On 8 October 2026 I started building an AI telemarketing agent at Zaman IT. This is a working note, not a retrospective — the system is still in progress and I am not claiming any results yet.
The problem
Most outbound calls end in a no or an unanswered phone. A salesperson's time is best spent on the few people who are actually interested, but someone still has to find them.
Start narrow
The agent is given one job: ask whether the person is interested, and answer basic questions about the company if they come up. It does not negotiate or sell. A small scope is easier to train, easier to test and much easier to trust.
The flow
- 01Upload the contact database.
- 02Train the agent on the company's information.
- 03The agent calls and checks interest.
- 04Interested: alert the sales team with the number, recording and transcript.
- 05Busy: ask for a good time and call back then.
The record matters as much as the call
Handing a lead to a human is only useful if they know what was said. So every call is kept end to end: outcome, recording, transcript, callback requests and attempts. The salesperson opens the lead already knowing the context.
Callbacks are a product feature
A busy contact is not a rejection. Letting the contact choose the callback time — with a sensible retry window when they do not — turns a wasted call into a scheduled one.
Status
In progress. I will add follow-up notes as the call logic, the training flow and the sales alerting take shape.
Written by Huzaifa Ahmed