Approaches for applying retention labels in SharePoint Sites and Microsoft Teams

As many of you will know I’ve been pretty exclusively focussing on retention in Microsoft 365 for the past decade. During this time, I’ve helped dozens of organisations radically improve the way that they manage their digital records, by introducing architectures that centre on information management lifecycles.

However, while many organisations have now taken the plunge into managing records in SharePoint and Microsoft Teams, the vast majority still haven’t taken advantage of the powerful retention capabilities that are included within their Microsoft 365 licences.

This blog post outlines all of the different approaches that Microsoft provides, which allow you to apply retention labels to your content across SharePoint Sites and Microsoft Teams. My objective here is to help inform your decision around which of the various approaches might be most suitable for you to use when planning your organisation’s adoption of retention labels.

There are seven approaches that can be used to apply retention labels to your content in Microsoft 365:

• Manual application
• Custom logic/processes
• Trainable Classifiers
• Sensitive Information Types
• Exact Data Match
• Keyword Query
• Content Processing (SharePoint Premium)
• Default labels

Each of these approaches are described in more detail in the sections below:

Manual application of retention labels

Manually tagging content is the most straightforward (yet possibly least successful) method of applying retention to your content in Microsoft 365. With this approach you publish your retention labels so that they are available for staff to use and then typically spend a lot of time trying to remind people about the benefit of applying them to files.

In the vast majority of organisations this approach simply won’t work – let’s face it, staff rarely tag their information (and even when they do, they rarely tag content with any real accuracy). Not only does this manual approach introduce a new responsibility for staff, in my experience it tends to result in fairly poor results; with relatively small volumes of content being classified and frequently with incorrect labels having been selected.

That said, I have seen it work in some organisations – but only where staff are already acutely aware of the importance of tagging content accurately. For example, you might find that manual application of retention labels is a viable option for a small pharmaceutical company undertaking clinical drug trials, where every ‘i’ is already dotted, and every ‘t’ is routinely crossed.

However, it is worth noting that if your organisation is limited to using E3 licences (or below) and doesn’t have any of the add-on Purview licences, then unfortunately, this manual approach will be your only option for applying retention labels to your content. As such, all of the other approaches detailed in this blog post are only available for organisations with more advanced licences (typically E5 or equivalent).

Custom logic/processes

The most flexible approach for applying retention labels to files is to introduce your own custom logic. Whether you are using PowerShell scripts, C#, or running Power Automate processes, it’s perfectly possible to call into Microsoft 365’s APIs to apply retention labels to your content.

This approach forms the basis of many of the more reliable records management solutions that I’ve seen introduced across Microsoft 365. However, unfortunately, this approach will invariably require the development (and subsequent maintenance) of the bespoke logic or processes. Typically, this makes the overall cost of applying retention labels via custom logic relatively expensive, something that significantly reduces the viability of this approach for many organisations.

My understanding is that applying retention labels programmatically in this way requires your users to have an E5 licence (or one of the compliance add-on licences). Certainly, that was the case when I published Retention Licensing Update | Blog in December 2023. However, at the point of writing (March 2025), I can no longer see this specifically mentioned in Microsoft’s guidance around “Licenses for other retention label application methods”.

Trainable Classifiers for applying retention labels

Having been around for a several years now, Trainable Classifiers provide an AI-driven approach to automating the application of retention labels. They can be an invaluable tool which allow you to automatically label content regardless of where it is stored across your SharePoint architecture.

Microsoft provides you with a range of pre-trained classifiers, which you can choose to make use of, however, as these models have not been trained on your own information, you will definitely want to closely verify the results. Instead, most organisations who are exploring Trainable Classifiers tend to build their own models, by training the AI on their organisation’s own content.

To build your own custom Trainable Classifier, you will, need a fairly large number of files, which all contain the same type of information. This type of content should ideally have a standard document structure, as this will allow the AI to more accurately determine whether each file should be labelled or not.

You will then need to seed the AI with at least 50 positive examples of this type of file, alongside 150 negative examples. From my testing, you might well want to exceed the minimum number of seed files to help improve the accuracy of the AI. The Trainable Classifier then takes a day or two to appraise the seeded content, and if it is successful, allows you to publish it across your tenant.

While using AI to automatically apply retention is a bit of a utopian dream for many Records Managers, there are a few drawbacks that you will probably want to consider. Firstly, to go to the effort of training a model, you will need to have hundreds (or more likely thousands) of files containing the same type of information, and for each of these files to have fairly consistent structures. This could certainly be the case for some of the information assets you possess, however, it’s likely that many of the types of content you store will not be suitable for Trainable Classifiers. As a result, this approach will probably only be able to apply label to a minority of the content that you possess.

In my opinion the biggest drawback to be aware of is that Trainable Classifiers only work on content that is less than six months old (see: Automatically apply a retention label to Microsoft 365 items | Microsoft Learn). This means that Trainable Classifiers aren’t suitable for applying labels to legacy content.

Sensitive Information Type

Sensitive Information Types are becoming central to many of the capabilities found in Microsoft Purview. They are used to scan information in your tenant to find personally identifiable information (PII) that is stored in a consistent ‘codified’ format. What I mean by this is that Sensitive Information Types allow you to find content containing codes or IDs such as Passport Numbers, Social Security Numbers, Asset IDs, or Credit Card Numbers.

Where Sensitive Information Types really shine is if you happen to have a unique ID applied to the data that you process. For example, a Housing Agency might well have files tagged with Property ID’s and Tenant ID’s, which could be easily identified (and then labelled) by using a custom Sensitive Information Type.

I should emphasise that the more unique the format of your ID is, the more accurate your Sensitive Information Type will likely be. For example, if your ID is a five-digit number, it’s very likely that the Sensitive Information Type will find (and apply labels to) files containing other five digit numbers (such as the first half of a mobile phone number). Whereas a 12-digit alpha-numeric code, with a consistent format of characters is far more likely to be unique (and as such, far less likely to be mis-labelled by a Sensitive Information Type).

In the right situation Sensitive Information Types are superb. However, if you don’t have a unique ID within your content then the value of this approach is minimal for Records Managers. Even if you do have a unique ID that you can use, you then need to ask yourself which retention label should be applied to all of the content containing that ID. Very often, you might have multiple information assets that all contain a specific ID, but which need to be retained for different durations. This is why, I’ve personally come to the conclusion that Sensitive Information Types are typically much better at applying sensitivity labels, than they are at retention.

Exact Data Match (EDM) Sensitive Information Types

Exact Data Match provides an alternative way of allowing a Sensitive Information Type to apply labels to your content, by leveraging exact, or near exact data matching as opposed to patterns. However, instead of looking for a codified ID, Exact Data Match allows you to look for your organisation’s own data when it is stored inside your content.

For example, if you have a HR system containing employee data, or a CRM containing information around your sales opportunities and clients, you can look for the actual data contained within these repositories across your Microsoft 365 content. This approach allows you to find all of the information that you store that contains, say, the name and date of birth of one of your staff, employee ID, or perhaps the post code of one of your specific customers and then use this as a way of applying labels to your content.

In many ways, EDM is an incredibly powerful tool, which can provide far greater accuracy and precision than standard Sensitive Information Types. You aren’t just labelling a file containing a sequence of numbers resembling the format of a phone number, EDM lets you label a file containing one of your customer’s actual phone numbers.

The challenge with EDM is that it is very reliant on the data you have available to use – this means that every organisation will need to plan how the existence of specific data will cause the application of which specific labels. This could form part of an internal process where HR systems are interrogated and stored in a format the EDM can reference, which is updated on a set frequency to ensure accuracy of the data supplied.

Exact Data Match has huge potential, although I suspect in most scenarios it will be far more useful for applying sensitivity labels than it will be for retention. As with standard Sensitive Information Types, it will also only ever be able to label the minority of your files, meaning that it probably won’t form the backbone of your records management architecture in Microsoft 365.

Keyword query

Probably the most powerful of the automated approaches provided for applying retention labels is Keyword Query – a function that Microsoft currently refers to as the ability to ‘Apply label to content that contains specific words, phrases, or properties’.

This is effectively a tool that allows you to define a series of conditions, which determine which files that retention label is applied to. For example, you could choose to create a query which applies a specific retention label to all of the files in your tenant that contain the phrase “Commercially Confidential”, use the “Corporate Policy” Content Type, and have also been tagged with SharePoint metadata indicating that they belong to the “Human Resources” department. As you can imagine, if you are controlling the conditions that you include within the query, will have full control over precisely which files are automatically labelled

While the tool is incredibly flexible, it can be a little bit technical to configure, as the conditions need to be defined in Keyword Query Language – as such, frequently some IT assistance is typically required.

The major challenge that many encounter with this approach, is trying to set appropriate conditions. If your conditions are too restrictive, then many files might be excluded from the scope; while conversely if your conditions are too liberal, then too many items will likely receive an incorrect label. It’s for this reason that Keyword Query isn’t typically used as the primary tool for applying retention labels.

Document Processing with SharePoint Premium

SharePoint Premium’s Document Processing models use AI to automatically find and extract metadata from within your files. For example, you could use Document Processing to scan Purchase Orders and automatically extract metadata from them, such as the PO Number and total value of the of the goods/services in the order.

The reason I mention Document Processing in this blog post, is that it can be used to automatically apply retention labels to every file that it successfully identifies. Let’s be honest, you won’t be purchasing Document Processing to meet your retention goals (you’d buy it for its metadata extraction capabilities). However, if you are planning to use Document Processing, you should certainly consider how it can apply retention labels to the files that it appraises.

Document Processing requires an additional pay-as-you-go licence which is currently priced at 5 US cents per processed page.

Default labels

In my opinion, setting default retention labels on all of the document libraries in your Microsoft 365 tenant is by far the most effective solution for applying retention to your content in Microsoft 365. This approach has been at the heart of many of the most successful retention architectures that I’ve seen.

Once default retention labels have been applied, all users need to do is upload their files and start working on them. The content then becomes automatically labelled, based on the location that the user has chosen to store it in. Importantly, staff can choose to manually change the default label, if they feel a different label is more appropriate to apply.

The only challenge that this approach presents is for organisations to work out how they are going to set the right default retention label on each of their libraries and folders. Making this process effortless is one of the main reasons Orinoco 365 has been introduced. The product allows you to automatically apply default retention labels that are contextual to the activity being undertaken across every new SharePoint site and Microsoft Team that is requested. Even if a user creates a new document library, Orinoco 365 automatically intercepts it and applies your chosen default retention label for you.

Speak to us today to find out how Orinoco 365 automates the application of retention labels, allowing you to easily take advantage of Purview’s Records Management capabilities.