Slow response on Microsoft Foundry, Document Inteligent

Question

Slow response on Microsoft Foundry, Document Inteligent

Matias Haller 0

We are sending a batch of 2 documents, there are invioces 1 page each invoice. The documentAI model is taking as 1 hour to get this process.

We Would like to understan why this is happening.

Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-06T21:48:41.04+00:00
Hello @Matias Haller ,

Could you please help us with the following details for better assistance.

API version and model (prebuilt-invoice vs. custom) being used?

Region the resource deployed in

How is the service being called (Foundry flow step, REST, SDK)

If there are any error or throttling messages in logs

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-08T16:52:18.64+00:00

Hello @Matias Haller ,

Checking in to see if you had any chance to review the above response.

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-11T17:56:14.98+00:00

Hello @Matias Haller ,

Just checking in to see if the above response was helpful

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-12T14:15:43.4866667+00:00

Hello @Matias Haller ,

Just checking in to see if you have got a chance to see my response.

Since I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted with an upvote? This helps others in the community with the same question find the solution more easily.

Thank you!
Matias Haller 0 Reputation points

2026-05-12T16:47:47.52+00:00
API version and model (prebuilt-invoice vs. custom) being used?
We are using FormRecognizer, prebuilt-invoices with Tier S0 Standard

Region the resource deployed in
US-EAST

How is the service being called (Foundry flow step, REST, SDK)
Python SDK

If there are any error or throttling messages in logs
No error, only that two month ago all batches returned in 5 minutos, a now there are returning between 30 to 60 minutes.

We would like to know if this is a normal behiviour, and there is documentation on how much time could it take. So we can evalute the solution with all data.

Thank you!!!****
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-12T17:56:39.5666667+00:00
Hello @Matias Haller ,

Thank you for sharing the details.

Based on the observed behavior and recent service patterns, the increased processing time is most likely related to backend queue delays driven by regional load and request distribution, rather than any regression, configuration issue, or SDK-related concern.

This is a known and occasionally observed behavior under high-demand conditions, and applying the above optimizations can help improve consistency and overall processing times.

At present, there are no active service-wide outages specific to Document Intelligence in the East US region. However, latency issues can occur under certain backend conditions, even when requests are successfully accepted and processed.

This behavior is typically associated with internal processing dynamics rather than request failures, and the following factors are known to influence processing time:

Regional load and demand patterns as East US is a high-traffic region

Queue waiting time before execution begins

Batch processing behavior and request aggregation

Document size, complexity, and page count

Temporary fluctuations in service throughput or backend health

Since Document Intelligence operates on an asynchronous processing model, the total processing duration includes both:

Queue time - waiting for processing slot

Execution time - actual document analysis

Under higher load conditions, queue delays can significantly increase, which leads to longer overall completion times without generating any explicit errors.

Please let us know if the response was helpful and if the latency issue has been resolved

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-13T20:04:50.09+00:00

Hello @Matias Haller

Checking in to know if the above response was helpful.

Since I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted with an upvote? This helps others in the community with the same question find the solution more easily.

Thank you!

2 answers

Your answer

Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-06T21:48:41.04+00:00

Hello @Matias Haller ,

Could you please help us with the following details for better assistance.

API version and model (prebuilt-invoice vs. custom) being used?

Region the resource deployed in

How is the service being called (Foundry flow step, REST, SDK)

If there are any error or throttling messages in logs

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-08T16:52:18.64+00:00

Hello @Matias Haller ,

Checking in to see if you had any chance to review the above response.

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-11T17:56:14.98+00:00

Hello @Matias Haller ,

Just checking in to see if the above response was helpful

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-12T14:15:43.4866667+00:00

Hello @Matias Haller ,

Just checking in to see if you have got a chance to see my response.

Since I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted with an upvote? This helps others in the community with the same question find the solution more easily.

Thank you!
Matias Haller 0 Reputation points

2026-05-12T16:47:47.52+00:00

API version and model (prebuilt-invoice vs. custom) being used?
We are using FormRecognizer, prebuilt-invoices with Tier S0 Standard

Region the resource deployed in
US-EAST

How is the service being called (Foundry flow step, REST, SDK)
Python SDK

If there are any error or throttling messages in logs
No error, only that two month ago all batches returned in 5 minutos, a now there are returning between 30 to 60 minutes.

We would like to know if this is a normal behiviour, and there is documentation on how much time could it take. So we can evalute the solution with all data.

Thank you!!!****
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-12T17:56:39.5666667+00:00

Hello @Matias Haller ,

Thank you for sharing the details.

Based on the observed behavior and recent service patterns, the increased processing time is most likely related to backend queue delays driven by regional load and request distribution, rather than any regression, configuration issue, or SDK-related concern.

This is a known and occasionally observed behavior under high-demand conditions, and applying the above optimizations can help improve consistency and overall processing times.

At present, there are no active service-wide outages specific to Document Intelligence in the East US region. However, latency issues can occur under certain backend conditions, even when requests are successfully accepted and processed.

This behavior is typically associated with internal processing dynamics rather than request failures, and the following factors are known to influence processing time:

Regional load and demand patterns as East US is a high-traffic region

Queue waiting time before execution begins

Batch processing behavior and request aggregation

Document size, complexity, and page count

Temporary fluctuations in service throughput or backend health

Since Document Intelligence operates on an asynchronous processing model, the total processing duration includes both:

Queue time - waiting for processing slot

Execution time - actual document analysis

Under higher load conditions, queue delays can significantly increase, which leads to longer overall completion times without generating any explicit errors.

Please let us know if the response was helpful and if the latency issue has been resolved

Thank you
Karnam Venkata Rajeswari 3,830 Reputation points Microsoft External Staff Moderator

2026-05-13T20:04:50.09+00:00

Hello @Matias Haller

Checking in to know if the above response was helpful.

Since I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted with an upvote? This helps others in the community with the same question find the solution more easily.

Thank you!

Answer 1

Hello @Matias Haller

We have seen few outages last month on DI service and product group is actively working to improve it.

You can explore other OCR extraction options like

Create AI search index and interact with LLM models for insights.
Container support for document intelligence to scale https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/containers/disconnected?view=doc-intel-4.0.0

Attached relevant thread

https://learn.microsoft.com/en-us/answers/questions/5820282/batch-jobs-are-stuck-on-validating-and-then-all-fa?comment=question&translated=false

Have tried to reach you on teams for discussing above

Thank you.

Answer 2

Hello @Matias Haller ,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

For a batch of two single‑page invoices, an end‑to‑end processing time of approximately one hour is significantly higher than expected. Under normal operating conditions, this type of workload typically completes within seconds to a few minutes.

A review of current service status indicates that there are no publicly reported service‑wide outages or known incidents affecting Document Intelligence at this time. While this rules out a global platform issue, temporary regional capacity pressure, backend queue congestion or workflow‑level delays may still occur without appearing as a service‑wide incident which can contribute to the latency issues.

However , extended processing time is not related to document size or invoice complexity but rather to how the request is executed within asynchronous workflows. When requests are submitted through Foundry flows, SDK‑based orchestration, or batch pipelines, the operation may spend time in backend queues before document analysis begins. In addition, polling intervals, retries, and orchestration logic can significantly increase the total observed duration, even when the actual document extraction completes much faster.

To help isolate the root cause, please check if the following validations are helpful, listed in order of likelihood:

Validating Execution Method and Workflow Timing Confirm whether processing is triggered via:
- Azure AI Foundry flow
- REST API
- SDK
- another orchestration layer
Verify whether the workflow uses asynchronous polling or batch execution Compare:
- request submission time
- analyze operation start time
- completion retrieval time This helps determine whether the delay originates from queueing, polling, or actual document analysis.
Validating Service Tier and Quotas
1. Lower tiers such as F0 may experience throttling and queue delays
2. S0 tier or higher is recommended for sustained or production workloads
3. Check for throughput, concurrency, or shared resource usage that could impact performance
Reviewing throttling, retries, and quota signals Review logs for:
- 429 (Too Many Requests) responses
- retry patterns or back‑off behavior
- timeout activity High retry volume can significantly extend overall workflow duration.
Validating Regional Behavior
Verify Azure Service Health for the deployed region
Test the same workload in another Azure region, if feasible
1. Process the same invoices directly in Document Intelligence Studio or via a simple synchronous API call If processing completes quickly in these tests, the delay is likely related to workflow orchestration rather than the Document Intelligence service itself.
Reviewing document characteristics While less likely for this scenario, validation is recommended if delays persist:
- File size and format (PDF, PNG, JPG)
- Image resolution or DPI
- Presence of embedded images or malformed metadata

The following references might be helpful , please check them out

Thank you

Matias Haller 0 Reputation points

2026-05-18T13:30:41.3733333+00:00

Hello,

We try a new test and we sent PDFs files in 48 batches of 100 files each batch, one or two pages per file.

One of the batches took about 20 hours. Most of them took 9 hours to process.

We have tier S0, the size of the files are 30kb to 503kb.

Is there a way to have some one to check why is taking to long time to process?, if we need to change our support tier we can do that.

Regards
Matias

Share via

Slow response on Microsoft Foundry, Document Inteligent

2 answers

Your answer