Release management bot
A Teams bot that answers change-management questions and does the routine coordination, so release engineers stop digging through records.
- For
- A global consumer technology company
- My part
- Designer and developer
- Status
- Completed
The problem
Change and release management runs on information that lives in a dozen places. Release engineers kept looking up the same things: change records, schedules, owners, who’s on call tonight. Then they did the same coordination by hand: invitations, reminders, CAB communications. None of it was hard. All of it was slow.
How it works, in 5 parts
Ask in Teams
What's on tonight's change calendar? Who owns this change? Who's on call? Asked where the team already works.
Understand the ask
The bot works out what's being asked. The later version used Copilot Studio for the conversation.
Look it up
Flows query ServiceNow and the team's own lists for change records, queues, watchlists and on-call details.
Answer with a card
Results come back as an Adaptive Card with the details and buttons for the obvious next step.
Do the routine part
Send the CAB invite, add a change to a watchlist, create or close a request for change.
What I built
A bot in Microsoft Teams, where the team already spent its day. Ask it about a change and it finds the record in ServiceNow and answers with a card. Ask what’s in the queue or on your watchlist and it tells you. It can send the calendar invitations, pull on-call details and handle the routine CAB communication.
The later version, built in Copilot Studio, went further: it could create and close requests for change and pull schedules from a conversation, while the actual record changes still ran through ordinary, predictable flows.
The hard parts
Conversation on top, rules underneath. Production changes don’t tolerate guesses. The AI handles understanding the question. The lookups and updates stay deterministic, so the same request always does the same thing.
Fitting the real process. A bot that skips a step in change management isn’t helpful, it’s a risk. Every action had to match how the team already worked and what they were allowed to do.
What I learned
Most of the value wasn’t the chatbot. It was removing small bits of friction that people had stopped noticing. And AI is at its best at the edge of a system, understanding what someone wants, not in the middle deciding what happens to production.