Emversity's inside sales team was scaling in headcount — but not in outcomes. With 40% monthly attrition, only 3–5% of calls audited, and FIFO-based calling that couldn't distinguish high-intent leads, the funnel was busy but not learning. Mis-selling ran at 13%. Refund rate at 12%. CAC had climbed to ₹95K. The bottleneck wasn't lead volume — it was invisible execution failure at scale.
← All projects
02 / 08
Featured project
AURA — AI Sales Intelligence Suite
Transforming an IST of 250+ agents from a people-dependent sales operation into a system that listens, learns, and guides at scale — before, during, and after every call.
The Problem
The core problem
The core insight: this wasn't a training problem or a hiring problem. It was the absence of an intelligent system that could listen, learn, and guide. Sales execution had to move from being people-dependent to system-supported.
The Solution
Building intelligence
AURA was built as three tightly integrated intelligence layers — each closing a different gap in the sales lifecycle.
01
AI Call Audit
100% call coverage via AWS Transcribe + GPT-4. 6-dimension scoring. Daily ISM reports. Replaced 5-member QA team.
02
AI Intent Prediction
Post-call LLM predicts Hot/Warm/Cold + cross-BU fit. Re-routed junk leads to right programs. Revived 68 admissions.
03
PitchMate
RAG chatbot embedded inside LSQ CRM. Real-time objection scripts, NBA, and program context — without leaving the workflow.
Product Interface
←
→
Impact at a glance
20x
QA coverage increase
3% → 100% of calls
↓70%
Mis-selling rate
13% → 4%
₹35L
Annual QA cost saved
6 FTEs → 1
↓42%
Sales cycle time
17–20 days → 12 days
65%
CAC reduction
₹95K → ₹70-75K
+48%
IST productivity
48 → 71 calls/day
68
Admissions revived
₹3.75 Cr revenue
+10pts
Agent confidence
2.8 → 4.4 / 5
My Role
End-to-end ownership
Sole PM end-to-end. Framed the problem as a system-level execution failure — not a training issue. Defined product vision, sequenced the 3-layer rollout with strict metrics-gated expansion, wrote specs, and coordinated data science, engineering, and sales leadership. Shipped in phased rollout — post-call audit first, then intent re-routing, then PitchMate — each layer unlocking only after the prior one stabilised.
Key Learnings
What this taught me
01
The hardest part wasn't the AI — it was ensuring agents trusted the system enough to act on it. Adoption required framing AURA as reducing call anxiety, not as surveillance.
02
35% of 'junk' leads were actually eligible for a different BU. The system wasn't losing bad leads — it was losing good ones by misclassifying them early.
03
Improving speed or productivity in a funnel that lacks intelligence just pushes more volume through the wrong decision paths. Fix the funnel first.