I officially signed up for SharePoint Syntex, assigned the license to my own account, and have started to dig in to what the world of AI models will mean for customers looking to intelligently apply information governance and retention at scale across their tenant workloads.
There are many features included in SharePoint Syntex and it will take awhile to fully understand it all.
My technology focus is on security, compliance, and ways it can be automated – this is one of the key pieces of functionality SharePoint Syntex will provide. At the heart of being able to apply this at scale, a new site template called the Content Center is now available to us. Let’s start by provisioning one of those… a default content center is created when I select Automate content understanding and Get started in the Microsoft 365 Admin Center under Setup… Files and Content.
Although not required, you can have more than 1 Content Center in your tenant. This would be a way of separating our your AI Models across different administrators/Business Units in your organization.
To walk-thru a specific use-case in this post, I’m going to build a Document Understanding (DU) model around Statements of Work (SOW) for my own company – fortunately something I’ve been creating a lot of lately!! 🙂
I store SOWs across disparate sites in my tenant, (each one in a different Modern Team site relating to the client). I would like to be able to apply a SOW DU model to any library across my tenant to allow me to protect and retain those documents in a targeted and specific way.
There are 2 types of AI models within SharePoint Syntex:
Form Processing Models
This will allow end-users to train a form processing model and apply it to a document library. This is done for semi-structured documents – a great example of this is a customer form that follows a prescribed format. In my setup, I’ve ensured all document libraries across the tenant will show the option to create a form processing model.
Note: you will require AI builder credits for forms processing.
Document Understanding Models – THIS POST!
These models are based on Language Understanding models in Azure Cognitive Services. Prior to building one of these, you must create a content center to house them in. These models are for unstructured content and is what my SOW example will fall into. In my tenant, I’ve created a new Content Center called Syntex Content Center.
The SharePoint Syntex site will appear in the list of SharePoint sites in the SharePoint Admin Center and permissions can be administered from there.
Follow along as I create my first Document Understanding (DU) model for my SOWs:
An SOW Content Type has been created in the Content Type Gallery in my tenant called Statement of Work. I’ve published the content type to my Customer Hub in my tenant. (To manage all collaboration with my customers, I create a new Modern Team site for each one and join it to my Customer Hub. Therefore, all SOWs will exist somewhere within the Customer Hub architecture)
It took about 30 minutes before I could see the Statement of Work content type in the SharePoint Syntex Content Center:
Step 1: Create a model
Provide a name and associate a content type to your model. You can create a new content type in-the-moment or use a previously created one. In this post, I’ll use the Statement of Work custom content type published from the Content Type gallery from the previous step to create a new DU model, StatementOfWork.
Once created, the model has a .classifier extension: StatementOfWork.classifier and is stored in the Models document library on the Content Center. You can, of course, have many models defined in the Content Center.
Add Example Files
Step 2: Add SOW example files (5 positive examples, 1 negative example)
In the Training Files library, I created a folder to house the example files specific for the StatementOfWork DU Model. I then uploaded 5 positive examples and 1 negative example into the folder, selected them, and added them to my model:
Classify files and run training
Step 3: Train the model to identify the files by labeling them and adding an explanation. This is adding some intelligence to your content type!
Click Train classifier…
This step will go thru each of the 6 files individually and ask if the file is an example of a Statement of Work.
Once enough files have been labeled to satisfy the model, a green toolbar message will appear advising you of this and that you can proceed to the Train step.
Step 4: Add explanations for your example files. Explanations help the model distinguish the SOW from other types of documents.
Once you add your explanations, you can train your files by selecting Train Model.
Try uploading other examples of both positive and negative files to see how smart your model is… it will tell you if it considers it a match or a mismatch.
Create and Train Extractors (Optional)
Step 5: You may want to extract words, phrases, and/or numbers from the example files. This will equate to library columns where you’ve applied the model. I have 2 site columns in the Statement of Work content type (Customer and SOW Effective date) and definitely want to pull these values from the SOW documents.
Add extractors – because I have associated my custom content type, Statement of Work, with this DU Model, I can add an extractor for each of the 2 site columns to have them automatically populated in document libraries when I apply this DU Model:
I created 2 custom extractors:
- Customer Extractor
- SOWEffectiveDate Extractor
Go thru your positive example files and select the value within the document that you want to populate the content type column. I hit Save for each file to see the Label that was applied:
Walk thru your trained files to see the document content that will be associated with your extractor and populated in your document library columns.
This is the part we’ve been waiting for… now to apply it to document libraries across your tenant to see if the DU Model can accurately detect an SOW, assign the content type and the metadata columns.
Step 6: We’ve added some intelligence to the Statement of Work content type so let’s see if it will be automatically applied to new SOWs added to a library. You will be prompted with the familiar ‘Frequent sites’ and ‘Recent sites’ dialog when determining where you’d like to apply your DU Model. I’ll select a site that has some SOWs in the library where the Content Type is not yet associated to it, NexNovus Opportunites.
The model is now active on the library and will automatically run when new files are added. For existing files, you can select a new toolbar option, Classify and extract to schedule the classification and extraction process using the model against the file.
I selected a few SOWs in the library (you can run the model against more than 1 document at a time) and in a few minutes the content type was automatically added to the library (it wasn’t previously) and was applied to the SOWs. The extractors also set the metadata, including the Customer name (Managed Metadata term) and the SOW effective date column. (Take note of the additional columns added to the library from SharePoint Syntex: Confidence Score, Classification Date, and Model URL)
Let’s add some Compliance!!!
Now that we’ve tested our model, let’s look at how we can associate a Retention Label to it so all Statement of Work content can be retained as per my retention requirements.
If you click the gear icon while in your model, a fly-out pane will have a Security and compliance section. From the dropdown, I’ll associate the Statement of Work retention label to the model…
I’ve previously created and published the Statement of Work retention label to all SharePoint sites.
I’ll re-apply the model to the document library and rerun the Classify and extract process on the documents. As you can see, not only has the custom content type and its metadata been applied, but also the retention label. Sweet.
That was a whirlwind trip thru SharePoint Syntex. Lots yet to learn, but I’m extremely impressed with how quickly I was able to learn the process and get a Document Understanding model built, trained, tested, and deployed to a library in my tenant. I’m looking forward to blogging and speaking about this capability and working with some of my larger customers to intelligently apply governance and compliance at scale across their tenants using this new capability.
Thanks for reading.