Extract data

Extract data uses an AI model to pull structured values out of one or more existing columns and write them into new output columns. It's useful when the information you need is embedded inside free-text fields, such as pulling a state abbreviation out of a full address column.

Availability: Requires the WorkflowsAiExtractData feature. Contact your OneSchema support representative if you do not see this transform.

What it does

Input: one or more CSV lists.
Output: the same CSV list with new output columns added, populated by the model based on your source columns and prompt.
The transform reads the values in your chosen source columns for each row and asks the model to produce the requested output values.

When to use it

A single free-text column contains multiple pieces of data you need split out, such as a "Notes" field that sometimes contains an order number or reference code.
You need to normalize or derive a value (like a category or classification) from unstructured text in existing columns.
Manual review currently reads through a text column to manually tag or classify rows.

How to configure it

In the MFF builder, add an Extract data node after the node that produces the CSV list.

Settings

Setting	Description	Default
Prompt	A plain-language description of what to extract from the source columns and how to format it.	Empty
Source column names	The columns the model reads from when performing the extraction.	Empty list
Output column names	The new column(s) that will be added to the list to hold the extracted values.	Empty list
Row limit	Optional cap on the number of rows processed, useful for testing on a sample before running on a full file.	None

Prompt tips

Be specific about format: "Extract the 2-letter US state abbreviation from the Address column. If none is found, leave blank."
Handle ambiguity explicitly: "If multiple order numbers appear in Notes, return only the first one."

Example

Input

Customer	Notes
Acme Co	Ref #: ORD-4821, delivered late

Output

Customer	Notes	Order Number
Acme Co	Ref #: ORD-4821, delivered late	ORD-4821

Troubleshooting

Output column is mostly blank

Confirm the source column actually contains the data you're asking for in a sample of rows.
Make the prompt more specific about the pattern to look for.

Extraction is slow on large files

Use Row limit while testing your prompt, then remove or raise it for the full run.

Values are inconsistent across similar rows

Add explicit formatting rules and edge-case handling to the prompt (for example, what to do when no match is found).