Hi everyone, good day.
I’m working with an AI Skill for document extraction in Appian, specifically targeting a structured and standardized document used in Mexico: the Constancia de Situación Fiscal (a tax ID certificate).
My goal is to extract the following fields:
RFC (tax ID)
Business or legal name (Razón Social)
Start date of activities
List of economic activities (shown as a table in the document)
The issue I’m facing is that the AI Skill successfully extracts all the individual fields (RFC, name, start date), but it completely skips the table that contains the economic activities.
I’ve already configured the document type, field mappings, and extraction zones where possible — I’ll share screenshots of my current setup below.
Has anyone encountered a similar issue with table extraction in structured documents?
Any suggestions would be greatly appreciated.
Best regards, Víctor Valenzuela
Discussion posts and replies are publicly visible
AI Skill may not extract table data if the tables are not clearly machine-readable, or the headers do not exactly match the expected field names. Also, don't allow manual labeling of tables during training - "you won't label the fields in the items table, but that information will still be extracted" automatically by the pre-trained model. However, the model doesn't recognize the unique format of the table in document. The quickest fix is to create a new AI Skill using "Unstructured" document extraction with a specific prompt to extract the table(Extract the complete table from the document, including all columns: clave, descripción, fecha inicio, and porcentaje). Alternatively, ensure your structured extraction has the table field set as type "Table" (not Text), properly named nested fields, and at least 50 training documents. Since the pre-trained model wasn't designed for this document, unstructured extraction with targeted prompts will give you better immediate results than relying on automatic table detection.https://docs.appian.com/suite/help/25.2/create-skill-doc-extraction.html
Víctor Valenzuela I agree that using the Generative AI Skill is a good idea as it is flexible to changing formats and working with complex tables. In fact, we have an intelligent document processing solution backed by these AI Skills that was released in 25.2 called the AI Document Center to serve this exact need and a lot more. The documentation lists how you can set up an extraction model and extract from tables. Please let us know if that helps!