What is Convin & Rule Engine?
About Convin
Convin is an AI Agent Platform designed to optimize sales, customer service, and collection teams through three powerful agents:
AI Conversation Intelligence Agent → Analyzes customer-agent conversations across channels, audits calls, and delivers insights.
Live Agent Assist → Offers real-time prompts, objection handling, and resource support during calls.
AI Voice Agent → Handles outbound calling tasks autonomously.
The Rule Engine lives inside the Conversation Intelligence Agent, acting as the brain behind automated auditing — defining what should be checked, how, and when, all without code.
Why did we build it?
Context & Problem Framing
In contact centers, agents often struggle to perform consistently. Despite investing millions in coaching, most companies face:
Long onboarding ramps
Reliance on expensive external trainers or senior agents
Frustrated agents who hear only criticism, not actionable feedback
We saw a pattern:
Peer-clip based coaching (our existing method) wasn't helping many agents improve.
Agents who didn’t understand the core concept (like upselling) couldn’t apply peer tips effectively.
Some didn’t get relevant calls for days — causing them to forget what they learned.
Our hypothesis: Agents need contextual learning, delivered right after mistakes — not just tips or recordings.
Strategic Insight: “Sequence Matters”
While our initial idea was to show the agent where they went wrong first, user research said otherwise:
“Nobody wants to hear what they did wrong before they even understand what right looks like.”
Inspired by ed-tech patterns (Coursera, Khan Academy, Duolingo), we restructured the experience into a 5-step adaptive learning model:
Learn → Listen → Reflect → Test → Practice
Learn: GPT-generated content that explains the concept (e.g., “How to upsell”) based on client-uploaded documents
Listen: AI voice explains what good sounds like
Reflect: See a snippet from your own call where you went off-track
Test: Quick quiz to reinforce learning
Practice: AI mock call for instant application (GPT + TTS + dynamic prompts)
My Role Across Both Versions
Version 1: I acted as Product Owner + Sole Designer, directly working with the CEO. The goal was to get a basic coaching MVP live (peer-clip based).
Version 2: Led the design for the structured AI coaching upgrade. Worked with:
1 Junior Product Designer (who I mentored and later handed full ownership to)
Sr. Visual Designer
APM (step model concept originated here)
Devs, QA, and customer success
I led weekly critiques with my junior designer, paired on flows, and empowered her to own the Reflect step revamp.
I translated the core learning theory into actionable flows, structured learning states, and easy navigation.
Research & User Insights
Interviewed 10+ client teams (QA heads, managers, agents)
Shadowed onboarding + live coaching workflows across domains (sales, support, collections)
Key insights:
Coaches spent 2–4 hours daily manually assigning or re-explaining the same concepts
Agents preferred structured, short bursts of learning — not hour-long decks
Motivation dropped when learning wasn’t immediately applicable
Peer clips alone didn’t help: one QA head told us, “He keeps repeating the same mistake because he doesn’t even understand what upselling is.”
These led us to prioritize real-time application (mock call), dynamic pacing, and personalization.
UX Process & Frameworks
Our goal was to design for motivation, minimal friction, and retention.
Frameworks Used:
Goal-gradient theory → Visible progress indicators to increase completion
Chunking principle → Breaking content into micro-learning steps
Progressive disclosure → Show only one concept at a time
Mobile-first patterns → Recognized that many agents consume content during downtime on phones
Tone of voice tuning → We tested 3 TTS voices to find one that felt friendly and human
Accessibility Considerations:
Large tap areas
Support for regional languages and simplified English
Color-safe design across components
UX Challenges & Design Decisions
Challenge 1: How do we motivate agents to complete sessions without making it feel like a classroom?
Solution: Gamified progress indicators, contextual AI voice prompts, snackable UI per step
Challenge 2: How do we build feedback into the system without hurting morale?
Solution: Always show the ideal way before showing personal mistakes
Challenge 3: How do we balance autonomy for managers with self-serve agent learning?
Solution: Agents could auto-enroll via performance flags; managers could still assign manual sessions
What Didn’t Work:
Initially placed the mock call at the start (before concept) — completion dropped
Agents skipped longer clips — we trimmed all audio to <45 seconds
First TTS voice felt robotic — caused disengagement until we changed tone
Design Learnings:
Step-by-step format increased completion rates in pilot testing
A/B tested mistake-first vs. concept-first → 70% preferred concept-first
Simplified UX by borrowing from Duolingo and Coursera’s visual rhythm
Technical Constraints & AI Integration
Content was generated by GPT using custom prompts based on client-uploaded training material (from our LMS)
Voiceovers powered by TTS (AI voice), tuned for clarity and friendliness
Mock calls were dynamic (GPT + keyword-based branching + speech evaluation)
Results & Business Impact
While exact client metrics are under NDA, observed directional results:
+7.2% CSAT uplift for teams with consistent coaching engagement
~40% reduction in manual coaching time for managers
4.5 days faster onboarding for new hires compared to previous cohort
Coaching usage expanded to QA training, team leads, and onboarding cohorts
Behavioral Wins:
Peer-clip only sessions had <30% completion; structured flow hit >70% in week one
Manager-assigned sessions dropped from ~10/week to ~3
This shifted Convin’s positioning from just an analysis tool to a performance enablement platform, giving us new pricing levers and market positioning in sales demos.
TL;DR Summary
Project: AI Coaching & LMS – Learning System for Contact Center Agents
Role: Product Owner (V1), Lead Designer (V2), Mentored Junior Designer
Skills: UX Strategy, Learning Systems, AI Integration, Motivation UX, Agent Enablement, Cross-functional Collaboration
Impact:
Structured coaching model adopted across verticals (Sales, Support, Collections)
Reduced coaching time for managers by ~40%
5–10% agent performance uplift
Design-first evolution from clip-based training to full adaptive learning flow
