Skip to main content
Question

How Do You Handle Documents With Inconsistent Layouts in Document Automation?

  • Nov 18, 2026 7:10 AM UTC
  • 2 replies
  • 10 views

renvik04
Ground Control | Tier 1

I'm interested in understanding how Automation Anywhere handles document processing when files follow different layouts.

For example, invoices and business forms may contain the same type of information but place fields in different positions. This can make it challenging to extract data consistently without manual corrections.

I'd like to understand a few things:

  1. Is it better to use a single extraction model or create separate models for different document layouts?

  2. How does Document Automation handle missing or incorrectly detected fields?

  3. What is the recommended approach for improving extraction accuracy when document formats change frequently?

For those who have implemented document processing workflows, what approach has given you the most reliable results?

I'd appreciate practical suggestions and experiences.

2 replies

Aaron.Gleason
Automation Anywhere Team
Forum|alt.badge.img+6
  • Automation Anywhere Team
  • Nov 18, 2026 12:55 PM UTC

​@renvik04 The answer to all your questions is... Give it a try!

You can use a single "learning instance" for all your invoices (for example) or just a specific, difficult-to-extract invoice. 

Missing fields or fields not matching confidence levels bring that document into validation, where extraction accuracy is improved. 

Try it out in our free Community Edition.


KhaledMostafaMe
Most Valuable Pathfinder
Forum|alt.badge.img+2
  • Most Valuable Pathfinder
  • Nov 18, 2026 5:58 PM UTC

@renvik04  Invoices with the same fields in different positions are what this feature was built for, so start small and let the layouts teach the instance.

Use one learning instance per document type, not one model per layout. Layouts it recognizes extract automatically, and unfamiliar ones go to validation, where your corrections teach it.

  1. Single model or separate models? Start with one learning instance for invoices. Extraction uses object detection: the engine stores a fingerprint of each document's keys and their locations and extracts from any layout it knows. A new layout adds a fingerprint, so you don't build a model per template.

  2. Missing or wrong fields. A missing required field is an error to fix. A low-confidence field or type mismatch goes to validation, and the confidence threshold sets how many documents land there. In the Validator, select/draw the extraction region around the right value; that training teaches the layout.

    Upload invoice ➔ Extraction ➔ Validator (training region) ➔ Feedback saved for that layout

  3. Improving accuracy when formats change. Keep Improve accuracy using validation selected when you create the instance; it is on for all fields by default. Mark fields that may be absent as Optional, or valid invoices get flagged as errors.

  4. What has worked best. I can't speak from broad experience here. Check the validation queue after each batch to see which layouts still need training.