An owner-led commercial insurance brokerage was losing nearly four in ten business-hours callers to rings that ended in nothing, and at lunch nobody answered live at all. We read every call leg, found the causes, rebuilt the answer chain with an AI assistant behind the people instead of in front of them, and measured the result with one engine before and after.
The client is an owner-led commercial insurance brokerage whose customers are mostly contractors — people who call because a job is waiting on a certificate or a quote. Nearly all of that demand arrives by phone. Before any advertising was switched on, we asked a narrower question: of the people who already call, how many reach someone? The answer came from the phone system's own call log, read call by call, and it was worse than the summary reports suggested.
Across the full weeks of June and July, 2,136 calls reached the main line during weekday business hours. 1,077 of them (50%) reached a person. 817 (38%) reached nobody and left nothing — no answer, no voicemail, no message. The lunch hour was the sharpest edge: a company rule sent every call between noon and 1 p.m. to a voicemail box, and across those two months not one of the 384 lunch-hour calls was answered live.
Lost callers rarely have a single cause. One was that lunch rule. Another was a time zone: the owner's extension was set to a zone that ignores daylight saving, so a 9-to-6 schedule ran an hour late and every call before 10 a.m. went to voicemail without ringing. The audit that found it found the same setting on 18 of 28 users. A third problem sat on the calls going out: 10 of the firm's 32 numbers were broadcasting the placeholder caller ID NO CALLED ID, the kind of signal that gets a callback labelled as spam. All three were found in call legs and account settings, not in a summary report.
The answer chain now ends in a person or a captured message instead of a dead end. The front desk rings first. What it misses goes to an AI assistant that answers in English or Spanish, puts callers through to the person they ask for by name, rings the whole team at once when a caller wants a quote, and otherwise takes a structured message. Before real callers could reach the assistant, it had to pass 14 scripted acceptance calls, including repeated attempts to extract pricing and coverage advice it must refuse. 12 passed; the two failures were fixed and re-tested, one of them on a live call. A fire drill before opening took the assistant's line down on purpose and exposed a dead end; a watchdog was built the same day and proven in a second drill, rerouting in seconds with zero customers touched. Every user was moved to the correct time zone, all 32 numbers now present one caller identity, 29 are registered with the major caller-reputation providers, and a 911 call from any office line now alerts the front desk by text and email, where before it alerted nobody.
The formula was written down and approved before the first post-change number existed. A call counts as reaching a person only when a human leg on a staff extension connects; an answer by the assistant never counts as a person. The same engine produces the weekly call report the owner receives, so the before and after on this page come out of one piece of code, read over the same line and the same hours. The pass bar was set in advance too: callers reaching nobody at or under 15%.
In the four full weeks from August 17, 936 calls reached the main line in business hours. 651 (70%) reached a person and 118 (13%) reached nobody and left nothing; the rest left a voicemail or a message with the assistant. Week by week, the share reaching nobody read 13%, 14%, 11% and 13% — under the bar every week. In the latest week, half of the calls that reached a person did so within 8 seconds. Lunch changed shape rather than disappearing: every lunch-hour call is now picked up by the assistant instead of a silent voicemail box, but in the week of September 7 only 4 of those 33 calls reached a person, and that hour still held 29 of the 61 calls that did not.
Call volume was lower after the rebuild — about 234 business-hours calls a week against about 310 in June — and the cause has not been established. Several routing changes landed inside the weeks measured, so this page reports their combined effect, not any single change. Four weeks is a start, not a season. Test calls into the support teams set up in mid-August are still owed. And registering numbers with reputation providers is a filing, not a result: whether carriers now label those numbers differently has not been measured.
| Metric | Value | Source |
|---|---|---|
| Callers who reached nobody and left nothing | 38% → 13% | Weekly call-report engine, pinned formula: 817 of 2,136 calls (full weeks, Jun–Jul 2026) vs 118 of 936 (Aug 17 – Sep 13) |
| Callers who reached a person | 50% → 70% | Same engine and populations: 1,077 of 2,136 vs 651 of 936 |
| Lunch-hour calls answered live, June and July | 0 of 384 | Call-log baseline, full months Jun–Jul 2026 (a lunch rule routed every call to voicemail) |
| Lunch-hour calls picked up by the assistant | 33 of 33 · 4 reached a person | Weekly call report, week of 2026-09-07 |
| Users whose schedules ran an hour late | 18 of 28 → 0 | Account audit with per-write read-back, 2026-08-14 |
| Numbers broadcasting NO CALLED ID | 10 of 32 → 0 | Phone-system API read-back, 2026-08-05 |
| Scripted acceptance calls passed before real callers | 12 of 14 | Acceptance matrix, 2026-08-04; both failures fixed and re-tested by 2026-08-10 |
| People alerted when 911 is dialled from an office line | Nobody → front desk (text + email) | Configuration read before and after, 2026-08-04 |
If your business runs on inbound calls, the cheapest audit you can commission is a read of your own call log: every leg, every hour, one formula written down before anyone looks at the result. Most owners have never seen the share of callers who reach nobody. It is usually the first number worth fixing — and it should be fixed before a dollar goes into ads that make the phone ring more.
No pitch deck. Just your CRM, your ad accounts, and a straight read on what the next 90 days could look like.
Leo No Brainer Media AI · replies in seconds A live demo of the assistant we install for clients.
