Sarah, our AI floor manager, was asked one simple question: what is he doing that nobody else is? Two days later she had the answer. It was not a sales trick. It was one man, calling in again and again, a new "homeowner" every time.
He joined our solar floor at the start of the month. By week three, 13 of the last 22 inbound appointments on the whole campaign were his.
His booking rate on incoming calls was 2.0%. Everyone else on the floor: 0.25%. Eight times better than a room full of good reps, on the same calls, with the same script.
Every manager has dreamed of this rep. You do not question him. You study him, and you teach the floor to do what he does.
So that is what we asked Sarah to do.
Sarah started where she always starts: not with the pitch, but with the data behind every call.
When a homeowner calls a solar floor, it is almost always because we called them first. They missed the call, saw the number, and rang back. On this floor, about 85% of incoming callers are people our dialer had already called.
Then Sarah looked at his 13 bookings.
Zero. Thirteen people "called back" a business that had never called them. And it got stranger:
Sarah stopped looking for a sales secret. She opened a case file.
Our dialer rotates the caller ID it shows. A number that rings homeowners on Monday may be gone on Tuesday. To call one back, you have to see it first.
Sarah went through every call the rep had ever made by hand. One phone number stood out. He had typed it in himself in his first week. It was on no client list. He dialed it 9 times, and every call lasted 0 to 6 seconds. Never a conversation.
Each of those short calls flashed that day's caller ID onto the other phone. Then the "homeowners" started calling it.
Each row is one day. The blue diamond is his 0-6 second dial to the private number. The yellow dots are the "homeowner" calls that followed, on the exact caller ID that dial had just shown. Hover any mark.
| Day | Call-ins after the secret dial (minutes) |
|---|
All 16 caller IDs used by the fake bookings had just been shown by one of his secret dials. No other rep on the floor showed those numbers by hand all month. The autodialer had shown them only 0 to 5 times in the hour before, almost all to answering machines.
Our phone system sends each incoming call to the rep who has been waiting longest. At every one of the bookings, 2 to 6 reps were free. He was the one waiting longest in 12 of 13.
By chance, that happens about 1 time in 110,000.
So how? Sarah found 14 more calls on his line: short, silent, 3 to 29 seconds, from numbers we never dialed. They came in bursts, minutes after a secret dial. Somebody was ringing in, listening, and hanging up, again and again, until the call landed on the right desk.
Every real call carries a little of the world in it: a fan, a TV, a dog, traffic, a kid in the next room. Sarah measured the silence between the caller's words.
Share of calls
Share of calls
On 7 of the calls, the silence sat on one frozen value: the idle signal of a phone code used outside North America. US and Canadian phones use a different one. Not one of 70 normal callers had it.
The voices also reached the very top of the phone band, which happens when a call starts on a high-quality internet app and is squeezed down later. And the 13 calls came from just two calling setups that took turns by date. Not 13 homes. Two.
These odds overlap, so we never multiply them together. Each one alone is far outside normal.
Then Sarah did the thing that ended it. She took the caller side of every suspicious call, cut 40 seconds from each, and joined them into one recording. Then she asked a speaker-counting system a simple question: how many different people are talking?
Each circle is one call. Calls that share a voice are drawn together.
Fifteen real homeowners came out as 15 voices. The fifteen "homeowners" on his line came out as 1.
Sarah did not stop at one system. She ran eight different voice-matching models, calibrated on our own phone calls. She compared every pair of the 15 calls: 105 pairs. All 105 leaned "same man". Out of 10,878 pairs of real, different men, not one scored as high as a typical pair on his line.
She screened 12,082 recordings from the whole floor for that voice. It showed up on exactly 15 calls, all his. On 0 of 657 other reps' calls.
And it was not the rep himself putting on an accent. His own voice, on the other side of the same calls, did not match.
Sarah transcribed 7,877 of the rep's calls, every one of 8 seconds or more that could be played, plus 1,000 calls from other reps to compare.
Of his 27 appointments, 16 were fake. Sixteen people who never existed, booked into a real company's calendar.
His contract was ended for cause the next night. His logins were already off. Every measurement and every chart sits in a sealed evidence file with a fingerprint on each of its 79,846 files.
And the price of the whole forensic run, the rented graphics card plus the phone lookups? About $9. Less than a pizza.
Early on, one test produced a jaw-dropping "1 in 10 to the 33rd power". It looked great. It was also the wrong math: it multiplied results that are not independent. So we withdrew it and used only numbers that stand on their own. If a check could not prove something, we say so below. That is how we work, and it is how we will check your leads.
Including the ones that found nothing. A check that comes back empty is still a check.
Sarah is our AI floor manager. She runs the floor with our human managers: she reads the dialer, listens to the calls, coaches the reps and runs the morning huddles.
A human manager sees a star rep and feels lucky. Sarah sees a star rep and asks why, then checks every number until the answer is solid. She does not get bored on call 7,000. She does not like or dislike anyone. She follows the data.
This time the data led to a man who did not exist, sixteen times.
If you pay for appointments, live transfers or "booked" calls, some of them may be staged. Your vendor will not tell you. Your star rep will not tell you. The calls will.
We match every "callback" to the dialing that should have caused it. People who appear from nowhere get flagged.
Cell, landline or a free app number. Where it is from, and whether that fits your territory.
We trace the caller ID a caller rang back and who showed it to them, call by call.
Which rep gets the calls, how often, and how likely that is by chance.
Room sound, line codes and app fingerprints, measured in the silence between words.
We compare your calls with each other, so one person playing ten customers shows up as one.
Phrases, answers given before the question, and word habits that repeat across "different" people.
Every flagged call, why it was flagged, how sure we are, and what to do next. No guesswork, no hype.
Voice checks compare calls with each other. They do not name anyone and they are not lie detection. We run them only on recordings you own, and only where the law in your state allows.
