How a CBSE Publisher Verifies Teachers and Delivers Specimen Books with AI Calls

The publisher

A CBSE exam-prep book publisher that sells sample papers and question banks across India. Its growth depends on teachers: when a teacher likes a book, their students buy it. So each season it sends free specimen books to teachers.

The challenge

• Free books attract the wrong people. A specimen programme only pays off if the book reaches a practising teacher of the right class and subject.
• Thousands of teachers, short season. Book decisions happen in a narrow window, and a sales team can't call every name in time.
• Language. Teachers across the Hindi belt switch freely between Hindi and English, often in the same sentence.
• Did it arrive? After dispatch, nobody knew whether the book reached the teacher or what they thought of it.

What was built

Two outbound AI voice agents that speak natural Hinglish and switch to Hindi or English to match the teacher:

• Teacher outreach agent. Confirms the teacher, explains the free specimen book, checks they genuinely teach, captures class and subject, and registers them. It answers objections like "are you an AI?" and "how do I trust this?" honestly.
• Post-delivery follow-up agent. A short call after dispatch: was the book received, was the gift received, a quick highlight of what's inside, any requirement, and the next step.

What gets measured on every call

Every call is scored automatically after it ends, so the team sees the campaign without listening to recordings:

• Outcome: registered, not interested, not a teacher, callback, wrong person, no answer, dropped early.
• Why calls end early, and whether the listener suspected an AI and at what point.
• Teacher verdict, script adherence and, on follow-ups, delivery status and a call score.
• Unanswered dials are counted too, so connect rates are real, not flattered.

What made it work

• Verify before you give. The agent qualifies the teacher before offering the free book.
• Score after the call, not during it. Checks run on the finished conversation, which keeps the live call natural.
• Measure every agent. The same scoring runs on the follow-up agent, not just the first one.
• Plan for Hinglish transcripts. English terms can come back written in Hindi script, so qualification doesn't depend on exact English keywords.
• Short follow-ups. The post-delivery call is designed to finish in under a minute.

Results

This deployment went live in August 2026 and the season is ongoing. We'll add rounded, dated results here once the campaign data is reviewed with the publisher.

Related

• Download this case study as a one-page PDF
• AI voice agent for education
• How a publisher reaches teachers with AI calls
• Olympiad registration reminder calls
• Another real deployment: solar site surveys in Bengali