The inbound line
The college's number was answered by an AI admission assistant with a knowledge base built from the prospectus, the website's course API, notices and scholarship pages: programmes, eligibility, fees, the scholarship matrix, recognition and campus details. Retrieval answers a question in 27–85 ms, so there is no pause before the AI speaks. Scholarship and programme questions were answered correctly in testing; a counsellor's number was set for warm transfer.
Round one: 3,941 enquiries
The opener named the college, mentioned the student's earlier interest in IT and asked whether they were still considering a bachelor's. Of 3,941 contacts, 422 turned into real conversations. We reviewed every one by hand: 52 follow-up leads, of which 12 had a next step agreed (6 campus visits, 3 counsellor calls, 2 will apply, 1 call-back), 16 were interested and 24 needed a follow-up. The admissions team got a password-protected report page with each lead's recording and a dial list.
Round two: 6,929 numbers in one afternoon
Six days later the college sent a larger follow-up list. The AI dialled all 6,929 numbers between 11:57 and 18:01 on about fifteen parallel lines, with retries the same afternoon.
| Outcome | Contacts | Share |
|---|---|---|
| Could not be dialled (wrong, dead or barred numbers) | 3,013 | 43.5% |
| No answer | 785 | 11.3% |
| Answered but hung up during the opener or said nothing | 2,196 | 31.7% |
| Voicemail | 16 | 0.2% |
| Talked | 919 | 13.3% |
919 conversations, 673 talk-minutes, 44 seconds on average. Hand-sorted: 51 leads: 14 with a visit or counselling time agreed, 21 interested (mostly "after Dashain and Tihar"), 16 asking for a call-back. About 300 said they were not interested and 36 had joined elsewhere, which is useful data too.
What the admissions team received
A report page with both rounds, a switch between them, every lead's tier, note and recording, and three CSV downloads (summary, leads, all calls) with Devanagari-safe encoding for Excel. The next morning the counsellors had a dial list of 103 names with the context of what each student had said.