Reducing repeated IT support questions with AI for a media and entertainment company

The challenge

A large German media and entertainment company had a repeat IT support problem. Its central service desk was answering the same first-level questions again and again for a workforce of several thousand people.

The most visible example was VPN support. Employees also needed help with MFA on new phones, printer mapping after device changes and password resets for specific systems.

The answers usually existed, but they were scattered. Some lived in Confluence. Some were in PDF runbooks. Many were buried inside old ServiceNow tickets, where an AI Agent had already written the fix and closed the case.

First-level support was spending too much time finding answers that already existed. Skilled AI Agents were opening systems, checking old fixes, reading guidance and sending the same answers back to employees.

A simple chatbot would not have fixed that. A guessing chatbot would have made it worse. The AI Agent needed to retrieve the right answer, show where it came from, protect personal data and hand the question to a human when it could not answer safely.

• Repeated IT questions were taking up first-level support time.
• Useful answers existed, but were spread across Confluence, PDF runbooks and ServiceNow.
• Curated documentation and real ticket fixes did not always tell the same story.
• Resolved tickets contained practical fixes that had not always been turned into formal guidance.
• Support data included names, contact details, identifiers and other personal data.
• Low-confidence answers needed a clear ServiceNow handover, not a dead end.

Capabilities applied

The work needed more than a chat interface. It needed a controlled way to connect existing knowledge, protect personal data, answer with evidence and hand unresolved questions back to the service desk.

• Information Security, Data Architecture & Intelligent Systems (LLMs)
• AI Engineering, Prompting & Workflow Delivery
• AI Quality Assurance, Governance & Risk
• AI Strategy & Opportunity Design
• AI Product, Experience & Interaction Design
• Adoption, Enablement & Operational Readiness

What the work involved

• IT support workflow review
• Knowledge source connection
• Confluence and ServiceNow content handling
• Personal data masking
• Source-based answer retrieval
• Question intent matching
• Answer citation
• ServiceNow ticket handover
• Cloud service delivery
• Monitoring and improvement

The solution

The team built an IT support AI Agent that used the organisation's existing knowledge instead of starting from a blank chat window.

The AI Agent connected formal guidance from Confluence and PDF runbooks with practical fixes from previously resolved ServiceNow tickets. That mattered because documentation showed how things should work, while resolved tickets often showed how issues were actually fixed.

Personal data was treated as a front-door control. Names, contact details and identifiers were detected and masked before content was stored or made searchable.

The AI Agent first worked out what the employee was really asking. A broken VPN connection, a request for VPN access and a contractor VPN question may all mention VPN, but they need different answers.

When the AI Agent had enough confidence, it answered with a source reference. When it did not, it created a ServiceNow ticket from the conversation and gave the employee a ticket number in the same chat.

• Connected Confluence, PDF runbooks and ServiceNow ticket knowledge.
• Cleaned and prepared source material before making it searchable.
• Masked personal data before storage and indexing.
• Used Azure OpenAI, PromptFlow and Azure AI Search behind the scenes.
• Used FastAPI and Azure Container Apps to deliver the service.
• Returned answers with citations back to the relevant source.
• Created a ServiceNow ticket for low-confidence or out-of-scope questions.
• Used monitoring across production and non-production environments.

Impact and evidence

The AI Agent changed the support pattern. Instead of asking a human AI Agent to search across multiple systems for repeat questions, employees could receive a sourced answer when the available knowledge supported it.

It also made the failure path more useful. Questions that were too uncertain or outside scope became ServiceNow tickets, so the employee still had a clear route to help.

The source material does not provide approved figures for ticket reduction, containment rate, user feedback, time saved or ROI. Those outcomes should be assessed through containment rate, category-level ticket trends and user feedback before any public performance claim is made.

Business impact

• Employees: Routine IT questions could be answered from internal knowledge when a safe, source-based answer was available.
• First-level support: AI Agent could spend less time acting as a search layer and more time on issues that needed judgement, access, action or follow-up.
• Knowledge management: Resolved tickets became a useful knowledge source rather than closed records that buried practical fixes.
• Data protection: Names, contact details and identifiers were masked before content was stored or made searchable.
• Service workflow: Questions the AI Agent could not answer became pre-filled ServiceNow tickets instead of abandoned chat sessions.
• Measurement: The case identified containment rate, ticket trends and user feedback as the evidence needed to assess value after go-live.

The result

First-level support was not replaced. The repeated search work was moved into a governed support flow, while people stayed responsible for exceptions, judgement and follow-up.

The work also made knowledge quality harder to ignore. If a source was wrong or out of date, the assistant could still point to it. That made content ownership important after go-live.

For this client, repeated support questions were already visible in ServiceNow data. The practical move was to automate retrieval and handover, not the people.