Hi Bogdan Pechounov ,
Thank you for reaching out on the Microsoft Q&A .
Custom template models in Azure Document Intelligence skip complex neural networks. Instead, they rely on templates you define. During training, you provide documents with labeled data points. The model learns the positions of these points relative to the template and extracts text using OCR. This approach is fast for well-defined forms, easy to understand, and allows you to see how data is extracted.
I hope this helps! If you have any further questions, feel free to ask.
If the information is useful, please accept the answer and upvote it to assist other community members.