Microsoft’s transition from Office 365 to Microsoft 365 is nearly complete, bringing together its operating system, management tools, and productivity platforms into a single subscription offering, underpinned by the connective layer that is the Microsoft Graph. At its recent Ignite event in Orlando, Microsoft announced what it describes as the fourth pillar of Microsoft 365 — Project Cortex.

Perhaps best thought of as the latest iteration of knowledge management, Project Cortex is a way of bringing together key aspects of Microsoft 365 to help you find answers to questions about your organisation and the work it does. Building on familiar tooling, it’s perhaps best thought of as an evolution of the Microsoft Search tools available through Bing that add work queries to your browser.

Microsoft Search was focused on context-based search, using the tool you’re searching from to refine the underlying query. A search in Outlook would be primarily for emails, and in Word for documents. However, it’s still an intranet search tool, albeit one that’s better at providing the answers you’re likely to be looking for.

Going beyond Microsoft Search

Building on that foundation, Project Cortex goes further, using machine learning to generate content based on documents and enterprise social network conversations, constructing what can perhaps best be described as a ‘corporate Wikipedia’. By using elements of Azure’s Cognitive Services and the Microsoft Graph, content can be tied to key individuals (much like in Microsoft Search) and delivered in channel-appropriate formats.

Most interactions with Project Cortex will be through what Microsoft is calling ‘Topic cards’. These pop up at appropriate places — initially in Word, Outlook, Sharepoint and Teams. From the initial view we’ve had, as well as showing key information about content and people, they let you follow specific topics as well as suggesting any edits. Topics are reflected in Outlook’s ‘People cards’.

For example, you’ll be able to look for people related to specific projects in Outlook, with acronyms and definitions in Word, and quick responses via bots and ‘Adaptive cards’ in Teams. Project Cortex applies existing role-based access controls to content, so a commercially sensitive project’s content will only be visible to project team members. Other content might be limited to summary form only, keeping detailed content for only those with the appropriate access rights.

Adding machine learning to knowledge management

Project Cortex isn’t only for Office documents. Using Azure’s Cognitive Services, it can use image and text recognition to work with scanned content, images, and other file formats such as PDF. It can even use rules to define form structures, so that key information can be extracted from scanned forms and other common document types, allowing you to build a model of where projects are spending money by parsing purchase orders and invoices. Extracted information is used as metadata to provide context around documents, helping users find the content they need.

You’re not limited to structured document types. Another Azure Cognitive Service, LUIS, forms the basis of Project Cortex’s Machine Teaching. Here you can build new document models that look for key terms, allowing classification of, say, contracts which will differ from contract to contract, with different content and different formatting. Once a model is trained it can be used across your entire document store, improving search and increasing your organisation’s underlying knowledge model.

SEE: Microsoft HoloLens 2: An insider’s guide (TechRepublic download)

Project Cortex comes with a set of connectors for third-party repositories, so you can take advantage of investments in enterprise content management tooling and in older knowledge management platforms. The initial set includes support for Azure Data Lakes, ServiceNow and Salesforce. An ingestion API will allow other vendors to quickly add their own connectors.

There’s some element of content control in the tooling: if, for example, the generated content for a corporate acronym or project name is thin or inaccurate, internal editors can flesh it out with more information. There’s no need to learn new skills here, as the underlying content platform is based on SharePoint.

Extending the Microsoft Graph with Project Cortex

It’s important to understand that none of the information generated by Project Cortex leaves the Microsoft Graph. Instead it adds new metadata to your documents, using an automatically generated taxonomy, the Managed Metadata Service. As it’s a managed service, you can add your own metadata or create new tagging definitions. The resulting metadata, stored as knowledge entities in the Graph, is used to define the topics that deliver Project Cortex results.

One key driver behind Project Cortex is demography: the bubble of baby boomers is rapidly leaving the workforce as they retire, taking knowledge and skills with them. Much of that knowledge isn’t completely lost, but it’s trapped in the many terabytes of files of documents that have built up in our business systems. We need a way of transferring that knowledge to new generations of workers, across all our businesses. By using Project Cortex to capture and index that information, and with AI tools to build ways of navigating and extracting that data, we can turn corporate document stores from a memory hole into a dynamic learning environment.

SEE: Managing AI and ML in the enterprise 2019: Tech leaders expect more difficulty than previous IT projects (TechRepublic Premium)

Once that information is available to everyone, it can be used as the basis for dynamically generated FAQs and for corporate search results, with just-in-time access to essential content that not only helps with skills transfers, but can also be used to scale-up expanding businesses. Project Cortex doesn’t just capture knowledge from documents; it can also work with tools like Yammer and Teams to extract the implicit knowledge in our businesses — the things everyone knows informally and that are transferred by asking the person you sit next to rather than by reading a corporate handbook.

This is where much of Project Cortex’s machine learning comes in, helping to extract and categorise that informal knowledge sharing from corporate social networks, as well as from more formal documents. It’s an ambitious move for Microsoft, because past efforts to deploy knowledge management tools at scale haven’t gone well. It may be that Project Cortex’s mix of Microsoft Graph and machine learning can finally bridge the gaps between data, information and knowledge.

Also see