AI-Powered Agent Development

Client

Coaching & LMS

Year

2023-2024

Closing the loop from audited calls to structured agent growth.

Scope of Work

Learning Design
UX Research
Product Strategy

40%

less manager coaching time

4.5 days

faster agent onboarding

70%+

course completion

CSAT ↑

for coached agents

The problem

Rule Engine caught what agents were doing wrong. Live Agent Assist nudged them in the moment. But neither helped an agent improve over time — coaching remained manual, inconsistent, and dependent on manager bandwidth.

Convin’s early LMS had low completion, under 30%, and coaching that did not map back to what calls revealed agents needed help with.

The insight: sequence matters

Research showed the LMS was failing for a structural reason, not a content reason. It taught concepts after showing agents their mistakes, when the more effective order was concept first, mistake second. That sequencing change became the backbone of the redesign.

What didn’t work

Two early bets did not survive contact with real agents. Placing mock calls too early interrupted learning momentum. TTS voice for training scenarios undermined how seriously agents took the material, so we moved to real recorded call snippets. Both were cut rather than shipped half-working.

Results

Once the sequence and mock-call placement were corrected, completion rose from under 30% to more than 70%. Coaching time dropped by roughly 40%, and new agents ramped about 4.5 days faster.

What I’d take forward

The order information arrives in matters as much as the information itself. Cutting a feature that is not working is a design decision, not a failure.

“Seeing the correct approach before my own error made it land, rather than feeling like criticism.”

Test participant, during research

Product Designer, Learning Experience & Research

UX Research · Learning Design · Iteration Under Real Usage Data

UX Research · Learning Design · Iteration Under Real Usage Data