Microsoft says changes to its employee support agent helped raise the share of staff starting their support journey with it from 27% to 50%, while IT support tickets fell by 30%. The company’s account makes a useful case that an AI help desk needs more than accurate answers: it needs to remember what people have tried and offer a clear route to a person.
Microsoft AI Watch analysis
What happened
Microsoft’s internal IT organisation studied why employees were not choosing its Employee Self-Service agent, launched in autumn 2025. The company says research with its employees found that confidence depended on the whole support experience, not just the agent’s technical performance.
Microsoft says the team changed the agent to acknowledge previous troubleshooting, respond to users’ context, explain its limitations and make escalation to a live support worker clear. The company describes these as design changes informed by user research, rather than a new model or a standalone product launch. Read Microsoft’s account.
Why it matters
A support agent that sends someone through steps they have already tried can make automation feel less like help and more like a locked door with excellent grammar. Recognising prior efforts and making hand-off clear could make self-service useful without forcing staff to pretend it solved everything.
Microsoft reports a 30% overall reduction in IT support tickets and says the share of employees beginning their support journey with the agent rose from 27% to 50%. Those are results reported by Microsoft about its own operation; the article does not provide a comparison that isolates the effect of the trust-focused changes from other factors.
Our read
The practical lesson is refreshingly unglamorous: an agent should know what has already happened, say when it is out of its depth and make the next step obvious. That is a better service design brief than asking a chatbot to sound reassuring while repeating the same troubleshooting script.
Microsoft’s figures are encouraging, but one company’s internal case study is not a universal forecast. Teams considering an AI help desk should measure whether it actually resolves requests, whether people choose to use it and how smoothly it hands difficult cases to staff.
What to watch
- Whether Microsoft shares more detail on how it measured ticket reductions and agent use.
- Whether similar changes improve outcomes in other organisations and support settings.
- Whether users can reach a person promptly when self-service is not enough.
Discussion spark: Should an AI help desk be judged mainly by how many requests it resolves itself, or by how well it helps people reach the right support, including a person?
Sources and evidence
not affiliated with or endorsed by Microsoft