How many SIP trunks do you need?

Count concurrent calls, not people. A channel carries one conversation at a time, so the number you need comes from how many calls overlap at your busiest moment, plus a margin. For a ten-person business taking forty calls in its busiest hour, that is eight channels, not ten and not forty. Here is the arithmetic.

The mistake worth avoiding first: headcount and daily call volume are both the wrong input, and they fail in opposite directions. Sizing on headcount usually overbuys. Sizing on daily volume usually overbuys wildly. This is one of the sizing questions collected in SIP and networks.

Averages hide the only thing that matters. A business taking a hundred calls spread evenly over eight hours and a business taking a hundred calls of which forty land between nine and ten need different numbers of channels, and the daily total is identical.

So the measurement is a peak, and for most small businesses an hour is the right window. If your day has a genuine rush, a specific ten minutes when the school run ends or the clinic opens, use a shorter window, because a steady hour containing a five-minute stampede will under-report the stampede.

Two figures come out of it:

• Calls in the busiest hour. Inbound and outbound together. Anything that crosses to the public network counts, including calls forwarded out to a cell phone.
• Average call length, from answer to hang-up, including any time spent on hold. Most small-business calls run three to five minutes. If you do not know yours, look at a phone bill rather than guessing, because guesses cluster at three minutes and reality often does not.

Multiply the two, divide by sixty. The result is measured in erlangs, and one erlang means one conversation continuously in progress for the whole period.

40 calls x 4 minutes = 160 minutes of talking 160 minutes / 60 minutes of clock = 2.7 erlangs

Two point seven erlangs is the floor. It says that during that hour, on average, 2.7 conversations are happening at once. Three channels would be exactly enough if calls arrived on a perfectly even schedule, which they never do.

The idea and the unit are old, and standardised: measuring traffic intensity this way is the subject of ITU-T Recommendation E.500, "Traffic intensity measurement principles". Nothing about voice over the internet changed the arithmetic, only what the channels are made of.

Calls do not arrive evenly. They cluster, and the margin above your offered load exists to absorb the clusters. How large it should be is not a matter of taste; there is a model for it.

The relevant one is Erlang B, which assumes a caller who finds every channel busy goes away rather than waiting. That is exactly what a trunk does to them, and it is why Erlang B and not Erlang C is the right model here: Erlang C describes a queue, and there is no queue outside a trunk. There is nowhere to stand.

Read the figure as the answer to "should I just buy plenty". Going from four channels to eight takes the chance of an engaged tone from about one caller in six to about one in two hundred. Going from eight to ten takes it from one in two hundred to one in three thousand, for the same money per channel per month. The curve falls off a cliff and then flattens, and the flat part is where overbuying lives.

For most small businesses, targeting one blocked caller in a hundred is the right place to stop. Somewhere tighter is defensible if a missed call is expensive; what missed calls are actually worth is how to put a number on that rather than a feeling.

Four things people include that should not be:

Internal calls. Desk to desk, or extension to extension, stays inside your phone system and never crosses to the public network. It costs you nothing.

People. A ten-person team is not ten simultaneous conversations unless all ten answer at once, which for most small businesses happens approximately never. Our worked example has ten people and needs eight channels, and that is a coincidence of the numbers rather than a rule.

Phones. Twenty registered handsets on eight channels is a normal, correct configuration.

And one thing people leave out that should be in: calls forwarded out. A call that arrives on your number and is sent on to somebody's cell phone occupies two channels for its whole length, one in and one out. If you forward a lot, that alone can double your requirement.

If your phone system is hosted rather than sitting in your building, none of this appears on your invoice, because the trunks are the provider's and they have already been sized. What is SIP trunking explains which situation you are in.

The arithmetic still matters, but the question it answers changes. On a hosted system, capacity stops being about channels and starts being about people: not how many conversations can exist at once, but how many of them somebody can actually answer. That is the same offered-load figure fed into a different model, and our free Erlang C calculator does it, along with what one more person on the rota would change. How many people do you need to answer calls is the long version.

There is one thing to check on a hosted plan regardless of what the marketing says: whether there is a cap on simultaneous calls, and what happens at it. A published number is a number you can plan against. An unstated one is a number you discover on your busiest morning.

A caller who reaches a queue and waits produces a record. You can count them, look at how long they waited, and decide whether to do something about it.

A caller who reaches a full trunk produces nothing. The call never arrives at your system, so there is no record, no missed-call notification and no voicemail. From the caller's side there is an engaged tone or a message from a network they have never heard of, and no way to tell the difference between "they are busy" and "they are gone".

This is the strongest argument for a margin rather than an exact fit. The failure mode is invisible on your side and unambiguous on theirs, which is the worst combination a business phone system can have.

Ringfully runs the switch and the trunks behind it, so channel sizing is not on your invoice or your desk. Billing is per seat with the minutes included, and what a plan costs is the whole of it. If you are moving off a trunk you sized yourself, bring the two numbers from step one with you: they are still the right inputs, and they are the fastest way to say what your busiest hour actually looks like.