Part 2 of 3 — Agentic AI Success Stories
Someone still typed the booking ID.
Part 1 was support that talks to customers. This part is back-office work that writes — refunds, journals, tickets — and how easy it is to call that “autonomous.”
Microsoft’s Cineplex customer story (26 March 2024) is the one to read. Roundups say an agent processed thousands of refunds on its own. The page says something more boring, and more useful.
Cineplex — copilot plus flows
Cineplex had already automated some refunds with Power Automate: capture and validate from Guest Services email, process the refund, adjust seats, journal into a legacy app. Unattended flows in parallel. That layer, Microsoft says, saves 5,000 hours a year.
Then they wired those flows to Copilot Studio. A Guest Services agent enters a booking ID and date. The copilot recognises the refund type and starts the desktop flows — ServiceNow and other backends included. Handling time dropped from 5 to 15 minutes per request to typically about 30 seconds. Monique Binder is quoted: the copilot processed over 5,000 refund requests in five months.
human enters booking ID + date
│
▼
copilot picks refund type
│
▼
Power Automate flows write
│
▼
legacy app / ServiceNowCompany-wide, Microsoft also cites 30,000 hours a year across 400+ Power Platform solutions. That number is for the whole Power Platform programme, not the refund copilot alone. Do not add it to the 5,000 requests.
This is orchestration with a start button. The person is the first step. The stop is still there — a wrong ID is a bad input, not a smarter model. Memory is still the write into the journal. When not to split still applies: you do not hire five refund personas because five systems are involved.
When roundups say “autonomous agent, no human,” that claim fails the hub’s question about whether a human was still in the loop. The person who typed the ID is the first step.
Takeaway: If a human has to start the write, say so. Faster handling can still be a real win.