Microsoft wants to help organizations use their data to deliver experiences, insights and outcomes. But careful design, clear ethics and full transparency are required to keep the AI in check.
When Microsoft first added metrics about employee experience for Endpoint Manager, they focused first on network speed and other back-end issues, and then on device measurements like how long PCs take to boot. The now-infamous Productivity Score was about measuring what might be getting in the way of productivity, whether that was multiple security agents slowing down reboots or employees clinging to pre-Office 365 habits and mailing documents as attachments rather than collaborating on them in OneDrive and Teams.
Microsoft 365 has also been automatically sending staff Workplace Analytics that try to help them pin the people they talk to most for easier access, and remind them to book some ‘focus time’ on their own calendars if they’re spending so much time in meetings that they can’t get anything else done.
SEE: How to highlight duplicate values in Excel (TechRepublic)
There was an outcry when those two separate views of the information available in the Microsoft Graph seemed to be merging, with a preview feature that added details like how long people had specific applications open — and potentially showed that to managers who might use that as a measure of how hard someone was working. Microsoft quickly cancelled the preview feature, and declined to talk further about what the useful aspects of it would have been.
But transforming the modern workplace is a key part of Microsoft’s current strategy. Microsoft has just launched WorkLab, a site for sharing science and research about how work is changing, and this week CEO Satya Nadella will be talking about ‘reimagining the employee experience’. Some of that transformation is new features in Teams and Yammer, and tools that aim to make it easier for people to support and encourage colleagues. It’s also the way Microsoft uses your own data to help you: building a list of the acronyms your organization uses; drawing the graph of who you have meetings with to make it easier to find the documents they shared with you (that you’ll probably be discussing in the meeting) via Microsoft Search, while keeping the data private to your Office or Microsoft 365 tenant; or highlighting who in your business is an expert on a subject.
The combination of the Microsoft Graph with tools based on what Microsoft has learned with Bing is delivering a new version of the old ‘information at your fingertips’ slogan. Project Cortex is one of the ways that’s being delivered, using machine learning to extract and summarise the content and relationships inside an organization — and building that by working with partners and customers to make sure the tools deliver useful results.
This approach is what Jeffrey Snover (best known for PowerShell, but now the CTO of Microsoft’s modern workplace transformation efforts) refers to as delivering ‘outcomes’, rather than products and services, and the beginning of what he calls “Microsoft V3” (although Microsoft told us that isn’t an official term the company is using).
Speaking at Microsoft’s HiEd Higher Education conference, Snover talked about the V1 era of Microsoft as “producing great products” without knowing “whether people were successful with them or not”.
“That wasn’t good enough: we’ve got to make sure this product actually works and it’s running all the time, that it’s up to date and it’s secure.” Microsoft V2, in Snover’s view, was taking responsibility for the products working. “We ran them as services; we installed them, we configured them, we maintained them, we made them available throughout the world.”
But even that wasn’t good enough, Snover said. “We have to make sure people are getting the desired outcomes out of those products: are they getting business value, are they being more effective?”
According to Snover, that focus on business outcomes puts the company “on the verge of Microsoft V3” with three paths to ‘modern workplace transformation’: “information and people at your fingertips, healthy and productive people and organisations and…effective businesses”.
Even after 32 years of Microsoft Office, it might feel like “the information I want is always one inch away from where I am; it’s never quite there”, Snover said. That’s about to get dramatically better, he predicted (“stunningly good” in his words) by using the combination of Bing techniques and Microsoft Graph to find files, people, support documents and all the other resources on a topic in one place, rather than having to search each information store individually, using complex queries and customising the results.
“Everything that you store in Microsoft 365 is either housed in the substrate, or a digital twin is replicated to the substrate,” Snover explained. That ‘substrate’ is a planetary-scale NoSQL data system (Microsoft has just open-sourced JET Blue, the Extensible Storage Engine that acts as the core data engine for the Microsoft 365 substrate).
What that produces isn’t a list of files or links; it’s an entity graph that you can pivot — and that can be consumed by multiple applications and tools. Project Cortex uses that to create ‘synthetic’ nodes in the graph for entities that aren’t people or files, but projects or products your organization creates.
“It turns out that the way people talk about projects is pretty common. So we’ll scan all the emails, all the conversations all the chats, all the documents, and the AI looks at this and does some probabilistic identity matching and says ‘I think there’s a high probability that they’re talking about project ‘foo’ here, so I’m going to create an entity called the project, and I’m going to link to it, all the people talking about that project, all the documents and the meetings and the websites about that project’. So then when I look at a project manager, I could see what projects she is working on, navigate to the project and from there see all the people talking about the project.”
SEE: How to avoid a disappearing page number in Microsoft Word (TechRepublic)
As well as productivity, there’s an increasing emphasis on employee experience, especially with the stress of the pandemic — working later, blurring the boundaries between work and home, feeling more stressed and less connected because of video meetings — and the burnout that can bring. “A modern workplace is not just about personal productivity; it’s about organisational productivity, and neither of these are possible if the people and the organisations are not healthy,” Snover pointed out.
The wellbeing and productivity experiences will be there to help managers as well as individuals (giving a hint as to why the Productivity Score might have moved to include usage patterns) and Snover mentioned recommendations that are reminiscent of the nudging that Reinforcement Learning is promising: “They’ll get personalised insights to their role, along with recommended actions to how to change habits or create new habits.”
That will be done in a “privacy protecting way”, Snover promised. “The data is yours, not ours, and the data is only used for you and only used for you in a way that protects privacy. We never give information up to a higher level of the organisation in a way that it can identify an individual. We use differential privacy to fuzz the results so you can never identify an individual person.”
Individuals will get suggestions, like the daily emails from Cortana that reminds you of the commitments you’ve made as well as suggesting documents to help you prepare for upcoming meetings. Snover talked about this feature “Engineering the sense of psychological closure: I’m worried all the time about all the commitments that I’ve made. Who am I disappointing today?”
What a manager will see is more that there’s been a noticeable drop in the mood of the organisation, or reminders of who they haven’t spoken with recently. “We want to help managers be effective, so we keep track of things; we know good leadership is connecting with people, so we keep track and suggest ‘it’s been a while since you had a one-on-one with this person, would you like to schedule some time?’.”
One of the predecessors of the Microsoft Graph was a Microsoft Research tool called SNARF, the Social Network and Relationship Finder that offered to filter and triage email based on the interactions you’d had with people and — in an internal version that wasn’t released publicly — your connections in the Microsoft reporting structure.
At the time, Microsoft researcher Marc Smith pointed out that email had no understanding of human relationships, and that’s one of the things Microsoft is trying to improve, Snover said. “It used to be true that you had to understand the tools to get your job done. AI attempts to understand a person in order to achieve some result. We use AI to understand you. Why? To help you achieve your objectives, to help you go farther go faster, to bring the information to your fingertips. We’re trying to understand you to help you accomplish your goals.”
That might be something fairly simple like setting an out-of-office announcement, where Outlook spots what you’re trying to do and pops up a dialog where you can set the dates and what your automatic reply should say. “Instead of you having go out to find all the tools, we bring the tools to you as soon as we figure out what you’re trying to accomplish. We understand your objectives and then bring in line experiences to help you achieve them,” Snover explained.
Those objectives might also be more complex business outcomes, and what Snover will be focusing on in his job is tuning Microsoft 365 tools for vertical industries and first-line workers. “We have a bunch of great systems of engagement. And then there are these systems of record as well. And we’re trying to figure out how to marry these two and figure out where the gaps are and fill those gaps, so that you can use both systems of record, and systems of engagement to better drive your business.”
That includes insights and analytics about the key performance indicators of a business, as well as the critical workflows or specific industries. In healthcare that might involve building specific features in Teams like sending a message to whoever is in a specific role for the current shift, or repeating an urgent message sent to a doctor until they acknowledge it or sending it to another doctor if they don’t respond within a set time.
SEE: 69 Excel tips every user should master (TechRepublic)
Instead of one-off solutions, those will be reusable ‘building blocks’ using tools like task bookings and priority notifications. In the past, Microsoft has talked a lot about using PowerApps for organisations to fill in those gaps where commercial software doesn’t deliver exactly what an organization needs, but those are only reusable within one organisation. There are also Dynamics 365 applications that offer specific insights and analytics for business metrics like customer churn, and ways of creating them in Power BI. And Microsoft has long talked about combining systems of record and systems of engagement with Dynamics 365, so while Microsoft V3 may not be an official term, it fits with many long-term developments and priorities.
The Microsoft 365 team can undoubtedly build on those options and it’s likely that Microsoft will continue to have multiple ways of doing this depending on how much businesses want to build themselves and how much they want as part of a prebuilt service. Some of this will also tie into the vertical clouds that Microsoft is delivering, like its retail and healthcare cloud offerings.
But the concerns about Productivity Score reflect valid concerns about AI moving into a very human area. A tool pointing out that I used to meet with someone frequently and now I don’t is much more personal than a word processor spotting that I’m referring to a specific company project and offering to link to resources about it. The reasons why those meetings stopped could be even more personal: that colleague might have changed jobs, fallen ill — or I might be avoiding them because they’ve been bullying me.
These are human management issues that AI isn’t close to being able to help with, and the underlying problem isn’t a technology one. What’s needed is careful design, clear ethics and full transparency if we’re to avoid AI-powered suggestions about human interactions in the workplace that make people uncomfortable.