What a 99.9% availability commitment does not say
The figure is comparable from one offer to the next. What it covers is not: the measurement window, the excluded scope and the remedy alone determine its real reach.
A 99.9% availability commitment allows eight hours and forty-five minutes of interruption per year. The same figure, measured monthly, allows forty-three minutes. It is the same percentage and two different undertakings, and the gap between them appears nowhere in the comparison table of offers.
Three parameters that change everything
The measurement period. An annual rate allows a long interruption to be concentrated in a single month without ever breaching the commitment. A monthly rate forbids it. For an administration whose workload concentrates into a few weeks of the year the difference is decisive, and it costs nothing to negotiate at tender stage.
The excluded scope. Planned maintenance windows, outages attributable to site power, faults in the terminal equipment and force majeure almost always fall outside the calculation. On some contracts the exclusions add up to more downtime than the commitment covers.
The measurement point. An operator measures the availability of its service up to its own equipment. The site operator measures the availability of the application its staff use. Between the two sit the firewall, the switch, the UPS and sometimes a second supplier. Two honest measurements can differ by several hours without either party being wrong.
The remedy is not compensation
The usual contractual penalty is a credit proportional to the duration of the outage, capped at a fraction of the monthly subscription. For a service billed at a hundred thousand francs a month, a full day of interruption gives rise to a few thousand francs. That amount bears no relation to the real cost of the day lost to the organisation.
An availability commitment is therefore not an indemnity mechanism. It is an indicator of the standard the supplier holds itself to, and a contractual warning signal when it no longer meets it. Treating it as insurance leads to negotiating the wrong parameter.
What actually protects
The committed restoration time weighs more than the annual rate. An organisation absorbs four one-hour interruptions better than a single four-hour one: the second case passes the threshold beyond which degraded procedures stop being sufficient. That threshold is determined function by function, and it is written into the contract as a maximum response time and a maximum restoration time.
Then comes the escalation clause: who is reachable, at what hours, and after how long the case moves up to a higher level of decision. An on-call number that reaches voicemail on a Saturday is worth less than a lower annual rate backed by a real duty roster.
What to require
The monthly outage report, sent without being asked for, with the opening and closing timestamp of each incident. The explicit list of exclusions. The measurement point, named. And the right to terminate after a defined number of consecutive failures, which is the only remedy a supplier genuinely weighs.
