Microsoft 365, Artificial Intelligence, and Privacy

Last November, Microsoft experienced a great deal of criticism from privacy campaigners following the introduction of the Productivity Score feature. The resulting hubbub evolved into a PR issue which caused Microsoft to abandon display of individual user details, even if the data is easily available and widely used elsewhere inside Microsoft 365.

In a nutshell, people got upset because it was theoretically possible for managers to use the information presented in the Productivity Score to rate individual employees. No manager worth their salt would do such a thing because the data is simply not rich or revealing enough to create a reliable picture of someone’s productivity. In any case, the good ship Productivity Score is holed beneath the waterline and it wouldn’t surprise me if the feature disappears in due course.

No Disquiet About Using User Data Elsewhere in Microsoft 365

Even if I disagree with the concerns expressed about Productivity Score, it’s good that people worry about privacy. I’m surprised that more disquiet is not expressed about the growing influence of artificial intelligence within Microsoft 365. Or rather, how Microsoft is using data gathered about user activity to develop new features.

We started along this path with the arrival of Delve. I recall the reaction of part of the audience at an Ignite 2015 conference session when Microsoft explained that signals about user activity were gathered in the Office Graph (now part of the Microsoft Graph) to allow intelligent applications like Delve help users find information of interest. Concepts like being able to view “documents trending around me” seemed interesting (obviously, trending documents are those being worked on by important people), but the concerns expressed by some organizations (mainly in Europe) focused on how to control the set of signals about user activity collected in the Graph. Apart from a setting in Delve to allow users prevent their documents being found by other users (often because the SharePoint permissions set were too loose), I can’t recall Microsoft ever providing controls over the signal collected by the Graph.

Artificial Intelligence Spreading Across Applications

Signal collection continues unabated today, both in the Graph and the Office 365 audit log. In addition, many Microsoft 365 applications leverage user data to power features. Consider:

  • Outlook text predictions use artificial intelligence to figure out what someone is likely to type next given what they have typed before.
  • MyAnalytics reads email to generate insights about tasks and things for users to follow up on, like a commitment made in an email response to someone.
  • Microsoft 365 roadmap item 66189 says that sharing links will include “at a glance” information extracted from Office documents (March 2021).
  • The Editor feature in Word assesses the quality of someone’s writing to suggest improvements.

If you believe that Microsoft is hell-bent on compromising user privacy, you could make the case that the data used for these features could be harvested to build a picture of user productivity. I guess anything’s possible, but you’d have to figure out how to grab the information from where it’s stored in mailboxes and other places.

Cloud Transition Delivers Data to Providers

The point is that the transition to cloud-based applications has made it possible for providers to accumulate vast amounts of data about how people use their services. As evident in the move towards intelligent search across Microsoft 365, the trend will continue. In effect, when you use a cloud service, you allow the applications to gather information about what you do and how you do it. And if you don’t like this development, you need to go back to on-premises systems or find a different cloud provider who doesn’t believe quite so ardently in the benefits of machine learning and artificial intelligence.

Still Not Possible to Measure User Productivity

I’m not worried on a personal level. The combination of raw statistical data like number of messages sent and received together with the more nuanced view of how well I meet commitments is not accurate or insightful enough to tell anyone how productive I am in email. The same is true for writing stored in SharePoint and OneDrive, chatting and meeting in Teams, closing tasks in Planner, or videos uploaded and viewed in Stream.

I’ll get worried when the Microsoft Graph meshes the statistical view of user activity with a qualitative measurement of my interaction across the full spectrum of Microsoft 365 applications. At that point it will be time to retire.