Skip to main content

Importing Historical Data with the AI Assistant

Drop a file with your past results into a draft experiment and the AI assistant turns it into historical data for you. It reads spreadsheets, CSV files, PDF and Word reports, works out which columns are your variables and results, shows you a preview, and imports the data once you approve it. There is no need to reformat the file into a strict CSV first.


What You Can Import

The assistant accepts the following files:

  • Spreadsheets — Excel (.xlsx, .xls) and CSV (.csv)

  • Reports — PDF (.pdf) and Word (.doc, .docx), when the results are laid out in a table

  • Text — plain text (.txt) and Markdown (.md) holding a table


How to Import a File

  1. Open your draft experiment and go to the Historical data section.

  2. Drop your file onto Drop or click to import data with the assistant, or click it to pick a file. You can drop several files at once.

  3. The assistant chat opens and asks the assistant to import the file as historical data for this draft. If the assistant is still answering a previous message, the file waits in the message box — send it once the assistant is ready.

You can also attach the file directly in the chat and ask the assistant to use it as historical data.


What the Assistant Does

  1. Reads the file. It summarizes what the table holds and tells you how it reads the layout: whether each run is a row or a column, whether the file uses a dot or a comma as decimal mark, and how several tables fit together.

  2. Proposes the mapping. It suggests which columns are variables and which are results, and which runs to leave out, such as failed attempts. It only asks about what the file does not make clear.

  3. Adds what the draft is missing. Each column is linked to the variable or result of the same name in your draft. If you want a column the draft does not have yet, the assistant adds it to the draft first, with a range or categories that cover the values in the file.

  4. Shows a preview. You see the first rows as they will be saved, and every problem it found, in plain words. Ask for changes until the preview is right.

  5. Asks for your approval. A card titled Import this data into the draft? shows the dataset name, the file and tables it reads, and a table of every column with its role, unit and where it is read from in the file.

Click Import it to save the data, or Change something to tell the assistant what to adjust.


After the Import

The assistant tells you how many runs it imported. The data is saved as a dataset and appears in the draft's Historical data list, already mapped to your variables and results.

If your draft has variables or results the file has no column for, the assistant tells you. Before you can start the experiment, open Configure datasets and give each of them a column or a Fixed value — see Attaching Historical Data to an Experiment.


Good to Know

  • Drafts only. The assistant imports historical data while the experiment is a draft. To add data to a running experiment, upload a CSV file in the Historical data section — see Preparing Your Data File (CSV Format).

  • Values are copied, not retyped. The import copies the numbers from your file exactly. The assistant never writes the values out for you to paste in.

  • Check the preview, especially for PDFs. The assistant reads the text and tables extracted from your file, and PDF tables in particular can come out with headers merged or shifted. Compare the first rows of the preview with your file before you approve.

  • Partial imports. If only some runs can be read, the assistant imports those and tells you which runs it left out and why, instead of dropping the whole file.

  • Upload limits. Files count toward the assistant's per-conversation limits — see Uploading Documents to the AI Assistant.

  • Data quality still matters. The optimizer trusts imported data like new measurements, so only import runs from the same or a very similar setup — see Why Use Historical Data?

Did this answer your question?