Azure Document intelligence prebuilt model - West US 2 Slow

Question

Azure Document intelligence prebuilt model - West US 2 Slow

Erik Nylander 5

I've been noticing for the past 24 hours that prebuilt-model invoices are taking 15-20 mins to run. Even using the samples from DI Studio are also experiencing the same issue.

Are there issues happening with West US2?

Anonymous

2025-09-17T19:23:40.96+00:00

Hello Erik Nylander

Over the past 24 hours, prebuilt-model invoices have been taking 15–20 minutes to run, even when using samples from Document Intelligence Studio. There are no reported outages or performance degradations in the West US2 Azure region, which is currently operating normally. However, known issues with Azure’s prebuilt models—such as missing fields and ModelNotFound errors—can occur due to API version mismatches, regional availability, or endpoint differences.

These models may behave differently when used via API versus DI Studio. Internal documentation also recommends using custom prebuilt models for better performance and accuracy. To resolve the delay, verify model ID and API version, confirm regional support, and consider switching to a custom model. Azure Service Health can be used for proactive monitoring.
it is advisable to verify that the correct model ID and API version are being used, and that the model is supported in the West US2 region. If the delays persist, consider deploying a custom model using AI Builder or Form Recognizer Studio, which offers greater control and optimization for your specific invoice types. Additionally, setting up personalized alerts through Azure Service Health can help monitor any future regional disruptions more proactively.

Reference Links: Document Intelligence invoice model

Hope it Helps!

Thank you
MC 0 Reputation points

2025-09-17T20:38:51.39+00:00

Same here, it's clear that we had a West2 outage today but Azure Team doesn't even acknowledge it. This is unacceptable!!!
Erik Nylander 5 Reputation points

2025-09-17T22:01:45.6833333+00:00

We are using API version 2024-11-30 for prebuilt-invoice and prebuilt-layout. We have not had any issues, and no code changes when we experienced the slow response times.

Currently no plans to use a custom model except for edge case documents that we see.

I understand that the service health says it's ok, but clearly there are not enough resources, or something happened on west us2 that caused such a massive slow down.

Thanks
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-18T07:15:32.2733333+00:00

Hi Erik Nylander

I just tested some documents in West Us 2 regions. It got processed under 10 second for me.

You can split the larger pdfs to smaller ones if the size is high for processing.

Suggesting testing other regions like Canada east and East US if those are prebuilt models.

Requested details in private message for further investigation.
Sergio Rodenas Romero 15 Reputation points

2025-09-18T08:08:12.5533333+00:00

Hi @Anonymous , @Manas R Mohanty

I am experiencing the same issue. I am in West Europe and using the prebuilt Invoices with API version 2024-11-30 (4.0 General Availability). I have tested calling the Document Intelligence service both from a microservice and directly through the Azure portal using Document Intelligence Studio. The result has been the same since yesterday: processing times of over 10 minutes.

This happens with all invoices I have tested, both my own and the sample ones provided by Azure.

Best regards.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-18T09:27:23.99+00:00

Sergio Rodenas Romero

Thank you for clarifying that you are operating from West Europe.

Updating the metric screenshot in concerned PG channel thread.

Waiting for inputs from Erik Nylander for needed support ticket and PG escalation.Could you create a new thread and tag me to get the details of your resources.

Thank you.
Muhammad Zubair 20 Reputation points

2025-09-18T10:21:42.0633333+00:00

I’m currently testing the performance of Azure Document Intelligence prebuilt models in West US 2.
Sergio Rodenas Romero 15 Reputation points

2025-09-18T14:08:23.3133333+00:00

@Manas R Mohanty

As you suggested, I opened a separate thread with the details of my case, but it was closed shortly after. I assume this may have been because it looked very similar to the original thread.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-23T02:50:26.56+00:00

Hello Erik Nylander

Good day.

Please help me with details requested in private message if the issue is not resolved at your side.

We need those details create support ticket for seeking product group attention.

Thank you for understanding.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-23T02:54:28.5566667+00:00

Hi Sergio Rodenas Romero

Threads are moderated actively, please don't include any metric screenshot or PII information (like subscription, email etc.) and share me your thread.

Thank you.

1 answer

Your answer

Anonymous

2025-09-17T19:23:40.96+00:00

Hello Erik Nylander

Over the past 24 hours, prebuilt-model invoices have been taking 15–20 minutes to run, even when using samples from Document Intelligence Studio. There are no reported outages or performance degradations in the West US2 Azure region, which is currently operating normally. However, known issues with Azure’s prebuilt models—such as missing fields and ModelNotFound errors—can occur due to API version mismatches, regional availability, or endpoint differences.

These models may behave differently when used via API versus DI Studio. Internal documentation also recommends using custom prebuilt models for better performance and accuracy. To resolve the delay, verify model ID and API version, confirm regional support, and consider switching to a custom model. Azure Service Health can be used for proactive monitoring.
it is advisable to verify that the correct model ID and API version are being used, and that the model is supported in the West US2 region. If the delays persist, consider deploying a custom model using AI Builder or Form Recognizer Studio, which offers greater control and optimization for your specific invoice types. Additionally, setting up personalized alerts through Azure Service Health can help monitor any future regional disruptions more proactively.

Reference Links: Document Intelligence invoice model

Hope it Helps!

Thank you
MC 0 Reputation points

2025-09-17T20:38:51.39+00:00

Same here, it's clear that we had a West2 outage today but Azure Team doesn't even acknowledge it. This is unacceptable!!!
Erik Nylander 5 Reputation points

2025-09-17T22:01:45.6833333+00:00

We are using API version 2024-11-30 for prebuilt-invoice and prebuilt-layout. We have not had any issues, and no code changes when we experienced the slow response times.

Currently no plans to use a custom model except for edge case documents that we see.

I understand that the service health says it's ok, but clearly there are not enough resources, or something happened on west us2 that caused such a massive slow down.

Thanks
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-18T07:15:32.2733333+00:00

Hi Erik Nylander

I just tested some documents in West Us 2 regions. It got processed under 10 second for me.

You can split the larger pdfs to smaller ones if the size is high for processing.

Suggesting testing other regions like Canada east and East US if those are prebuilt models.

Requested details in private message for further investigation.
Sergio Rodenas Romero 15 Reputation points

2025-09-18T08:08:12.5533333+00:00

Hi @Anonymous , @Manas R Mohanty

I am experiencing the same issue. I am in West Europe and using the prebuilt Invoices with API version 2024-11-30 (4.0 General Availability). I have tested calling the Document Intelligence service both from a microservice and directly through the Azure portal using Document Intelligence Studio. The result has been the same since yesterday: processing times of over 10 minutes.

This happens with all invoices I have tested, both my own and the sample ones provided by Azure.

Best regards.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-18T09:27:23.99+00:00

Sergio Rodenas Romero

Thank you for clarifying that you are operating from West Europe.

Updating the metric screenshot in concerned PG channel thread.

Waiting for inputs from Erik Nylander for needed support ticket and PG escalation.Could you create a new thread and tag me to get the details of your resources.

Thank you.
Muhammad Zubair 20 Reputation points

2025-09-18T10:21:42.0633333+00:00

I’m currently testing the performance of Azure Document Intelligence prebuilt models in West US 2.
Sergio Rodenas Romero 15 Reputation points

2025-09-18T14:08:23.3133333+00:00

@Manas R Mohanty

As you suggested, I opened a separate thread with the details of my case, but it was closed shortly after. I assume this may have been because it looked very similar to the original thread.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-23T02:50:26.56+00:00

Hello Erik Nylander

Good day.

Please help me with details requested in private message if the issue is not resolved at your side.

We need those details create support ticket for seeking product group attention.

Thank you for understanding.
Manas R Mohanty 17,270 Reputation points Moderator

2025-09-23T02:54:28.5566667+00:00

Hi Sergio Rodenas Romero

Threads are moderated actively, please don't include any metric screenshot or PII information (like subscription, email etc.) and share me your thread.

Thank you.

Answer 1

Hi Erik,

I just tested the same prebuilt models in the West US 2 region on my end, and they are processing normally without longer delays (Around 10 to 15 Seconds). This suggests the issue may be transient or related to specific requests or document sizes.

As mentioned earlier, to mitigate or troubleshoot delays, you can:

Verify configuration: Ensure the correct model ID and API version are used.
Consider a custom model: Deploying a custom model via AI Builder or Form Recognizer Studio can improve performance and accuracy.
Split large documents: Breaking up large PDFs can reduce processing time.

Test other regions: If feasible, compare latency in regions like Canada East or East US.

Note that Azure AI Document Intelligence is asynchronous and multitenant, so temporary slowdowns can occur even when the service reports healthy.