Gokan's Studio

SharePoint “Premium” (formerly Microsoft Syntex) has been unbranded as Document Processing – but under the covers, the AI-driven classification and metadata extraction features remain the same. It’s just the same story repeated. In other words: Project Cortex, SharePoint Syntex, Microsoft Syntex, SharePoint Premium, and Content AI are dead, long live Document Processing. These tools are now part of Microsoft 365 Content AI and, somehow, Power Platform, enabling organizations to automatically read, classify, and tag documents of all kinds.

Why this article? Why is the SharePoint team so confusing (the million-dollar question)? Because with announcements and rebranding, it’s hard to keep track. One month, you’re exploring Syntex templates; the next, Microsoft launches Copilot in Teams; then you hear about a new Knowledge Agent preview in SharePoint. No one expects you to know all the details by heart. What you need is guidance on where to focus: which tool fits each scenario, how to get started quickly, and how to avoid wasteful trial-and-error.

That’s why I wrote this guide. Over the past few years, I’ve worked with many organizations on document automation projects. Time and again, I saw teams struggle: building a proof-of-concept with AI Builder only to realize Syntex would have been faster, or vice versa; or simply not using a tool because they didn’t realize it existed. I also know your schedules are packed, and you need clear answers.

In the chapters ahead, we’ll break down these options in plain terms and real scenarios. You’ll find:

• Side-by-side comparisons of Syntex, AI Builder, Copilot, Knowledge Agent, etc., showing which fits everyday tasks (e.g., invoice processing, contract management, content Q&A).
• Practical examples and tips from real deployments (so you don’t reinvent the wheel).
• Guidance on licensing and cost considerations (which features have free tiers, which need add-ons).
• Pointers to the latest updates, sprinkled through the text, so you’re up to date on new capabilities.

My goal is to save you the frustration of figuring this out by trial and error. By the end of this introduction, I hope you feel that you won’t have to navigate this alone. We’ll cut through the jargon and focus on what you need to know. Think of this guide as your map through the new world of AI document processing – friendly, practical, and ready to help you make the most of these time-saving tools. More detailed articles will follow up! Ready? Let’s go (gotta catch them all)!

Unstructured Document Processing

Unstructured document processing (again, the old name) uses AI to classify and extract information from Single Class documents (e.g. contracts, reports, letters, in Word, PDF, PowerPoint, and many other document types). You train a model by showing example files (for instance, sample invoices or contracts) so it learns to recognize document types and pull out key fields (dates, customer names, amounts, etc.) In practice, you create or apply an unstructured processing model in a SharePoint Content Center. When you use it in a library, the model is linked to that library, and every document is automatically analyzed.

The most challenging SharePoint AI model is the Single class, but it is also the most powerful because you have complete control over creating, managing, and extracting data.

You need to add classifiers and extractors to your unstructured document processing models to perform the following actions:

• Classifiers are used to identify and classify documents that are uploaded to the document library. For example, a classifier can be “trained” to identify all contract renewal documents that are uploaded to the library. The contract renewal content type is defined by you when you create your classifier.
• Extractors pull information from these documents. For example, for each contract renewal document identified in your document library, columns displaying the Service Start Date and Client are provided for each document.

Use example files to train and test your classifiers and extractors. These files help the model learn what to look for when identifying and extracting data. For example, train your contract renewal model using real contract renewal documents from your organization. You can also use these files to validate your model’s accuracy. After publishing your model, use the content center to apply it to any SharePoint document library that you have access to.

• Advantages: Unstructured models work on any document layout, needing only example files to learn from. They also integrate with SharePoint search and compliance, making documents easier to find and govern.
• Drawbacks: These models require a training phase and some validation. Early outputs may need review, especially if the sample set is small. Accuracy depends on the quality and variety of your example documents. Setting up a model can take time, so it’s best for high-volume or mission-critical libraries. Also, it only works on document libraries (not lists) and requires documents to contain searchable text (image-only PDFs need OCR first).
• Licensing: As of 2025, Microsoft has shifted to a pay-as-you-go model: content AI features are enabled via Azure and billed per action (for example, auto-tagging calls are about $0.005 each). Existing Syntex licenses continue to work until they expire; after that, all usage goes through the consumption model.

Use Cases:

• My Finance (BE) and Manufacturing (NL) Customer: Automatically process thousands of invoices, contracts, receipts, or tax forms. For example, train a model on a few sample invoices, and then it will extract Invoice No., Date, Vendor, and Total from every new invoice PDF. This eliminates manual data entry and speeds up accounts payable and auditing.
• Military Customer (EU):  Organize publicly classified and diverse forms and reports. For instance, classify incoming supply orders vs. mission briefings, and auto-extract fields like unit ID, equipment name, or mission date. This makes it easy to track and retrieve essential documents from extensive archives.

Future: It’s not dead! I expect support for more document types, languages, and more intelligent AI. The core idea – automated content understanding in SharePoint – will only get more powerful and broadly used.

When not to use: If your documents are very uniform fixed forms, a structured form approach (below) might be more straightforward. If you only have a handful of files, manual entry or a simple flow could suffice. And if your data is ultra-sensitive, you may prefer manual tagging for anything mission-critical.

AI Builder Document Processing (Structured and Freeform)

Microsoft’s AI Builder (part of Power Platform) lets you build custom document-processing models. There are two types: Structured (fixed template) and Freeform. In a structured model, you tag fields on a few example forms (e.g., highlight the “Invoice Total” on several bills). AI Builder then applies that template to new forms of the same layout. In a freeform model, you label fields on a variety of documents that aren’t identical, and the AI learns to find those fields in different layouts.

When you apply an AI Builder model to a SharePoint library (via the Classify and Extract menu) or invoke it in Power Automate/Apps, every document added is processed by the model. The extracted data is appended as new metadata columns in the library. Behind the scenes, AI Builder stores training data in Dataverse, but the key point is that new files automatically get parsed and key data pulled out for you.

• Advantages: Structured models can be highly accurate for standardized forms (e.g., tax forms, invoices). Freeform models add flexibility for documents that change layout. These models integrate directly into Power Automate and Power Apps, enabling end-to-end workflows (for example, trigger on email arrival, form processing, then update a database or start an approval). Microsoft even provides some prebuilt models (e.g., for standard invoice or receipt formats) as templates to accelerate deployment.
• Drawbacks: Building these models requires sample documents and manual labeling, which takes effort. Processing happens in Microsoft’s cloud (Dataverse), so there’s some latency, and it consumes AI Builder capacity credits. If your documents are long narrative texts, these models will struggle more than Syntex’s unstructured approach. For very small or one-time projects, setting this up might not be worth the overhead.
• Licensing: AI Builder consumes capacity units. A Power Apps or Power Automate plan includes a base number of credits (typically 5,000 per user/month). For heavier use, you can purchase an AI Builder Capacity add-on (T1) for around $500 per month, which includes 100,000 credits. Each model training or document processed consumes credits based on file size and actions.

Use Cases:

• My Finance (BE) and Manufacturing (NL) Customer: Build a structured model for vendor invoices or receipts. The model can automatically extract fields like Vendor Name, Invoice Number, Date, and Amount from every uploaded invoice PDF. The extracted values become metadata columns, triggering downstream automation (e.g., approval flows).
• Military Customer (EU): Process standardized forms like travel vouchers or supply requisitions. For example, an expense form has fixed fields (soldier name, date, amount); an AI Builder model can pull these into columns for integration into the finance system.

Future: Microsoft will likely add more prebuilt models for standard documents (tax forms, permits, etc.) and improve the user experience (e.g., better auto-detection of fields during training).

When not to use: If your documents are long, unstructured narratives (choose Syntex/unstructured instead) or if you can’t allocate time to create training data, this may not be ideal. Also, if you only need raw text (no structure), a simple OCR tool might suffice.

Knowledge Agent (Autofill V2) in SharePoint

The SharePoint Knowledge Agent is a new AI assistant built right into SharePoint to help automate metadata and site management. When enabled (in Public Preview as of 2025), site owners see a floating “AI Actions” button in libraries and pages.

It offers context-sensitive tasks like “Organize this library,” “Ask a question,” or “Summarize”, adapting to your role (owner, contributor, etc.)

Figure: The SharePoint Knowledge Agent adds an AI menu to a document library for tasks like tagging and organizing.

A significant feature is AI-driven Autofill columns. The agent can suggest adding metadata columns (for example, “Document Type”, “Received Date”, “Department”) and then automatically populate them. In one demo, the agent proposed three columns and then filled them in: Document Type, Date, and Assigned Staff. Behind the scenes, SharePoint then auto-fills these columns for new and existing documents using advanced AI models.

• Advantages: This brings AI to business users with zero coding. It can quickly organize large libraries by adding metadata you didn’t have time to track. The Knowledge Agent also flags issues: it can highlight outdated pages, fix broken links on your site, and even generate content summaries. All of this makes your intranet more searchable and Copilot-ready. Because it uses natural language and AI, it might suggest metadata that manual tagging has overlooked. The biggest news of all: it now supports Managed Metadata Service!
• Drawbacks: The Knowledge Agent is still in preview, so expect its capabilities and accuracy to evolve. Currently, it only works on libraries (not on lists or pages). Because it’s AI-driven, it can make mistakes, so be sure to review any new tags it creates. Crucially, it requires a Microsoft 365 Copilot license to use: if you don’t have Copilot enabled, the agent won’t appear.
• Licensing: The Knowledge Agent is included with Microsoft 365 Copilot. Copilot is a paid add-on (roughly $30/user/month with an annual commitment) on top of an eligible M365 plan. Each user who triggers the agent or uses Copilot in SharePoint must have that license. From March 2025, SharePoint autofill columns are priced at $0.005 per transaction if using PAYG (without KA)

Future: When generally available, I expect more capabilities (support for list items, improved AI suggestions, extended admin controls). Microsoft will also likely connect this intelligence to more content (Teams files, OneDrive, etc.).

When not to use: If your library is already well-organized or if you lack the Copilot license, there’s no benefit. Also, if your content is extremely sensitive, you may prefer manual tagging for critical fields.

AI Prompts and AI Models

Microsoft also offers AI prompts and AI Models to speed up development. In Power Automate or Power Apps (Teams or web), you can choose from the AI HUB templates for common scenarios. For example, there are templates to process an invoice, analyse customer feedback sentiment, or detect forms in images. You pick a template and fill in a few parameters, and Power Automate generates the flow for you.

AI models, as shown hereunder, are very similar to the Freeform and Structured AI models.

The AI Prompt is quite different but very easy to understand and use, as it incorporates custom AI Actions in generative language.

AI Models | AI models are prebuilt or custom machine learning models that analyze data and return structured predictions or results. Examples:

• Form Processing (AI Builder) → extracts data from invoices or receipts.
• Prediction Model → predicts customer churn or product demand.
• Object Detection → recognizes items in images.
• Category Classification → classifies text (e.g., “complaint”, “praise”, etc.).

In short, AI models are data-driven and focus on structured prediction or extraction.

AI Prompts | AI prompts are instructions or natural language queries that tell a large language model (LLM) (like GPT) what to do — they don’t require training.

They’re used mainly in:

• Copilot in Power Automate or Power Apps
• AI Prompt capability in Power Automate Desktop
• Copilot Studio (formerly Power Virtual Agents)

Examples:

• “Summarize this email thread and extract the key actions.”
• “Write a friendly response to this customer complaint.”
• “Generate an introduction paragraph for this report.”
• “Turn this JSON into a readable summary.”

Similarly, Microsoft 365 Copilot offers built-in prompts and actions. In SharePoint or OneDrive, you can select files and click the Copilot button to “Summarize” them or “Compare” them. In a Copilot chat, you can ask natural questions like “What are the main points of this contract?” or “Show me the sales figures from these financial reports.” Copilot will search your accessible SharePoint content and summarize the answer from the documents.

• Advantages: Templates and prompts save time and lower the bar for non-technical users. You get a working solution almost instantly (especially for standard use-cases). Copilot prompts let you use AI conversationally, without knowing which models are running.
• Drawbacks: They are generic. A template might not fit your exact data fields, and you may need to tweak it afterward. Copilot’s answers depend on content quality: if files aren’t properly indexed or permissions are missing, results may be incomplete. Also, using Copilot prompts requires the Copilot license and that the files be stored in eligible apps (e.g., SharePoint or OneDrive).
• Licensing: Using a Power Automate or Power Apps template still consumes AI Builder credits (either trial or paid). Copilot actions in SharePoint require the Copilot license (as above). (Copilot Chat itself is free with many plans, but document summarization and advanced query features are part of the paid Copilot.)
• Future: Microsoft continually adds new templates (often industry-specific) and expands Copilot’s prompt library. Expect richer AI actions in Teams, Word, and beyond.
• When not to use: If you have no AI Builder or Copilot capacity, these won’t work. If your scenario is highly unique, a custom solution may be better. If your existing process is simple and working fine, a full AI deployment might be premature.

Copilot Document Processor Agent

Microsoft also released a Document Processor Agent as a Copilot Studio template. Think of it as an AI-powered mailroom clerk. Once you deploy the agent, it monitors an email inbox (or a Teams channel) for incoming document attachments (PDFs, Word, etc.). When a document arrives, the agent automatically extracts key information and exports it to your chosen system (Dataverse, SharePoint, SQL, etc.). It even handles review: if the AI is uncertain about a field, it can route the document for human validation.

You configure the agent by specifying which fields to extract (e.g., “Name”, “Order ID”, “Total Amount”). The agent uses built-in AI (no training required) to find and pull those fields from each document. There’s a dashboard to monitor processed documents, review any issues, and track performance.

• Advantages: It’s fast to set up (install and configure in Copilot Studio). It can process large volumes automatically, and because it uses Copilot and Dataverse under the hood, it’s enterprise-secure and connects easily to other Microsoft data. Essentially, it turns incoming documents into structured data with almost no coding.
• Drawbacks: As of 2025, it’s preview-only so that some features may be missing or changed. It assumes documents arrive by email or in a channel; other workflows need adaptation. You need a Copilot Studio setup, which means the Copilot license plus additional capacity. Extraction quality depends on the AI’s understanding; very unusual layouts might not parse perfectly. Heavy use consumes Copilot Studio “messages.”
• Licensing: You must have Microsoft 365 Copilot to use Copilot Studio and its agents. Copilot Studio usage is metered by messages: a Starter Pack is about $200/month for 25,000 messages. (Think of each message as one document processed or one AI response from the agent.)

Future: I expect more Copilot Studio agents (for expenses, contracts, etc.) and enhancements (trigger on folder drop, more output formats). As Copilot’s document AI gets better, extraction accuracy, language support, and customization will increase.

Licensing Comparison

The methods above involve different licensing models:

Document Processing (Syntex)/Content AI	Primarily pay-as-you-go via Azure. You activate it in the admin center (tied to an Azure subscription) and pay per transaction. Legacy Syntex per-user seats (~$5/user) are no longer sold; existing seats remain valid until expiration.
AI Builder (Forms/models)	Requires Power Apps or Power Automate plans. A Power Automate Premium license (~$15/user/month) includes a baseline AI Builder quota (5,000 credits per user). For larger workloads, purchase an AI Builder capacity add-on ($500 per T1 per month for 1M credits).
Knowledge Agent & Copilot	Requires Microsoft 365 Copilot. The Copilot add-on costs roughly $30/user/month (annual) on top of an eligible M365 plan. This covers the AI chatbot, Copilot in apps, and the Knowledge Agent in SharePoint. From March 2025, SharePoint autofill columns are priced at $0.005 per transaction if using PAYG (without KA)
Copilot Studio / Document Processor Agent	Also requires the Copilot license. AI calls meter Copilot Studio usage. For example, a Copilot Studio Starter Pack (25,000 messages) is ~$200/month. Higher tiers are available for heavier use.

Conclusion

Always check the latest Microsoft documentation or licensing guides for exact SKUs and prices, as offerings can evolve. In summary, Microsoft 365 now offers a rich AI toolbox for document processing – from SharePoint Content AI and form models to Copilot assistants and agents. Each has its strengths, costs, and ideal scenarios. By understanding the trade-offs (as detailed above), IT admins and business users can pick the right approach to automate content-heavy processes and prepare for an AI-powered future.

Hope this helps!