You have an app named App1 that uses a custom Azure AI Document Intelligence model to recognize contract documents. You need to ensure that the model supports an additional contract format. The…

Lowering the confidence score threshold would increase the likelihood of incorrect extractions and would not teach the model to recognize a new format effectively.

Creating an entirely new training set and a new model for the additional format is redundant and does not minimize development effort, as it disregards the existing trained model and dataset.

The most efficient way to expand a custom Document Intelligence model's capabilities is to add samples of the new contract format to the existing training dataset and then retrain the model. This incremental approach leverages the prior training, allowing the model to adapt and learn the new format without starting from scratch, thereby minimizing development effort.

Lowering the accuracy threshold of the application does not improve the model's ability to recognize new formats; it merely accepts less accurate results, potentially leading to unreliable data extraction.