversation partner.
Datahub is the step beyond pasting a spreadsheet into a chatbot. You bring the Excel data and its meaning together in the Datahub company brain. HERC, your built-in assistant, helps you query that source in plain language. Connected AI assistants can use the same definitions and permissions through MCP. This is not AI reading an isolated Excel file; it is asking questions of the shared knowledge you govern in Datahub.
Why an uploaded file gives unusable answers
The copy is outdated the moment it uploads. The file copied this morning is already stale by the afternoon. Every answer describes how things were, not how they are.
The model does not know the meaning. A column called "status" can mean anything. Without agreed definitions the model fills in the meaning itself, and it sounds more convincing than it is.
There is no traceability. Where does this number come from? Which rows counted? On a copy that cannot be checked, so it comes down to gut feeling.
The data leaves the organisation. Customer names, prices and plans end up in an external service. For many organisations that alone is a reason to stop.
The same question, a different foundation.
The AI-native order: govern first, then query
The answer is not a better prompt, but a different order. First the spreadsheet becomes a governed business asset, as described in Outgrown Excel as a database? Build on it with Datahub. Only then do you talk to it with AI.
- Import the file. Columns become concepts with a definition, so "customer" and "status" mean the same thing to everyone.
- The data is alive. Changes get a history. A question is about the current state, not a snapshot.
- Everyone queries the same source. Colleagues ask questions through the interface; AI assistants connect via MCP. Both use the same definitions and the same permissions.
- Every answer is traceable. From the answer back to the source, down to the row. Doubt becomes checkable.
Import, understand the meaning, ask and trace.
That is the difference between talking to a copy and talking to the truth. More about the foundation is in From your critical Excel file to the Datahub company brain.
From answers to actions
Once questions are answered reliably, the next step comes into view by itself. An alert when a value deviates. An automation that checks new rows. And for organisations that do not want to send their data to an external AI service: private AI, on your own hardware, with the same definitions and permissions.
Datahub starts from €499 per month, excl. VAT.
