Azure Document Intelligence: Analyzing Multiple Document Types Together

Question

Azure Document Intelligence: Analyzing Multiple Document Types Together

Steven Darby 1

Currently, we have used prebuilt models like "prebuilt-idDocument" for passport analysis, and we have also trained a custom model for specific academic test result documents.

There is now interest in supporting the scenario where uploaded documents may include various different document types. We want to be able to recognise that e.g. page 1 is a passport and analyse that accordingly, and page 2 is an academic test result and analyse that accordingly.

While researching, information about composing models and custom classifiers was found, but it appears that this option is limited to custom models, which restricts the inclusion of prebuilt models like "prebuilt-idDocument".

Is there a way to achieve what we want? If not, are there any plans for its implementation in the future?

Saideep Anchuri 9,015 Reputation points Microsoft External Staff Moderator

2025-06-16T13:29:53.3033333+00:00

Hi Steven Darby

Currently, Azure Document Intelligence allows for the use of composed models, which can group multiple custom models together for analysis. However, as you noted, this feature is limited to custom models and does not include prebuilt models like "prebuilt-idDocument."

where different pages of a document are analyzed according to their specific types (e.g., a passport on one page and an academic test result on another)—you would typically need to use a custom classification model to identify the document types first. Once identified, you can then invoke the appropriate custom extraction model for each type.

Unfortunately, as of now, there is no direct support for combining prebuilt models with custom models in a composed model setup. There are no current indications or plans mentioned in the provided context regarding future implementations that would allow for this integration.

Kindly refer below link: composed-models

custom-classifier

Thank You.
Saideep Anchuri 9,015 Reputation points Microsoft External Staff Moderator

2025-06-17T01:51:09.66+00:00

Hi Steven Darby

Did You get any chance to check above response.

Thank You.
Steven Darby 1 Reputation point

2025-06-17T08:17:35.0233333+00:00
Thank you @Saideep Anchuri . Just to confirm then, the available options for the example I gave include:

Train a custom classification model, which would involve doing my own training for passports

(Not ideal) Analyse with the prebuilt-idDocument model and get any passport documents it may have found, then try to analyse with our custom academic results model and get any academic result documents it may have found. Use both sets of analysis results.

Regarding #1, I suppose my training wouldn't have to be thorough, just enough to identify it as a passport - then I can use the prebuilt model for the actual analysis.
Saideep Anchuri 9,015 Reputation points Microsoft External Staff Moderator

2025-06-18T01:11:52.4966667+00:00

Hi Steven Darby

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet.

Thank You.

1 answer

Your answer

Saideep Anchuri 9,015 Reputation points Microsoft External Staff Moderator

2025-06-16T13:29:53.3033333+00:00

Hi Steven Darby

Currently, Azure Document Intelligence allows for the use of composed models, which can group multiple custom models together for analysis. However, as you noted, this feature is limited to custom models and does not include prebuilt models like "prebuilt-idDocument."

where different pages of a document are analyzed according to their specific types (e.g., a passport on one page and an academic test result on another)—you would typically need to use a custom classification model to identify the document types first. Once identified, you can then invoke the appropriate custom extraction model for each type.

Unfortunately, as of now, there is no direct support for combining prebuilt models with custom models in a composed model setup. There are no current indications or plans mentioned in the provided context regarding future implementations that would allow for this integration.

Kindly refer below link: composed-models

custom-classifier

Thank You.
Saideep Anchuri 9,015 Reputation points Microsoft External Staff Moderator

2025-06-17T01:51:09.66+00:00

Hi Steven Darby

Did You get any chance to check above response.

Thank You.
Steven Darby 1 Reputation point

2025-06-17T08:17:35.0233333+00:00

Thank you @Saideep Anchuri . Just to confirm then, the available options for the example I gave include:

Train a custom classification model, which would involve doing my own training for passports

(Not ideal) Analyse with the prebuilt-idDocument model and get any passport documents it may have found, then try to analyse with our custom academic results model and get any academic result documents it may have found. Use both sets of analysis results.

Regarding #1, I suppose my training wouldn't have to be thorough, just enough to identify it as a passport - then I can use the prebuilt model for the actual analysis.
Saideep Anchuri 9,015 Reputation points Microsoft External Staff Moderator

2025-06-18T01:11:52.4966667+00:00

Hi Steven Darby

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet.

Thank You.

Answer 1

Hi Steven Darby

You are correct in your understanding of the options available for analyzing multiple document types together using Azure Document Intelligence.

Custom Classification Model: You can train a custom classification model specifically for identifying passports. This model would require a minimum of five samples per class to train effectively. While thorough training would improve accuracy, you can indeed focus on just enough training to classify documents as passports, and then use the prebuilt model for detailed analysis.
Prebuilt Model Approach: Alternatively, you can analyze documents using the prebuilt-idDocument model to identify any passport documents it finds. After that, you can analyze with your custom academic results model for academic documents. This approach allows you to leverage both sets of analysis results.

Both options are viable, and your choice may depend on the specific requirements of your project and the resources available for training.

Kindly refer below link: document-splitting

models-and-development-options

Thank You.

Share via

Azure Document Intelligence: Analyzing Multiple Document Types Together

1 answer

Your answer