An availability figure is arithmetic, not a delivery promise
99.99% availability measured over a 30-day month permits roughly four minutes and twenty seconds of unavailability. That is the whole of what the number says. It does not describe how many messages reached handsets, how quickly, or whether the recipient acted on them.
That distinction matters because messaging spans systems with different owners. "Delivery guaranteed" cannot be a contractual commitment, because no provider controls the destination network, the handset or the regulator.
Where the measurement boundary sits
A gateway availability figure should name the components it measures. Typically that is the parts the provider runs:
- API ingestion: requests are accepted and queued.
- Routing: each message is dispatched to a route.
- Status processing: provider status updates are received and normalised.
- Callback delivery: status callbacks are dispatched to your endpoint.
Everything beyond that boundary belongs to someone else:
- Provider route and carrier behaviour. Filtering, throttling and network incidents happen outside the gateway. Detecting them and moving traffic is operations work, not availability.
- Destination network state. Congestion or an operator incident affects delivery while the gateway is fully available.
- Handset and recipient. Powered off, out of coverage, blocked sender, message ignored.
- Regulatory and registration changes. Re-registration of a sender identity or template re-approval can stop traffic without any gateway fault.
If an SLA does not state the boundary explicitly, it is not measurable by you.
Questions that determine whether the number is meaningful
- Which components are inside the measurement, in technical terms — API only, or something wider?
- What counts as unavailability: error responses, latency above a stated threshold, or full unreachability?
- Is availability calculated globally, or per region and per route? A global figure can absorb a long regional outage without breaching.
- What exclusions apply, and how is planned maintenance announced and counted?
- What is the measurement source, at what granularity, and can you see the underlying telemetry?
- How long is that telemetry retained for the purposes of a dispute?
- What is the remedy if the commitment is missed, what is it a percentage of, and what has to be claimed?
- What is the published response commitment for degradation that sits outside the measured boundary — a filtering change or a route problem?
- What evidence do you receive after an incident, and on what timeline?
- What visibility do you have into dependencies you do not control, such as route health and provider status?
Question 8 is usually the most revealing. Most real messaging problems are outside the availability boundary, so the response commitment for them matters more than the fourth nine.
Whatever a provider states in these areas — notice periods, check intervals, retention, credit mechanics, postmortem timelines — treat it as a commitment only where it appears in your agreement or a published service description. Flowstates' commitment is 99.99% gateway availability; anything beyond that should be confirmed contractually rather than inferred from an article.
Availability and operations are different products
The value of a managed gateway is mostly outside the availability figure:
- Continuous monitoring of routes and destinations against each programme's baseline.
- Escalation into providers and, through them, operators, during incidents.
- Routing changes when filtering or performance shifts.
- Registration and compliance handling as market rules change.
None of that compresses into an uptime percentage. It shows up as incident frequency, time to detect and application outcomes.
Supply model does not change the boundary
Flowstates sells and operates messaging routes, supports customers keeping their own provider contracts (BYOV), and supports hybrid arrangements. The measurement boundary is the same in each case: the managed gateway is measured on what it runs, and route and carrier behaviour sits outside it, whoever holds the supply contract.
The short version
Ask for the boundary, the calculation, the exclusions, the evidence and the out-of-scope response commitment. A provider that answers those has thought about operations. A provider offering a bigger number without a boundary has not.