Parlay
AI sales coaching platform that turns every conversation into a performance feedback loop.
The System
A four-stage pipeline that converts live sales conversations into actionable coaching — automatically.
Every call flows through the same path: captured audio is transcribed via Whisper, scored against six skill axes by the LLM layer, and surfaced as structured feedback with concrete next-step actions — all before the rep finishes their post-call notes.
What It Does
Platform metrics and core capabilities.
Scores reps across 6 core skills in real-time — clarity, influence, objection handling, discovery, delivery, close
Transforms raw conversations into structured insights within seconds
Generates "double down" improvement actions with clear coaching plans
Feeds performance data into CRM workflows (Salesforce, HubSpot)
Key Decisions
Architecture and product choices that shaped the system.
Speech-to-Text + LLM Pipeline
Chained Whisper for transcription into an LLM for analysis rather than relying on a single end-to-end model. Two discrete stages give us control over each layer independently.
↳ Tradeoff: Two-step latency for better accuracy and debuggability.
Structured Scoring Criteria
Locked scoring to 6 fixed skill axes instead of allowing free-form feedback. Every rep is measured on the same rubric, every time.
↳ Tradeoff: Less flexibility, more consistency across the org.
Rep-First Design
Coaching is delivered directly to reps in the field — not buried in dashboards for managers. The person doing the work gets the feedback.
↳ Tradeoff: Less enterprise admin appeal, more direct impact on performance.
CRM Integration
Automated note generation and workflow triggers push structured data back into Salesforce and HubSpot without manual entry.
↳ Tradeoff: Integration complexity for seamless day-to-day workflow.