This month the voice agents on our platform crossed 10,000 calls. About 9,500 of those rang on live client lines, and those are the ones this post is about. That is a round number and I am not sentimental about round numbers, but it is a large enough sample that the patterns in it are no longer anecdotes. So I pulled every call record and read the whole thing the way I used to read a call center's monthly report.
A few ground rules on the data. Everything below comes from live client phone lines: plumbers, property managers, a mortgage broker, a restoration company, a concrete contractor, a landscaper, a consultancy. Our own number, demo lines and internal agents are excluded. Nothing is projected. Where a figure depends on an assumption I say so.
The phone is busiest when nobody is at the desk
Eighty-four percent of calls arrive between 8am and 5pm on a weekday, which sounds like the phone is a business-hours problem. It is not, for two reasons.
First, the busiest single hour of the day is noon. The window from noon to 2pm carries 21% of all calls. For a business with one person on the front desk, that is the window she is at lunch, on a job site, or eating at her desk while trying to close out the morning. The phone does not know that.
Second, 23% of all calls arrive outside business hours entirely. Twelve percent land after 5pm on a weekday, 8% on a weekend, 2% before 8am. Nearly one call in four reaches a business at a time when, before we arrived, it reached voicemail. For the restoration company in the sample the figure is 42%, because basements do not flood on a schedule. Its busiest day ever was a flash flood on September 4: 138 calls hit our client lines that day, more than any other day in the history of the platform, and the agent on that line answered every one of hers and sent the owner an alert with the address and the problem already captured.
Three quarters of calls never needed a person
Of 9,450 inbound calls, 2,399 were transferred to a human. Everything else, 75%, was handled start to finish by the agent: a message taken, an appointment booked, a question answered, a service request captured with the details a dispatcher needs.
I want to be careful with that number, because it is easy to read it as "the AI replaced three quarters of the phone staff." That is not what it measures. It measures how much of a small business's inbound volume is routine. Twenty years of running call centers taught me the routine share is always higher than the owner believes, because the owner only remembers the hard calls. The data says the owner is remembering about one call in four.
The transfer rate also varies enormously by business, and it should. The property manager transfers 44% of calls, because a tenant with a maintenance emergency belongs with the property manager for that building and the agent's job is to get them there with the right unit and the right zone attached. The mortgage broker's line transfers none; every caller's details are captured and he calls back himself. Neither number is better. The right transfer rate is whatever the operation decides it should be, and the point of a built agent rather than a bought one is that you get to decide.
Most calls are short, and that is the honest part
The median call is 29 seconds. The mean is 52. Twelve percent run two minutes or longer, and those are the ones doing the heavy lifting: the full service request, the pre-qualification, the tenant walking through what happened.
Thirty-nine percent of calls end inside 20 seconds. Some of those are robocalls and wrong numbers. Some are people who wanted a human, heard an assistant, and hung up. Some are callers who got their answer in one exchange. I count them all, because they are real calls that rang a real business, and because a human receptionist would have picked up every one of them too. Any vendor showing you an average handle time of four minutes on a small business line is showing you a filtered sample.
Callers are not angry at it
Every call gets a sentiment read after the fact. Across the full sample, 97% of callers are rated neutral or better and 14% are rated positive. Two hundred fifty-two calls out of 9,450, under 3%, were rated negative. That is a small enough number to review one at a time, and when we do, a negative call is usually telling us something about routing or hours rather than about the agent.
The growth line
In August 2025 the client lines we ran took 62 calls. In August 2026 they took 1,462. Most of that is new clients, and some of it is existing clients routing more of their numbers to the agent once they trusted it. Both are the same signal.
What I would tell an owner reading this
Three things.
Your after-hours volume is larger than you think, and it is disproportionately the urgent, high-value kind, because people with a routine question wait until morning and people with water in the basement do not.
Your routine share is larger than you think. Three quarters of what rings your phone is a message, a booking, or a question with a known answer. Staffing a person for that is expensive, and not staffing it means voicemail, and voicemail in a service business is a lost job.
And the number that matters is not any of the ones above. It is how many of your calls currently go unanswered. Most owners have never seen it. If you would like to, that is a conversation I am glad to have, and the first thing we do is measure it.