← Selected work
CASE STUDY 01 · BPO SAASPRODUCT DESIGN · AI CALL ANALYSIS

The problem

How might a BPO team understand agent performance and act on call quality?

Teams needed to review calls scored by AI, help human agents improve, and give managers a clear way to follow up. I designed an MVP that connects those tasks across five levels of access.

From call score
to better action.
MY ROLESole Product Designer
DESIGN INPUTSStakeholder discussions · MVP requirements
STATUSDesign complete · In development

Pre-launch project: the role needs and workflows came from stakeholder input. There was no product user data to validate yet.

01 · Role-based views

One call. Five levels of responsibility.

Each portal focuses on the decisions that person needs to make. Clients can add custom roles at the level that fits their organization.

  1. 01Tenant adminOversees all client accounts.
  2. 02Client adminSets up teams and access for one client.
  3. 03ManagerTracks performance across supervisors.
  4. 04SupervisorReviews calls and coaches agents.
  5. 05Call agentSees call scores and review feedback.
Supervisor dashboard with team filters, call quality score, agent performance, and follow-up agenda.
Supervisor portal concept · Sample data
02 · QA calibration

Let people check the AI—and teach it from the result.

AI-scored calls can be reviewed by a separate calibration team. This is especially useful when a new program starts and the team needs to see how well the scoring fits its criteria.

01
Review a scored call

The calibration team checks the call and the evidence behind its score.

02
Add a human score and note

Reviewers record where their assessment differs and why.

03
Send the result to model training

The calibrated data feeds directly into training, creating a clear improvement loop.

QA calibration screen comparing reviewer and AI scores by call criterion, with fields for a consensus score and remarks.
Consensus workbench concept · Sample data
03 · Workflow automation

Turn a call signal into a clear next step.

Instead of manually chasing every quality finding, teams can define what should happen when a call meets a condition. Here’s one example from the workflow builder.

WORKFLOW CANVASFollow-up for low-scoring calls
Draft
WHEN A CALL IS SCOREDAI quality scoreTrigger · New evaluation
CHECK CONDITIONScore is below 70Only continue when true
THEN CREATECoaching reviewAttach call and score
AND NOTIFYSupervisorAdd to follow-up list
04 · Outcome

A complete MVP design, moving into development.

I completed the requested designs for role-based portals, call calibration, and workflow automation. Development is nearing completion.

The product is pre-launch, so there are no usage or performance results to report yet.

Back to selected work
Up nextSchool OperationsRead it →