"Managed server" can describe almost anything: a fully operated system at one end, and at the other a provider who installed the operating system once and will answer an email if it catches fire.
The word is close to useless without a list of responsibilities attached to it. So here is the list we think it should imply.
Somebody is watching it
A managed system is monitored — availability, disk space, memory pressure, hardware health — and the monitoring only counts if the alerts reach somebody who can act on them.
A graph nobody looks at is decoration.
Routine maintenance has an owner
Operating systems and exposed services need security updates. Some can be applied routinely; others need testing, a maintenance window, or coordination with the application.
Managed should identify who tracks those updates, who decides when they are installed, and who checks that the service came back afterwards.
Backups include recovery
A provider should be able to say what is backed up, how often, for how long and where the copies are kept. They should also be able to explain how a restore works and how long it takes.
If recovery is entirely the customer's problem, that is a legitimate model — but it needs to be stated, rather than left behind a checkbox labelled "daily backup".
Hardware failure is not your ticket to the data centre
On dedicated infrastructure, disks and power supplies fail, and occasionally whole machines do. The operator should own the relationship with the facility and coordinate the replacement.
The customer should not need to learn a rack number because a drive died.
Capacity is a conversation
Management does not mean infinite resources. It does mean watching the trend and saying something before a predictable limit becomes an outage.
If storage has grown five per cent every month and will be full in six weeks, discovering that at one hundred per cent utilisation is not management.
Application management is a separate boundary
Managing a server does not automatically mean managing every application on it. We may own the operating system while you own the application, or we may own both.
Either model works. What matters is that the boundary is written down before there is an incident to argue about.
The word "managed" should reduce ambiguity. If it is creating more of it, ask for the responsibility list.