The most valuable inquiries a service business gets do not arrive between nine and five. They arrive when something has gone wrong: a pipe at 11:40pm, a dead furnace on a Sunday morning, a roof leak during the storm that just woke everybody up.
Those callers are not shopping. They have decided to buy already. They just need somebody to answer, and they will accept the first company that does.
You cannot staff that. A night answering service costs more than the jobs it catches, and the people who work them cannot see your calendar, cannot qualify a job, and cannot book anything.
An AI front desk is not a robot pretending to be a person. It is a system that answers in under sixty seconds, asks the same five questions your best office manager asks, and puts real appointments on a real calendar.
What after hours costs
Count the hours you are actually uncovered. Nights, weekends and the middle of every job add up to roughly two thirds of the week.
Now the money. Say six after-hours inquiries a week currently reach voicemail or a dropped chat. Emergency work carries a higher ticket, so call it $1,100. If the agent converts even a third of them:
| After hours | Figure |
|---|---|
| Inquiries outside office hours, per week | 6 |
| Answered today | 0 |
| Booked by the agent at a 33% rate | 2 per week |
| Weekly value at $1,100 | $2,200 |
| Annual value | $114,400 |
Halve every assumption if you like. It still pays for itself several times over, and it does it on inquiries you have already paid to generate.
How the agent works, step by step
Here is the actual flow, in the order the customer experiences it.
What it should never do
The limits matter more than the capabilities, and any decent build starts with them.
- Quote a firm price on complex work. Ranges are fine when you have set them. Invented numbers cost you the margin or the job.
- Promise a specific technician. It does not know who is sick tomorrow, and a broken promise on the doorstep is worse than no promise.
- Argue with an unhappy customer. One temperature check, then a clean handover with the transcript attached.
- Handle anything outside its tested scope. If a scenario was not written down and run before go-live, the correct behaviour is to fetch a human. Deeper on this: the AI agent QA checklist.
An agent that never says let me get somebody for you has not been built carefully. It has been built optimistically, which is a different thing.
What changes once it is in
Before: the 11:40pm call goes to voicemail, gets heard at 8:15am, and by then the customer has been dry for six hours thanks to somebody else.
After: it is answered at 11:41pm, qualified by 11:44pm, and on the 7am schedule. You did not wake up. The customer never knew there was a difference.
The second change is quieter and larger. Daytime overflow disappears. When two calls land at once, nobody gets voicemail, because the second one is handled with the same five questions and the same calendar.
Your after-hours competition is not another company. It is voicemail, and voicemail has never booked a job in its life.
The measurable: appointments booked outside office hours. That line reads zero today. Two weeks after go-live it will not, and that is the entire business case in one number.
