Concurrency: Why One Receptionist Can't Take Six Calls at Once

Concurrency: Why One Receptionist Can't Take Six Calls at Once

Last winter I sat in the back office of an HVAC company in Birmingham on the first properly cold Tuesday of the year. Their receptionist, Dee, was very good at her job. Between 8:02 and 8:40 that morning the phone system logged 23 inbound calls. Dee answered nine of them.

Nobody had done anything wrong. The business had simply hit a ceiling that has a name, and the name is call concurrency.

What call concurrency actually means

Call concurrency is the number of phone conversations a system can hold at the same moment. A human receptionist has a concurrency of one, and that isn't one per hour, it's one at any given second. While Dee was booking a boiler replacement, the other fourteen callers got a few rings and then her voicemail greeting. Most of them didn't call back.

That's it. That's the whole mechanism, and I'd argue it's the most useful thing an owner can understand about their own front desk.

Here's what most people get wrong

Owners treat missed calls as a staffing problem. Hire another person, buy back some hours, problem solved. The reason that fix disappoints is that phone demand isn't spread evenly, and staffing is.

Call volume arrives in clumps. The first cold snap of the season does it, so does the Monday after a bank holiday, and so does the morning somebody shares your Facebook ad in a local community group. A. K. Erlang worked this out for the Copenhagen Telephone Company back in 1917, and the queueing formulas that carry his name are still what serious contact centres use for rostering (Erlang's 1917 paper is the foundation of call centre staffing maths)).

The uncomfortable part of Erlang's maths is that answer rates fall off a cliff rather than sliding down gently. Adding a second receptionist to a one-person desk does not cut your missed calls in half. It does far better than that during ordinary hours and still leaves you exposed at peak, because peak is where a small number of agents gets overwhelmed fastest.

So you're choosing between paying for idle capacity all day to survive forty minutes of Tuesday, or accepting that you lose those forty minutes. Most businesses quietly pick the second one and never put a number on it.

What our call data shows

Across the home services and clinic accounts we run at AI Placers, the pattern is boringly consistent. Somewhere between a fifth and a third of a typical week's inbound calls land in windows where two or more calls overlap. On the seasonal businesses, that concentration is worse. The Birmingham HVAC account had one 38 minute window carrying more call volume than the entire afternoon.

A voice agent's concurrency limit isn't one. It's whatever number you provision, and the marginal cost of the sixth simultaneous call sits close to the cost of the first. That's the actual difference, and it has very little to do with tone or scripting or the whole "AI versus human" argument people want to have. The difference is that six callers at 8:15 all get a conversation, instead of one getting a conversation and five getting a beep.

We ran a database reactivation campaign on the same account as a test: 30 warmed inboxes on the email side and a parallel calling motion. The email leg returned a 6% reply rate, which was fine. The calling leg outperformed it, and the reason had nothing to do with the script. It was that the calls went out and got picked up in batches rather than one at a time.

The three numbers to pull before you change anything

Open your phone system's reporting and find these:

Simultaneous call events per week. Any moment where call two arrived while call one was live. Your VoIP provider almost certainly logs this. If you're on a plain BT line, you won't have it, and that's its own finding.

Answer rate inside your busiest hour, compared to your daily average. If the gap is more than about fifteen points, concurrency is your bottleneck and not your staffing level.

Repeat-caller rate. What share of missed callers ring back within a day? On the accounts I've looked at, it's lower than owners expect. People who need a plumber at 8am call the next plumber, not you again.

Run those three and you'll know in twenty minutes whether this post applies to you. If your busiest hour answers as well as your average hour, you don't have a concurrency problem, and you should spend your money elsewhere.

Where concurrency stops being the answer

I'd rather say this plainly than sell past it. Concurrency solves simultaneous arrival. It does not solve a caller who needs judgement, a complaint that needs de-escalating, or a quote that depends on something only your senior engineer knows. A voice agent that takes all six calls and mishandles four of them is worse than a voicemail, because voicemail at least doesn't promise anything.

The version that works is narrow. The agent takes the calls a human would handle identically every single time: bookings, qualification, capturing details, routing to the right person. Anything outside that gets a warm transfer, or a callback commitment with an actual name attached to it. We've written more on what AI receptionists still get wrong if you want the honest version of the limits.

What this is worth in money

Take the Birmingham numbers. Fourteen missed calls in one morning. Say a fifth of them would have booked, and the average boiler job is a few hundred pounds. That's one morning, in one week, at one company.

You can run the same arithmetic on your own business in about five minutes, and we laid out the full method in our AI receptionist ROI breakdown. If the answer is small, ignore all of this. If it's big, the interesting question isn't whether to fix it but how fast you can.

Frequently asked questions

What is call concurrency?

It's how many phone conversations can happen on your line at the same time. A human receptionist handles one. A provisioned voice agent handles as many as you set it to.

How many concurrent calls does an AI receptionist handle?

It's a configuration choice rather than a hard limit. Most small businesses are fine with six to ten. Seasonal trades and multi-site clinics usually want more headroom for the two or three peak windows a week that actually matter.

Does concurrency cost more per call?

Not meaningfully. Pricing is usually per minute or per call, so ten simultaneous calls cost roughly what ten sequential calls cost. That's the economic difference from hiring, where you pay for the seat whether it's busy or not. Our pricing page has the current numbers.

Can't I just add a second receptionist?

You can, and for steady all-day volume it works well. It works poorly for spiky volume, because you're buying eight hours of capacity to cover forty minutes of demand.

What happens to the calls my voice agent can't handle?

They should transfer to a human or book a callback with a specific person and time. If your provider's answer is "it tries its best", that's the wrong provider.

How do I know if concurrency is my real problem?

Compare your answer rate in your busiest hour against your daily average. A wide gap means overlapping calls. A narrow gap means your issue is total hours covered, which is a different fix.

Does this apply to outbound calling too?

Yes, and more sharply. Outbound is where concurrency compounds, because you control the volume. A missed-call text-back compared against an instant callback shows the same effect from the other direction.

Will callers know they're talking to an agent?

Some will, some won't, and you should tell them either way. Disclosure costs you almost nothing and being caught out costs you a customer.

Do this next

Pull your simultaneous call events for last week. Just that one number. If overlapping calls are under 5% of your volume, close this tab and go work on something else.

If they're over 20%, don't rebuild your whole phone setup. Point something at the overflow on your worst hour and leave the rest alone. Our 15-day free trial exists for exactly that, and a fortnight is long enough to catch at least one of your bad Tuesdays.

Sources: A. K. Erlang's 1917 queueing work via the Erlang unit reference); internal AI Placers call data from home services and clinic accounts, 2025 to 2026.