Hello experts, I have a question regarding the IXP and Action Center–based HITL setup.
When a user manually annotates document fields in the Action Center to correct data extraction, does this feedback get used to train or improve the backend model, or is it only applied at the document level?
I could not find any setting for this in IXP so wanted to understand how the HITL helps here.
Any official documentation of Action Center + IXP training would also help and how is your personal experience in this regard.
hi @tejaskumar.darji
Action Center corrections in current UiPath IXP setups provide HITL validation at the document level only - they do not automatically train/improve backend IXP models.
IXP Status: No settings exist for Action Center feedback to train models. Improvement comes via prompt instructions/configurations only. Future releases planned for feedback loops IXP FAQ.
Document Understanding Alternative: DU collects Action Center exceptions for manual retraining via “Exceptions for review” feature DU Retraining.
2 Likes
Hi @tejaskumar.darji
The IXP with action center, user corrections are applied only to the processed document and are not automatically used to retrain or improve the extraction model. Model improvement requires explicitly exporting validated data and retraining a custom ML model through AI Center. HITL mainly ensures accuracy for current transactions, not continuous learning.
Happy Automation
1 Like
Hi @tejaskumar.darji
When a user corrects data in Action Center (HITL) for a document, that correction only applies at the document level by default. It helps the process continue with accurate data, but it does not automatically improve the backend extraction model just by validating it in Action Center. UiPath Documentation
For the model to actually learn from those corrections, you need to send the human-validated data back into a training pipeline (for example using AI Center data labeling and retraining workflows). In that setup, you collect the corrected data, upload it to the AI Center dataset, and trigger a retraining run so the model learns from that feedback. UiPath Documentation
Official UiPath docs on “Using Data Labeling with human in the loop” explain how to close the loop and use validated data for training. UiPath Documentation
In my experience, HITL in Action Center is great for improving extraction quality for that batch, but you must explicitly include a training step if you want long-term model improvements. Otherwise, the changes stay only at the document level.