An uptime figure with no method is decoration
An uptime figure nobody can verify, defined by the party quoting it, is not information. What matters is where a call goes when part of the system fails.
The short version
- Three questions dismantle most uptime claims: measured how, downtime defined as what, and what happens if it is missed?
- For a phone product, partial failure matters more than total failure and no uptime figure captures it.
- For a phone line the better protection is a fallback: calls fall through to a number you nominate, so an outage becomes a ringing phone.
- The useful question is not the percentage. It is what happens to a call when something goes wrong.
What a percentage does not tell you
Measured how, and by whom? A vendor pinging its own health-check endpoint from its own network will produce a very different figure from one measuring whether real calls connected and completed. Both can be described with the same word.
Downtime defined as what? If the system answers but the transcription fails, is that downtime? If calls connect but bookings do not write to the calendar, is that downtime? For a product like this the interesting failures are partial, and a single availability percentage is blind to all of them.
And what happens if the number is missed? An uptime figure with no service credit and no contractual consequence is a statement of intent. That is not worthless, but it is not a commitment either, and the two get presented identically.
What GreetKeeper does when something fails
Forwarding happens at your carrier, so a carrier-level fallback to your own number keeps working whatever happens further down the line. It takes a minute to set and one test call to prove.
Transfers have a second number. When the first person doesn't pick up, the call tries the backup you nominate, then the assistant takes a full message.
Every answered call is stored with its transcript and summary. The webhook retries with backoff when your system doesn't accept a delivery, so a brief outage on your side doesn't lose the record.
The question to ask instead
What happens to a call when your system is down? For a phone product there is a genuinely good answer available: the call falls through to a number the business specifies, so an outage degrades to a ringing phone rather than to silence. Ask every vendor whether that failover exists and how it is configured.
Ask also what happens when a component fails rather than the whole system. If the calendar write fails, does the caller get told, does the business get told, or does the booking simply not exist? A silent failure is worse than an outage because nobody goes looking.
Those two answers tell you more about engineering culture than any percentage will, and unlike a percentage you can test them.
Hear it handle one of your own calls
Your scenario, your greeting, a couple of minutes.