Fix Microsoft 365 Document Processing Problems
Why This Is Happening
Picture this: your IT team spent two weeks planning a rollout of Microsoft 365 document processing across your SharePoint environment. You set up the Azure subscription, got approval from finance, enabled the services in the admin center , and then nothing works. Models don't apply. Autofill columns sit blank. The optical character recognition feature quietly does nothing. The error messages, when you get them at all, are either missing entirely or so generic they're useless.
I've seen this exact situation on dozens of tenant configurations, and the root cause is almost always one of three things: a broken pay-as-you-go billing link between your Microsoft 365 tenant and your Azure subscription, a permissions gap that blocks the service from writing results back to a SharePoint library, or a licensing transition issue left over from the old Microsoft Syntex per-user model.
Microsoft 365 document processing , which you may know from its previous branding as Microsoft Syntex, is a suite of AI-powered services that includes autofill columns, document translation, eSignature, optical character recognition, content assembly, image tagging, taxonomy tagging, prebuilt document processing models, structured and freeform models, and unstructured classification models. Every single one of these runs on a pay-as-you-go billing model tied directly to an Azure subscription. If that Azure link is misconfigured, broken, or pointing at the wrong subscription, none of the services activate correctly, and Microsoft's admin center won't always tell you why in plain English.
Another thing that trips people up: the old per-user licenses for these services are no longer available for new purchases. Microsoft ended that model. If your organization still has active per-user licenses, you can keep using them for tasks like applying unstructured models, running prebuilt and structured models, and using content assembly, but the moment those licenses expire, you're moving to pay-as-you-go whether you're ready or not. Many organizations hit that wall without realizing it, and suddenly their document processing workflows stop dead.
Government Community Cloud (GCC) tenants face a different problem: pay-as-you-go licensing isn't available yet for GCC environments at all. Those organizations need to stay on per-user licenses until Microsoft makes that transition available. If you're on GCC and seeing document processing failures, that's very likely your answer right there.
I know this is frustrating, especially when these workflows are tied to business-critical processes like invoice extraction, contract management, or compliance tagging. The good news: most of these problems are fixable once you know exactly where to look. Browse all Microsoft fix guides →
The Quick Fix, Try This First
Before you go anywhere near PowerShell or Event Viewer, do this one check first. The majority of Microsoft 365 document processing failures, I'd say at least 70% of the cases I've worked, come down to the pay-as-you-go billing setup either being incomplete or silently broken. Here's the fastest way to confirm and fix it.
Open the Microsoft 365 admin center at admin.microsoft.com. In the left navigation, go to Setup > Files and content. Look for the Document processing section. If you see a banner saying "Pay-as-you-go is not configured" or the services are listed as inactive, that's your culprit right there.
Click through to set up or verify pay-as-you-go billing. You'll need:
- An active Azure subscription (not a free trial, those don't qualify)
- Owner or Contributor rights on that Azure subscription
- Global Admin or SharePoint Admin rights in your Microsoft 365 tenant
If billing is already configured but services still aren't working, navigate to the SharePoint admin center at your tenant's admin.sharepoint.com. Go to Settings and look for Document processing. Check whether the specific service you're trying to use, say, autofill columns or optical character recognition, is toggled on at the tenant level. Many admins set up billing but forget to enable individual services here. They're off by default.
Once you flip those toggles on and give it about 15–30 minutes to propagate, test by going to a SharePoint document library, clicking the Automate menu in the command bar, and checking whether the document processing options now appear. If they do, you're done.
This is the foundation. Microsoft 365 document processing costs are metered through Azure, so the two systems need to be properly linked. Without this connection, every service just silently fails.
Sign in to the Microsoft 365 admin center (admin.microsoft.com) with a Global Admin account. Go to Setup in the left menu, then select Use your content AI features or search for "document processing" in the setup search bar. You'll land on the pay-as-you-go configuration page.
Click Connect an Azure subscription. In the panel that opens, you'll see a dropdown to select your Azure subscription. If your subscription doesn't appear in the list, it typically means one of two things: your Azure account isn't linked to the same organizational tenant as your Microsoft 365, or your Azure account doesn't have the right role. You need at minimum Contributor access on the Azure subscription to link it here.
Once you select the subscription, choose your Resource Group (or create a new one, naming it something like rg-m365-docprocessing-prod keeps things organized). Then select the Azure region closest to your users. Click Save.
To confirm the link worked, open the Azure portal at portal.azure.com, navigate to your selected resource group, and look for a new resource of type Syntex or SharePoint. Its presence there confirms the billing bridge is active. If no resource appears after 10 minutes, the connection failed silently and you'll need to try again, often switching to a different browser session or using an InPrivate window resolves authentication caching issues.
Linking your Azure subscription only turns on billing. Each individual Microsoft 365 document processing service still needs to be explicitly enabled by a SharePoint Admin. This is a separate step and one that catches a lot of people off guard.
Go to the SharePoint admin center, the URL is https://[yourtenant]-admin.sharepoint.com. In the left panel, click Settings. Scroll down until you see the Document processing section. You'll see a list of services:
- Autofill columns
- Document translation
- eSignature
- Optical character recognition
- Content assembly
- Image tagging
- Taxonomy tagging
- Prebuilt document processing
- Structured and freeform document processing
- Unstructured document processing
Each one has its own toggle. Enable the ones your organization needs. Don't just enable everything speculatively, each service has its own pay-as-you-go meter in Azure, so enabling services nobody uses means you'll start seeing unexpected line items on your Azure bill the moment someone accidentally triggers one.
After enabling, click Save at the bottom of the page. Again, give it 15–30 minutes. Then navigate to a SharePoint document library, click Automate in the top command bar, and verify the relevant options now appear. For autofill columns specifically, you'll also see a new option when you add or configure a column, look for Autofill as a column type in the column creation panel.
Autofill columns is one of the most-used Microsoft 365 document processing features, it uses large language models to automatically extract or generate metadata from uploaded documents. When it stops working, it usually manifests as columns that stay blank after files are uploaded, or as a spinning indicator that never resolves.
First, confirm the column is properly configured. In the SharePoint library, go to Library settings (gear icon > Library settings > More library settings). Click on the autofill column name. Verify that the Autofill toggle is on and that a valid prompt or extraction instruction is set. If the prompt field is empty or contains a vague instruction, the LLM has nothing to work from.
Next, check whether the service is actually processing files. Upload a test document (a one-page PDF or Word doc works well), then open the file's properties panel. If you see a processing spinner icon next to the autofill column that never clears, the job may be stuck in queue. This often happens when your Azure subscription hits a throttle limit. Check your Azure portal under Cost Management + Billing, if your subscription has a spending limit set and it's been reached, processing jobs queue up but never execute.
For stubborn cases, try removing the autofill column entirely and re-creating it from scratch with a cleaner, more specific prompt. Autofill columns that were configured before pay-as-you-go billing was properly linked sometimes get stuck in a broken state and don't self-heal even after billing is fixed. Deleting and recreating forces a fresh initialization.
# PowerShell: Check SharePoint service health status
Connect-SPOService -Url https://[tenant]-admin.sharepoint.com
Get-SPOTenant | Select-Object *Syntex*
OCR for Microsoft 365 document processing lets you pull printed or handwritten text out of images and image-based PDFs, making that content searchable within SharePoint. Document translation creates translated copies of files in a SharePoint library while keeping the original format intact. Both services fail in very similar ways, so I'm grouping them here.
The most common OCR failure is that the feature appears to run, you can trigger it, but the extracted text never appears in the document's search index. This is nearly always a permissions issue at the site collection level. The service account running the OCR job needs at least Contribute permissions on the target library. Go to Site settings > Site permissions and check whether the SharePoint App Principal service account has the necessary rights.
For document translation failures, the most common error code you'll see is a generic service timeout, especially for large files. Microsoft 365 document translation supports all the languages and dialects listed in the official documentation, but files over 40MB or documents with complex embedded objects (charts, embedded Excel tables, SmartArt) can fail silently. Break large files into smaller chunks before translating, or flatten complex objects to images first.
If neither service responds at all, no error, no output, just nothing, run this diagnostic in PowerShell:
Connect-MgGraph -Scopes "Sites.Read.All"
Get-MgSiteList -SiteId [your-site-id] | Select-Object DisplayName, WebUrl
This verifies that Graph API access to your site is functioning. If this command fails with a permissions error, your tenant's Microsoft Graph API permissions may need to be reconsented, go to Azure Active Directory > Enterprise applications > find the SharePoint app > Permissions > Grant admin consent.
The AI models at the heart of Microsoft 365 document processing, prebuilt models for invoices, receipts, and contracts; structured and freeform models for extracting fields from consistent-layout documents; and unstructured models for classifying varied content, are where things get genuinely complex when they break.
Prebuilt models for invoices and receipts failing to extract fields is almost always a document quality problem, not a service problem. The model expects reasonably clean input. Scanned documents at below 150 DPI, heavily watermarked files, or PDFs that are password-protected will produce incomplete or empty extractions. Run your source files through the OCR service first if they're image-based, then process them through the prebuilt model.
For structured and freeform models that worked fine before but are now producing errors or not applying to libraries, check whether the model was built in a content center site and whether that content center still has an active model configuration. Navigate to your content center site > Models > select the model > Where the model is applied. If the library shows as disconnected, remove the association and re-apply the model.
Unstructured models used for document classification sometimes stop classifying correctly after major changes to your document library's content type configuration. These models are tied to content types, so if someone modified or deleted a content type that the model was trained to assign, the model will apply successfully but output nothing. Go to Site settings > Content types and verify the target content type still exists and is active on the library.
# Verify content type is available on the library
# Run in SharePoint PnP PowerShell
Connect-PnPOnline -Url https://[tenant].sharepoint.com/sites/[contentcenter]
Get-PnPContentType -List "[YourLibraryName]"
Advanced Troubleshooting
Once you've worked through the standard fixes and things are still broken, it's time to get into the deeper layers. These are the issues that usually show up in larger organizations, domain-joined machines, custom Power Platform environments, conditional access policies, and Azure cost controls getting in the way.
Conditional Access Policies Blocking the Service
Microsoft 365 document processing services authenticate through Azure Active Directory. If your organization has Conditional Access policies that require compliant devices or specific network locations, the background service processes that run document AI jobs may be getting blocked without any visible error surfacing to end users. Check your Azure AD Sign-in logs (Azure portal > Azure Active Directory > Sign-in logs) and filter for the application name Office 365 SharePoint Online and Failure status. Look for Conditional Access as the failure reason.
Custom Power Platform Environments for Structured or Freeform Models
If your organization processes structured or freeform document models through a custom Power Platform environment rather than the default one, there's a separate setup path required. The standard pay-as-you-go billing setup in the Microsoft 365 admin center doesn't automatically extend to custom Power Platform environments. You need to configure that environment specifically by going to Power Platform admin center > select the environment > Settings > Features > and enabling document AI capabilities there separately.
Azure Spending Limits and Cost Alerts
This one burns people regularly. Azure free and trial subscriptions have spending limits that, when hit, pause all metered services, including document processing. Visual Studio subscriptions have the same behavior. The symptom is that everything works fine for the first part of the month and then stops completely without warning. Check in Azure portal under Cost Management + Billing > Subscriptions > select yours > Spending limit. If you see a spending limit is active, either remove it (for production use) or switch to a Pay-As-You-Go Azure subscription type.
Event Viewer and ULS Log Analysis
For on-premises SharePoint hybrid setups, Windows Event Viewer on the SharePoint server (Event ID 6398 in the Application log is the classic SharePoint timer job failure event) can point to document processing job failures. For SharePoint Online, ULS logs aren't directly accessible, but the Microsoft 365 admin center under Health > Service health will show active incidents for SharePoint Online that may explain widespread failures.
Tenant-Level Service Throttling
High-volume tenants that upload thousands of documents in a short window can hit service throttling on the document processing side. Microsoft's document processing services are designed for steady-state usage, not burst imports. If you're doing a migration or bulk upload, stage your content in batches of no more than 500–1000 documents per hour and monitor your Azure billing meters to watch for processing queues backing up.
Prevention & Best Practices
After fixing a Microsoft 365 document processing issue, the last thing you want is to hit the same wall six months from now. Here's what I recommend to every organization I work with after getting their environment back online.
Set up Azure cost alerts the day you link your subscription. Go to Azure portal > Cost Management + Billing > Cost alerts > Add. Create an alert at 50%, 80%, and 100% of your monthly document processing budget. You want to know about spend trends before they become a problem, not after the service stops. This is especially important if you've enabled image tagging or taxonomy tagging at scale, those meters can accumulate faster than expected when applied to large libraries.
Document which services are active and why. It sounds obvious, but in enterprise environments with multiple SharePoint admins, it's incredibly common for someone to enable a service for a test, forget about it, and leave it running against a production library. Keep a simple SharePoint list or OneNote page that logs: which services are on, which libraries they're applied to, who enabled them, and the business justification. This makes troubleshooting dramatically faster.
When using autofill columns across document translation workflows, for example, extracting metadata from translated copies, set up processing rules to control the order of operations. Document translation creates a copy first; autofill columns should run on that copy, not the source file. Getting the sequence wrong means you're running AI extraction on documents in the wrong language, which produces garbage output.
Plan for the AI Builder credits end-of-life. Microsoft announced in October 2025 that AI Builder credits are being phased out progressively. If any of your document processing workflows depended on AI Builder capacity, those need to be migrated to the direct pay-as-you-go meters for the specific services used. Audit your Power Automate flows and Power Apps for any AI Builder actions that overlap with document processing tasks.
- Use the SharePoint cost calculator before rolling out document processing at scale, it gives you realistic monthly cost estimates based on your actual document volumes
- Enable only the document processing services your organization will actually use in the first 90 days; add more once you understand your usage patterns
- Create a dedicated content center site for model management, mixing model administration with regular team sites causes permission conflicts and makes governance harder
- Set a calendar reminder every 90 days to review your Azure billing meters for document processing services, catch silent cost creep before it becomes a finance conversation
Frequently Asked Questions
My Microsoft 365 document processing models stopped working after our Azure subscription was changed, how do I reconnect it?
When an Azure subscription is cancelled, transferred, or replaced, the billing link to your Microsoft 365 tenant breaks, and document processing services stop silently. Go to the Microsoft 365 admin center > Setup > document processing configuration, and reconnect your new Azure subscription using the same steps as the original setup. If your old subscription is still active in a transitional state, remove it first before linking the new one, otherwise you'll see a conflict error. Give it 30 minutes to propagate after saving, then re-test a document library.
Can regular users set up autofill columns, or does it need an admin?
Regular users with at least Contribute permissions on a SharePoint library can add and configure autofill columns themselves, they don't need to be a SharePoint Admin. The admin setup is only required once at the tenant level to enable the service and link billing. After that, column creation is self-service. The only exception is if your tenant admin has specifically restricted who can create columns via SharePoint site-level settings or a custom permission policy.
Is Microsoft 365 document processing available for GCC tenants?
Not yet for the pay-as-you-go model. As of April 2026, Government Community Cloud (GCC) organizations cannot use pay-as-you-go billing for document processing services. GCC tenants can continue purchasing and using per-user licenses until Microsoft makes pay-as-you-go available in that environment. Keep an eye on the Microsoft 365 roadmap at microsoft.com/microsoft-365/roadmap for updates on GCC availability.
What happened to Microsoft Syntex, is it the same as document processing?
Yes, exactly the same services. Microsoft rebranded Microsoft Syntex as "document processing for Microsoft 365", the features, functionality, and underlying technology are unchanged. If you have documentation, training materials, or IT runbooks that reference Syntex, they're still accurate. The rename was purely cosmetic from a technical standpoint, though the licensing transition from per-user to pay-as-you-go happened around the same time, which is where the real operational changes came from.
Why is optical character recognition not finding text in my uploaded images?
OCR quality drops sharply with low-resolution source files. The service performs best on images at 150 DPI or higher, with good contrast and minimal skew. Heavily compressed JPEGs, smartphone photos taken at an angle, or images with dense watermarks all produce incomplete extractions. Try re-scanning your source document at 200–300 DPI and re-uploading. Also check that the library has OCR enabled specifically, it doesn't inherit the tenant-level setting automatically in all configurations. Go to Library settings > Advanced settings and verify OCR is turned on for that library.
Do I get charged when a document processing job fails?
Generally, no, Microsoft's pay-as-you-go meters for document processing bill per successful transaction, not per attempt. However, the exact billing behavior varies by service. For example, optical character recognition bills per page of successfully processed content. If a job errors out before completing, you typically won't see a charge for that failed run. That said, jobs that partially complete, say, OCR that processes 8 of 10 pages before timing out, may bill for the pages that were processed. Check the specific pricing page for each service to understand the exact meter unit for that capability.