AI Voice AgentConversational AI0 → 1Founding PM
ANIKA — Multilingual AI Voice Agent
Re-architecting lead pre-qualification from a headcount problem into a system capability — enabling instant outreach, consistent qualification, and 20× cost efficiency at peak admissions scale.
A funnel that was busy but not learning
Emversity generated 6L+ leads per admissions cycle at a 0.5% lead-to-sale conversion. Over 40% were never meaningfully contacted. Response times stretched to 3.5–6.7 hours during peak season. 1.73 lakh leads saw fewer than 3 dial attempts in six months — despite the company paying to acquire every single one.
Agents handling ~20 leads/day were forced into high-volume, low-judgement screening that crowded out time for actual counselling. 60% monthly attrition made linear headcount scaling not just expensive — but operationally fragile.
"Pre-qualification wasn't a staffing problem. It was a system design failure. The organisation needed a dedicated, always-on qualification layer that separated volume screening from high-value human counselling."
A 24/7 qualification layer,
not a sales replacement
ANIKA — Multilingual AI Voice Agent was scoped deliberately — qualify and route, never sell. Its job: contact every lead instantly, enforce retry discipline, extract a consistent qualification signal, and hand only pre-qualified leads to humans.
01
Instant outreach
Contacts every lead within 20–30 seconds of acquisition. Enforces retry discipline across all channels, around the clock.
02
Qualification engine
PQ scoring across 4 dimensions: interest, program fit, availability, and fee acceptance. Score 0–100. Threshold-gated routing.
03
CRM intelligence
Writes structured outputs — PQ status, program, campus, disposition — directly into LeadSquared. Clean signals for ISTs.
Phased rollout — each gate is metrics-driven, not calendar-driven
Phase 1: cold & old leads
→
Phase 2: inbound channels + CRM writes
→
Phase 3: all sources + 24×7 inbound
Pilot results — bot vs 250 human ISTs
20-day pilot · ₹8L bot cost vs ₹62.5L human cost · Every number is real.
20×
Cheaper per PQ
₹391 vs ₹2,307
<30s
Time to first call
vs 3.5–6.7 hrs
13×
Higher PQ rate
6.03% vs 0.47%
99.8%
Lead persistency
vs ~40% human
↓92%
Inbound miss rate
37% → 5%
₹85K
Blended CAC
down from ₹1.2L
2,880
Enrolments enabled
vs 3,000 target
₹8L
Pilot cost (20 days)
vs ₹62.5L humans
End-to-end ownership,
from diagnosis to scale
Sole PM throughout. Started with problem diagnosis — CRM and dialer analysis revealed 1.7L+ leads with fewer than 3 attempts and TTFC stretching past 6 hours. Evaluated and rejected alternatives (predictive dialers, outsourced calling, expanded inbound) before committing to a product-led solution. Defined ANIKA — Multilingual AI Voice Agent's scope boundary deliberately: qualify and route, never sell. Led cross-functional rollout with sales ops, engineering, and leadership. Managed sales skepticism through off-peak pilots before peak-season expansion.
What this taught me
about building at scale
01
Accuracy over capability was the right call Constraining ANIKA to qualification — not selling — protected downstream conversion quality and sales trust during rollout.
02
The hardest problem wasn't technical it was change management. Sales managers feared replacement. Framing ANIKA as capacity amplification, not agent replacement, unlocked adoption.
03
Define boundaries early Pre-qualification needs to be a hard product boundary with PQ-to-enrolment success metrics baked in from day one — not validated downstream after launch.
04
Speed is the real moat In a seasonal, high-volume business, the real moat isn't lead acquisition — it's speed and consistency of first contact. ANIKA made that a system property, not a human one.