Fix Microsoft 365 Document Processing: Full Setup Guide
Why Microsoft 365 Document Processing Breaks (And Why the Error Messages Are Useless)
You're trying to get document processing working in Microsoft 365 , maybe you want OCR on scanned invoices, or you're rolling out autofill columns across a SharePoint library , and nothing works. The portal shows a greyed-out button, or an activation error, or it just silently does nothing when you apply a model. Sound familiar?
I've seen this exact situation on dozens of tenant deployments. The frustrating part isn't the technology, the underlying AI is genuinely good. It's the setup chain. Microsoft 365 document processing (previously branded as Microsoft Syntex) is a pay-as-you-go service that runs through your Azure subscription, and if any link in that chain is broken, the entire service stalls out without giving you a clear explanation of why.
Here's what's actually going on under the hood. Document processing services aren't just a Microsoft 365 feature you flip on in the admin center. They're Azure-backed, metered services billed through an active Azure subscription that you have to explicitly link to your Microsoft 365 tenant. That linking step, setting up pay-as-you-go billing, is the single most common place things go wrong. If you skip it, or if the Azure subscription isn't in the right state, every service from optical character recognition to eSignature will silently fail or remain unavailable in the UI.
The other major source of confusion right now is the licensing transition. Per-user licenses for these services are no longer available for new purchases. Microsoft announced the end of AI Builder credits in October 2025. Organizations that built workflows on AI Builder credits are now being migrated onto the pay-as-you-go model, and if your tenant hasn't been properly transitioned, you'll see features that appear available but refuse to process anything. Your tenant is essentially sitting in a licensing no-man's-land.
There's also a permissions layer that catches people off guard. SharePoint site collection administrator rights aren't enough on their own. The Microsoft 365 global admin or SharePoint admin has to enable the pay-as-you-go setup at the tenant level, and individual site owners need specific rights to apply models to document libraries. If your IT admin set up billing but didn't configure site-level permissions, you'll spend hours wondering why the "Apply model" button never appears.
Government Community Cloud (GCC) organizations have an additional complication: pay-as-you-go licensing isn't yet available for GCC tenants. If you're in a GCC environment, you can still use per-user licenses, but you can't access the new pay-as-you-go services until Microsoft rolls them out to that cloud environment.
I know this is frustrating, especially when it blocks real business work like invoice processing or contract classification. The good news is every single one of these issues is fixable once you know where to look. Browse all Microsoft fix guides →
The Quick Fix, Try This First
Before going through the full setup, do this check first. It resolves roughly 60% of Microsoft 365 document processing failures I encounter in the field.
Open the Microsoft 365 admin center at admin.microsoft.com. In the left navigation, go to Settings > Org settings. Scroll down and click SharePoint. On that page, look for a section called Document processing or Syntex (the label varies slightly depending on when your tenant was last updated). If you see a banner that says "Set up pay-as-you-go billing" or "Billing not configured," that's your answer right there.
Click that banner and walk through the wizard. You'll be asked to:
- Select an active Azure subscription (it must be in an active, not suspended, state)
- Choose a resource group within that subscription
- Select the Azure region closest to your SharePoint tenant's data residency location
Once that wizard completes successfully, go back to a SharePoint document library, click the Automate menu in the toolbar, and check whether document processing options now appear. On most tenants, services like optical character recognition and autofill columns become available within 5 to 10 minutes of completing billing setup.
If the wizard errors out during subscription selection, the most common cause is that your account doesn't have Owner or Contributor role on the Azure subscription. Have your Azure admin check your RBAC assignment in the Azure portal under Subscriptions > [Your Subscription] > Access control (IAM).
This is the foundation. Without a properly linked Azure subscription, no Microsoft 365 document processing service will work, full stop.
Sign in to the Microsoft 365 admin center with a Global Administrator account. Navigate to Settings > Org settings > Services tab, then find and click Document processing (Microsoft 365). You'll see a setup page with a prominent call-to-action to configure billing.
Click Set up pay-as-you-go billing. The wizard opens and asks you to pick:
- Azure subscription: Must be active and in good standing. Suspended subscriptions caused by missed payments will block setup.
- Resource group: You can create a new one named something descriptive like
rg-m365-docprocessing-prod. - Region: Choose the Azure region that matches or is nearest to your Microsoft 365 data center region. Mismatched regions can increase latency on OCR and model processing jobs.
After clicking Save, confirm the setup completed by returning to the same settings page. You should see a green status indicator and your Azure subscription ID listed. If you instead see error code BillingSetupFailed or AzureSubscriptionNotFound, verify in the Azure portal that your account has at minimum the Contributor role on that subscription via Subscriptions > Access control (IAM) > Role assignments.
A successful setup means you'll see the Azure subscription name displayed alongside a billing region. That's your confirmation that metered billing is live and services can now be activated.
Linking your Azure subscription doesn't automatically turn on every service. Each document processing capability, OCR, autofill columns, document translation, eSignature, image tagging, taxonomy tagging, and the various model types, needs to be individually enabled. This is by design so you only get billed for what you actually use.
Back in the Microsoft 365 admin center, go to Settings > Org settings > Document processing. You'll now see a checklist of available services with toggles. Enable only what your organization currently needs:
- Optical character recognition, for extracting text from scanned images and PDFs
- Autofill columns, for LLM-powered metadata extraction directly into SharePoint columns
- Document translation, to generate translated copies of files while preserving original format
- eSignature, for sending signing requests that keep documents inside Microsoft 365
- Prebuilt models, for invoices, receipts, contracts, and other common document types
- Unstructured models, for classifying documents that vary widely in layout and composition
After enabling a service, allow up to 15 minutes for it to propagate to SharePoint sites. A common mistake is toggling a service on, immediately navigating to a document library, not seeing the option, and assuming it failed, when it simply hasn't provisioned yet. Test on a specific site by going to a document library, clicking Automate in the command bar, and confirming the relevant options appear in the dropdown.
If a service is enabled at the tenant level but still not showing in a specific library, the next step is checking that the site collection itself has document processing features turned on, covered in Step 3.
Even with tenant-level billing and services enabled, individual SharePoint sites can block document processing features if the site collection admin hasn't activated the required features. This is one of the most overlooked steps in a Microsoft 365 document processing deployment.
Navigate to the SharePoint site where you want to use document processing. Go to Site settings (the gear icon > Site information > View all site settings). Under Site Collection Administration, click Site collection features. Look for Content center and Document processing features and make sure they're both set to Active. If they're deactivated, click Activate next to each one.
Next, ensure the user who will be applying models or configuring autofill columns has at minimum Site Member rights, but for model management, they'll need Site Owner or a custom permission level that includes the Manage Lists permission.
For organizations managing this across many sites, use PowerShell to activate features at scale rather than clicking through each site manually:
Connect-SPOService -Url https://yourtenant-admin.sharepoint.com
$sites = Get-SPOSite -Limit All
foreach ($site in $sites) {
Enable-SPOFeature -Identity "73EF14B1-13A9-416B-A9B5-ECECA2B0604C" `
-Url $site.Url -Force
}The GUID above corresponds to the document processing content center feature. Run this script with an account that has SharePoint admin rights. After execution, verify by spot-checking two or three sites to confirm the feature shows as Active.
Once site features are active, go back to any document library, click Automate, and you should see options like Apply a classification model, Set up autofill columns, and any other services you enabled at the tenant level.
OCR is the most-used document processing service for organizations migrating scanned paper records into SharePoint. When it's working correctly, it extracts printed and handwritten text from images and image-based PDFs, making them fully searchable. When it's not configured right, files remain unsearchable blobs.
To apply OCR to a document library, navigate to the library in SharePoint. Click Automate in the command bar, then select Set up optical character recognition. A configuration panel opens on the right side of the screen. Choose whether to process only new files going forward, or to also run OCR on existing files in the library retroactively (be cautious with the retroactive option on large libraries, each file consumes pay-as-you-go capacity).
After enabling OCR, upload a test image or scanned PDF. Within a few minutes, the file should have searchable text content. To verify, open the file in SharePoint's document viewer and attempt to select text, if OCR ran successfully, text will be selectable. You can also search for a word you know appears in the document using the SharePoint site search, and the file should appear in results.
If OCR runs but produces garbled or missing text, check the source file quality. Scans below 150 DPI produce poor extraction results. The ideal scan resolution for OCR accuracy is 300 DPI or higher. For handwritten text, accuracy varies significantly with handwriting legibility, Microsoft's OCR engine is strong, but heavily stylized cursive remains a challenge.
A processing error where OCR shows as "failed" in the library's column view usually means the file format isn't supported. OCR works on JPEG, PNG, TIFF, BMP, PDF, and a handful of other formats. Word documents and Excel files don't need OCR since they already contain selectable text.
Autofill columns are where Microsoft 365 document processing gets genuinely powerful, and where configuration errors most often produce wrong or empty results. This service uses large language models to read document content and automatically populate SharePoint metadata columns. When it works, it eliminates hours of manual metadata tagging. When it's misconfigured, it fills columns with confidently wrong data.
To set up autofill columns, go to a document library and click Automate > Set up autofill columns. Select which existing columns in the library you want to auto-populate. For each column, you'll write a prompt that tells the LLM what to extract, for example, for an "Invoice Total" column you might write: "Extract the total amount due from this invoice. Return only the numeric value with currency symbol."
The quality of your prompt directly determines the quality of the extracted data. Vague prompts like "get the price" produce inconsistent results. Specific prompts with output format instructions produce reliable data. Here are prompt patterns that consistently work well in my experience:
Column: Contract Start Date
Prompt: "What is the contract start date? Return in MM/DD/YYYY format only."
Column: Vendor Name
Prompt: "Who is the vendor or supplier named in this document? Return only the company name, no other text."
Column: Document Language
Prompt: "What language is this document primarily written in? Return the language name in English only."After saving the configuration, upload a test document and wait 2–5 minutes for autofill to process. Check the column values. If columns remain empty, the most common cause is that the document library's autofill setting is set to process only documents uploaded after the configuration date. Existing documents need to be manually triggered. Select the file, click Automate, and choose Run autofill now to process it on demand.
If column values are consistently wrong across multiple documents, revisit and refine your prompts. Iterating on prompt wording is the fastest path to accurate extraction, treat it like writing a job description for the AI.
Advanced Troubleshooting for Enterprise and Domain-Joined Environments
If you've worked through all five steps and things still aren't right, you're likely dealing with an enterprise-level configuration issue. Here's how to diagnose the most common ones.
Azure Billing Meters Not Showing Up in Cost Analysis
After setting up pay-as-you-go billing, document processing usage should appear as line items in your Azure Cost Management portal under the resource group you specified. If you're using services but seeing zero charges and zero usage in Azure Cost Management, the billing linkage may not have completed correctly despite showing as configured in the Microsoft 365 admin center. The fix: go back to Settings > Org settings > Document processing in the admin center, click on the billing configuration, and verify that the Azure Subscription ID shown matches exactly what you see in the Azure portal. A mismatch, which can happen if someone set up billing with a different admin account that has access to a different Azure subscription, means you're billing to the wrong subscription or billing isn't actually flowing.
Conditional Access Policies Blocking Model Processing
In organizations with strict Azure AD Conditional Access policies, service principals used by document processing services may be blocked by policies requiring MFA or compliant devices. These aren't user accounts, so they'll silently fail rather than producing an MFA prompt. Check your Azure AD sign-in logs (Azure Active Directory > Monitoring > Sign-in logs) and filter by Application, searching for Microsoft Syntex or SharePoint service principals. If you see blocked sign-in events, create a Conditional Access exclusion for the Microsoft 365 document processing service principal identities.
GCC Tenant Limitations
If your organization is on Government Community Cloud, you cannot use pay-as-you-go services. Any per-user licenses you have remain usable, you can still assign them to new users and run unstructured models, prebuilt models, structured and freeform models, content assembly, content query, annotations, and document library automation rules. But newer pay-as-you-go-only features like the expanded autofill columns with LLM prompting won't be available until Microsoft extends those services to GCC. Check Microsoft's GCC roadmap through the Microsoft 365 admin center's Message Center for timeline updates specific to your cloud environment.
Transition from AI Builder Credits
If your document processing workflows were previously powered by AI Builder credits, note that Microsoft announced the end of AI Builder credits in October 2025. Flows and models that consumed AI Builder credits need to be evaluated and either migrated to pay-as-you-go document processing services or rebuilt using alternative approaches. Check Power Automate flows that reference AI Builder actions, any that used AI Builder document processing models may need to be updated to call the newer document processing services directly via SharePoint automation or the Microsoft Graph API.
Event Log and ULS Log Analysis
For SharePoint on-premises administrators who've integrated with Microsoft 365 hybrid document processing, ULS (Unified Logging Service) logs are your best diagnostic tool. On the SharePoint server, use the SharePoint Management Shell to pull relevant log entries:
Merge-SPLogFile -Path "C:\Logs\DocProcessing_$(Get-Date -Format yyyyMMdd).log" `
-StartTime (Get-Date).AddHours(-2) `
-EndTime (Get-Date) `
-EventID "Document Processing"Look for correlation IDs associated with failed model runs. Those correlation IDs can be cross-referenced against the Microsoft 365 admin center's Health > Service health page to determine if the issue is tenant-wide or isolated.
Prevention & Best Practices for Microsoft 365 Document Processing
Once you've got document processing working, keeping it working, and keeping costs predictable, comes down to a handful of operational habits. I've seen well-intentioned deployments go sideways six months down the line purely because of missing governance around these services.
First, monitor your Azure costs from day one. Set up a cost alert in Azure Cost Management specifically for the resource group you created for document processing. A reasonable starting alert is at 80% of your monthly budget. Navigate to Azure portal > Cost Management + Billing > Cost alerts and create a budget with email notifications. It takes three minutes and saves serious pain later.
Second, be selective about retroactive processing. Autofill columns and OCR can be applied retroactively to entire document libraries, but each file processed consumes pay-as-you-go capacity. For large libraries with thousands of historical documents, do a pilot on a subset, say, documents uploaded in the last 90 days, before enabling retroactive processing across the whole library. Use the SharePoint cost calculator to estimate what full retroactive processing will cost before you click confirm.
Third, document your prompt library. If you're using autofill columns across multiple libraries and multiple column types, maintain a shared OneNote or SharePoint wiki page that records which prompts are in use, which version performed best during testing, and who owns each configuration. When prompts need to be updated because document formats change, you'll thank yourself for having this record.
Fourth, review your per-user license status before they expire. Per-user licenses can still be assigned to new users if you have remaining licenses, and they let you continue using unstructured models, prebuilt models, content assembly, and other services without pay-as-you-go metering for those specific users. Once they expire, you'll need pay-as-you-go for everything. Plan the transition so there's no gap in service for users who depend on document processing in their daily work.
- Set an Azure cost alert at 80% of your expected monthly document processing budget before enabling any services
- Create a dedicated resource group specifically for document processing so its costs are easy to isolate in Azure Cost Management
- Test autofill column prompts on 10–20 sample documents before applying the configuration tenant-wide, iterate until accuracy is above 90%
- Audit your Conditional Access policies for service principal exclusions before rolling out to production, to avoid silent processing failures in secured tenants
Frequently Asked Questions
Why is the "Apply a model" button greyed out in my SharePoint document library?
This almost always means one of three things: pay-as-you-go billing hasn't been set up in the Microsoft 365 admin center, the document processing feature hasn't been activated at the site collection level, or your account doesn't have sufficient permissions on the site. Start by checking Settings > Org settings > Document processing in the admin center to confirm billing is configured. Then go to Site settings > Site collection features and make sure document processing features are active. If both of those check out, confirm you have Site Owner rights on the site, Site Member rights aren't enough to manage document processing models.
How much does Microsoft 365 document processing cost per month?
It's genuinely pay-as-you-go, you only pay for capacity you actually consume, with no minimum monthly fee. Each service has its own pricing meter in Azure, and costs are billed through your Azure subscription. For accurate current pricing, use the SharePoint cost calculator available from the document processing settings page in the Microsoft 365 admin center. It factors in your organization's usage patterns and document volumes to produce a realistic monthly estimate. Microsoft also offered free trial capacity through December 2025 for organizations with pay-as-you-go billing already configured, covering selected services.
We're on GCC, can we use the new document processing AI features?
Not yet. Pay-as-you-go licensing and the services that depend on it aren't available for Government Community Cloud organizations. GCC tenants can continue using per-user licenses to run unstructured models, prebuilt models, structured and freeform models, content assembly, taxonomy services, content query, annotations, and document library automation rules. The pay-as-you-go model is expected to come to GCC at a future date, check the Microsoft 365 Message Center in your admin center for GCC-specific roadmap announcements.
Autofill columns keep returning empty or wrong values, what's going wrong?
Empty values usually mean the document type or layout doesn't match what the prompt is asking for, or the file isn't being processed at all yet (wait 2–5 minutes or trigger it manually via Automate > Run autofill now). Wrong values are almost always a prompt quality issue. Vague prompts produce inconsistent LLM outputs. Rewrite your prompt to include the exact output format you expect, for example, specify "return only the date in MM/DD/YYYY format" instead of just "what is the date." Test your revised prompt on 10 to 15 representative documents before saving the configuration to the full library.
What happened to Microsoft Syntex, is it the same thing as document processing?
Yes, exactly the same product, just rebranded. Microsoft renamed the pay-as-you-go services previously offered under the Syntex name to "document processing for Microsoft 365." The features, functionality, and underlying technology remain completely unchanged, only the name has been updated in documentation and the admin center. If you built workflows, documentation, or training materials that reference Microsoft Syntex, everything still applies. You may see a mix of both names in older SharePoint UI elements and documentation depending on when your tenant was last updated.
Can I still use my existing per-user licenses or do I have to switch to pay-as-you-go?
You can continue using active per-user licenses, they can even be assigned to new users as long as licenses remain in your tenant. What you can't do is buy new per-user licenses, as they're no longer available for purchase. Once your existing per-user licenses expire, you'll need to switch to pay-as-you-go billing to continue using document processing services. With an active per-user license, you can apply unstructured models to libraries, create and run prebuilt, structured, and freeform models, use content assembly and taxonomy services, and work with content query and annotations, exactly as before.