Per-user pricing charges for each person supported regardless of how many devices they use. It has become the more common default because it matches how people actually work — a single employee with a laptop, a phone and a desktop is one person to support — and because it does not penalise a client for issuing better equipment.
Per-device pricing charges for each managed endpoint or server. It is simpler to count and audit, and it fits environments where devices outnumber people or run unattended.
Tiered pricing packages levels of service — a basic tier covering monitoring and patching, higher tiers adding security tooling, backup or strategy time. It suits providers who want to sell upward over time rather than negotiate scope client by client.
All-you-can-eat agreements bundle unlimited support inside a fixed fee, which clients like and which requires disciplined scope to margin.
What matters more than the model is what sits outside it. Every agreement has a boundary — after-hours work, project work, onboarding a new site, supporting an application nobody documented — and the profitable providers are the ones who wrote that boundary down before the first invoice rather than discovering it during a dispute.