Contact center simulation
Lightning Lab
Design and test voice assistants, rehearse service procedures with agents, and give engineering evidence to scope a pilot.
Proof of concept
The problem
Operations, QA, and engineering need a shared way to test a service design before investing in an enterprise integration. They need to see what happened in the conversation, what action followed, and which version produced the result.
Design decisions
I designed two workflows around the same service procedure. Supervisors can test a voice-assistant draft and trace the findings back to the exact version. Human agents can rehearse the procedure separately, with a supervisor reviewing their work before publishing coaching.
What it does
Versioned voice-assistant design
Turn service policy and customer intent into a saved flow with approved knowledge, verification rules, permitted actions, and criteria QA can test.
AI customer tests and phone preview
Run simulated customer conversations or put an exact saved draft on the controlled phone test line. Formal flow acceptance requires passing tests for that draft and human review.
Evidence tied to the draft
Filter phone calls and AI tests by flow and time period. Open transcripts, recorded service actions, and assessments to trace findings back to the draft that ran.
Guided agent practice
Assign a fictional contact. Agents verify the account, check service availability, prepare a simulated transfer, and save their wrap-up notes.
Supervisor-reviewed coaching
Review the agent’s actions and notes against a rubric before publishing feedback. Agents can see their own evidence and published coaching.
Build Package for engineering
Download the product brief, technical specs, architecture and integration spec, and Jira specification with backlog and acceptance cases to scope and test a pilot.
Screenshot tour
Design a voice flow
Define the service goal, procedures, and approved knowledge. Then run AI customer tests or preview the saved draft on the controlled test line.
How AI helps
Find what breaks before a pilot
AI helps draft the flow, generates fictional customer cases, and tests the saved version in multi-turn conversations. Reviewers use the transcript, recorded service actions, and assessments to decide what to revise and retest.
Example
In Move Internet Service, QA can check whether the assistant verifies the account, checks options for the new address, reads back the move details, and confirms consent before preparing a simulated transfer.
What people decide
QA reviews passing test evidence for the exact draft before formal acceptance. A phone preview can run under lab limits before acceptance. Supervisors publish coaching separately for human agent practice.
Proposed enterprise pilot
The proposed design separates live call handling from queued evaluation and reporting so each can scale and recover independently. The pilot would use a dedicated Genesys training queue and synthetic service data. Enterprise routing, archive access, and real service-system writes are proposed integrations.
What the pilot would measure
At agreed peak and burst volumes, measure call completion, response time, evaluation backlog, and cost per completed session. Test recovery from timeouts and duplicate or missing callbacks. Compare time to reviewed findings, test coverage, and design rework with the existing process.
Built with
- Next.js
- Vercel
- Supabase
- Twilio
- ElevenLabs
- OpenAI
Next application
Build Factory