![\"\"](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2025\/07\/Learn-how-p-g.png\")
<\/figure>\n<\/div>\n\n\n\nRead our guide<\/strong><\/strong><\/strong><\/p>\n\n\n\n Learn how to deploy an Employee Self-Service Agent based on our experience here at Microsoft.<\/a><\/p>\n<\/div>\n<\/div>\n\n\n\n We already fundamentally understand that the power of AI and large language models has changed the game for many work tasks. The way employee support functions is no exception to this sweeping change.<\/p>\n\n\n\n \u201cA tool like the Employee Self-Service Agent doesn\u2019t know if something is true or false\u2014it only sees information it can use and present. That\u2019s why stale or outdated information is such a risk, unless you manage it up front.\u201d<\/p>\nDavid Finney, director of IT Service Management, Microsoft Digital<\/cite><\/blockquote>\n\n\n\n Instead of relying on human agents to answer employee questions or resolve issues, we now have AI agents trained on vast corpora of data that can find the answer to a complex question in seconds.<\/p>\n\n\n\n But in our drive to give these tools access to everything they might need, they sometimes end up consuming information that isn\u2019t helpful.<\/p>\n\n\n\n \u201cA tool like the Employee Self-Service Agent doesn\u2019t know if something is true or false\u2014it only sees information it can use and present,\u201d says David Finney, director of IT Service Management. \u201cThat\u2019s why stale or outdated information is such a risk, unless you manage it up front.\u201d<\/p>\n\n\n\n Before AI, support teams didn\u2019t need to worry as much about the buried issues with unstructured content because a human could generally spot it or filter it out manually. After we turned these tools loose, they began reading everything, including:<\/p>\n\n\n\n \u201cFor example, humans never saw the old, decommissioned SharePoint sites because they were automatically redirected,\u201d says Kevin Verdeck, a senior IT service operations engineer. \u201cBut AI definitely could find them, and it surfaced ancient information that we didn\u2019t even know was still out there.\u201d<\/p>\n\n\n\n A major part of the solution to this problem is better governance. We had to get a handle on our data.<\/p>\n\n\n\n \u201cWe needed to determine the owners of the sites and then establish processes for reviewing content, updating it, and defining how it should be structured. I would highly encourage that our customers think about governance first when they are launching their own AI tools, because everything flows from it.\u201d<\/p>\nOlivier Cherel, senior business process manager, Microsoft Digital<\/cite><\/blockquote>\n\n\n\n The first step was a massive cleanup effort, including removing decommissioned SharePoint sites and deleting references to retired programs and policies. The next step was making sure all content had ownership assigned to establish who would be maintaining it. This was followed by setting up schedules for regular content updates (lifecycle management).<\/p>\n\n\n\n Governance was the first priority for IT content, according to Olivier Cherel, a senior business process manager in Microsoft Digital.<\/p>\n\n\n\n \u201cWe had no governance in place for all the SharePoint sites, which were managed by the various IT teams,\u201d Cherel says. \u201cWe needed to determine the owners of the sites and then establish processes for reviewing content, updating it, and defining how it should be structured. I would highly encourage that our customers think about governance first when they are launching their own AI tools, because everything flows from it.\u201d<\/p>\n\n\n\n Content governance was also a huge challenge for other support areas, such as human resources. A coordinated approach was needed.<\/p>\n\n\n\n \u201cHR content is vast, distributed across multiple SharePoint sites, and not everything has a clear owner,\u201d says Shipra Gupta, an engineering PM lead in Human Resources who worked on the Employee Self-Service Agent project. \u201cSo, we collaborated with our content and People Operations teams to create a true content strategy: one source of truth, no duplication, with clear ownership and lifecycle management.\u201d<\/p>\n\n\n\n Cherel observes that this process forces teams to think about their support content in a totally different way.<\/p>\n\n\n\n \u201cPeople realize they need a new function on their team: content management,\u201d he says. \u201cYou can\u2019t simply rely on the knowledge found in the technicians\u2019 heads anymore.\u201d<\/p>\n\n\n\n The simple truth is that part of what makes unstructured data so difficult for agentic AI tools to deal with is that it\u2019s disorganized.<\/p>\n\n\n\n \u201cOur HR Web content already had tagging for many policy documents, which helped us get started. But it wasn\u2019t consistent across all content, so improved tagging became a big part of our governance effort.\u201d<\/p>\nShipra Gupta, engineering PM lead, Human Resources<\/cite><\/blockquote>\n\n\n\n AI works best with content that has as many of the following characteristics as possible:<\/p>\n\n\n\n This kind of formatting and metadata allows the AI tool to more clearly parse and sort the information, meaning its answers are going to have a much higher accuracy level (even if it might be a little slower to return them).<\/p>\n\n\n\n \u201cA good example here is tagging,\u201d Gupta says. \u201cOur HR Web content already had tagging for many policy documents, which helped us get started. But it wasn\u2019t consistent across all content, so improved tagging became a big part of our governance effort.\u201d<\/p>\n\n\n\n Be sure that as part of your content review, you\u2019re setting aside the time and resources to add this kind of structure to your unstructured data. The investment will pay off in the long run.<\/p>\n\n\n\n As AI tools grow more sophisticated, we\u2019re using them to directly work on AI-related challenges. This includes using AI on the challenge of unstructured data itself.<\/p>\n\n\n\n \u201cRight now, these efforts are primarily human-led, but we are applying AI to, for example, help write knowledge base articles,\u201d Cherel says. \u201cAlso, we\u2019re starting to use AI to determine where we have content gaps, and to analyze the feedback we\u2019re getting on the tool itself. If we just rely on humans, it\u2019s not going to scale. We need to leverage AI to stay on top of things and keep improving the tools.\u201d<\/p>\n\n\n\n Essentially, the future of such technology is all about using AI to improve itself.<\/p>\n\n\n\n \u201cWe\u2019re looking at building an agent to help validate content,\u201d Finney says. \u201cWe can use it to check for outdated references, old processes, or abandoned terms that are no longer used. Essentially, we\u2019ll have AI do a readiness check on the content that it is consuming.\u201d<\/p>\n\n\n\n Ultimately, the better the data is conditioned, the more accurate and relevant the agent\u2019s responses will be. And that will make the end user\u2014the truly important human in the loop\u2014much happier with the final outcome.<\/p>\n\n\n\n Key takeaways<\/p>\n<\/div>\n\n\n\n We\u2019ve highlighted some insights to keep in mind as you consider how to condition your own organization\u2019s data for ingestion by AI tools:<\/p>\n\n\n\n Try it out<\/p>\n<\/div>\n\n\n\n Learn how your organization can take advantage of the power of the Employee Self-Service Agent.<\/a><\/p>\n<\/div>\n\n\n\n Related links<\/p>\n<\/div>\n\n\n\n Anyone who has ever stumbled across an old SharePoint site or outdated shared folder at work knows firsthand how quickly documentation can fall out of date and become inaccurate. Humans can usually spot the signs of outdated information and exclude it when answering a question or addressing a work topic. But what happens when there\u2019s […]<\/p>\n","protected":false},"author":209,"featured_media":23022,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":true,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_hide_featured_on_single":false,"_show_featured_caption_on_single":true,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[882,1],"tags":[199,868,900,331,850,852,827,424],"coauthors":[841],"class_list":["post-23020","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-it-strategy","category-microsoft-digital","tag-ai","tag-ai-deployment-and-adoption","tag-ai-governance","tag-data-and-ai","tag-end-user-services-and-support","tag-it-and-business-operations","tag-microsoft-365-copilot","tag-power-platform","m-blog-post"],"yoast_head":"\nHow AI has changed the game<\/h2>\n\n\n\n
![\"A](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2026\/04\/David-Finney.png\")
<\/figure>\n\n\n\n\n
\n
Data governance is the key<\/h2>\n\n\n\n
![\"A](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2026\/04\/Olivier-Cherel.png\")
<\/figure>\n\n\n\n\n
Adding structure to the unstructured data<\/h2>\n\n\n\n
![\"A](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2026\/04\/Shipra-Gupta.png\")
<\/figure>\n\n\n\n\n
\n
\n
\n
Using AI to help condition data for use<\/h2>\n\n\n\n
![\"\"](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2025\/07\/Key-takeaways-badge.png\"){kind=icon}
<\/figure>\n\n\n\n\n
![\"\"](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2025\/07\/Try-it-out-badge.png\"){kind=icon}
<\/figure>\n\n\n\n![\"\"](\"https:\/\/www.microsoft.com\/insidetrack\/blog\/uploads\/prod\/2025\/07\/Related-links-badge.png\"){kind=icon}
<\/figure>\n\n\n\n\n