Uptime Budget Sheet

What each tier allows

DOWNTIME BUDGET

period x (1 − uptime %)

99% a year

3.65 days

99.5% a year

1.83 days

99.9% a year

8.76 hours — about 43 min a month

99.95% a year

4.38 hours

99.99% a year

52.6 minutes

99.999% a year

5.26 minutes

⚠ Each nine

divides the budget by TEN

⚠ Credits refund FEES

not what the outage cost you

Incidents this period

This period

Service

Period

Uptime target

BUDGET (min)

Downtime so far

BUDGET LEFT

ACHIEVED %

Partial outages counted how?

Provider's figure

Our measured figure

Credit due?

Before agreeing an uptime target

• Percentage converted to MINUTES for the billing period
• Target checked against how quickly a human can realistically respond
• Past three nines? Automatic failover budgeted, not just more care
• Definition of downtime read in the contract, not assumed
• Scheduled maintenance exclusions checked
• Remedy read — it is a fee refund, not compensation
• Partial and degraded service counting agreed IN ADVANCE
• Independent monitoring in place from outside the provider

The third nine is where it changes

99.9% allows 8.76 hours a year and can be run by people being careful. 99.99% allows 52.6 minutes, which one 3am page can spend entirely before anyone has diagnosed anything. Past three nines the answer is automatic failover, and the cost is architectural rather than operational — which is a budget conversation, not a discipline one.

An SLA is a refund policy

Missing the target usually means a percentage credit against your fee for that period, not payment for what the outage cost you. If an hour offline costs far more than an hour of hosting — and it usually does — the SLA is a statement of intent, not insurance. Redundancy you control is the only thing that actually transfers that risk.