Skip to main content

​
Why this matters

A lot of your most valuable data lives in spreadsheets — pipeline exports, financial models, survey results. But a spreadsheet isn’t prose: it’s rows, columns, and multiple sheets that only make sense as structured data. Chief handles spreadsheets on a dedicated path so it can reason over their structure rather than treating them like a wall of text. Knowing how that path works tells you when a spreadsheet answer will be reliable and how to set yourself up for accurate results.

​
How it actually works

Chief spreadsheets - a CSV file card with a structured table preview
Chief recognizes structured files (CSV and Excel) and routes them through a different pipeline from documents:
  • Excel files are split by sheet. An .xls/.xlsx workbook is converted into one CSV per sheet, and each sheet is then processed independently — so a multi-tab workbook becomes several queryable tables rather than one tangled blob.
  • CSV files are sampled and embedded. Chief reads the file, captures a representative sample of the data so it understands the shape and content, and indexes it so the structured data is searchable alongside the rest of your knowledge.
Because structured data behaves differently from documents, Chief asks you to add spreadsheets to a task individually to work with them directly. This keeps the model focused on the specific tables you want to analyze rather than mixing them in with unstructured files.
ℹ️ Important: In a Project with more than 50 non-spreadsheet files, spreadsheets are unavailable when every one of those files is selected in your task scope. Smaller Projects don’t lock spreadsheets. To work with structured data in a larger Project, deselect at least one file first, then add the spreadsheet you want. Also note that file counts shown in the prompt may include only files loaded in your current browser session — your Project may contain more.

​
A worked example

Chief spreadsheets - the Spreadsheets tab when adding a file to a task
You upload a workbook with three tabs — “Q1 Pipeline,” “Q2 Pipeline,” and “Notes.” Chief splits it into three separate CSVs and processes each one, so when you add the workbook to a task and ask “which deals slipped from Q1 to Q2?”, Chief is reasoning over two clean, structured tables rather than one merged sheet. If your Project held more than 50 other files and you’d left every one of them selected, Chief would prompt you to deselect at least one file before it could work with the spreadsheet directly.

​
FAQ

In a Project with more than 50 non-spreadsheet files, structured data is unavailable when all of those files are selected in your task. Deselect at least one file, then add the specific spreadsheet you want to analyze. Chief surfaces a notice and a link to enable spreadsheets when this happens.
Structured files are processed and reasoned over differently from documents. Adding them individually keeps Chief focused on the exact tables you want, which produces more accurate analysis than blending them into a large pool of unstructured files.
Chief captures a representative sample of the data to understand its structure and content, and indexes it for search. When a spreadsheet is in your task scope — including through the whole Project — Chief selects the sheets most relevant to your question and runs queries against their full tables. Totals, counts, and averages on a queried sheet cover every row, not just the sample. Chief queries at most 10 selected sheets in one search call, and a query that asks for individual rows returns at most 100 of them. Naming the specific columns, time ranges, or segments you care about still gives the most reliable answers.