AI voice experience

Making real-time AI coaching easier to configure.

Aigent’s AI voice tech was powerful, but setting up the brain behind it was a bottleneck that made teams give up. I rebuilt the Real-Time Coaches flow and the live-call view so supervisors could configure and manage AI guidance on their own.

Try prototype View design system

Role: Product Designer · UX/UI & Design System
Timeline: Jun – Oct 2022
Team: Data Scientist · Product Manager · Engineer · Customer Success
Tools & skills: Figma · Design Systems · User Testing · Data Visualization

Real-Time Coaches: Configuration flow redesigned
Call Monitor: Live-call view redesigned
Design system: Shared interface patterns

Aigent

Coach Builder · The real-time coaching workspace

The problem

A powerful AI platform held back by a painful setup.

I led the UX/UI redesign of Aigent’s platform and rebuilt how its real-time AI Coaches get created, the exact step that was killing adoption.

Aigent’s voice technology worked, but setting up the brain behind it, the Real-Time Coaches, was a dense, technical bottleneck. Supervisors couldn’t build guidance quickly, so teams lost interest and gave up.

The business needed supervisors to deploy new coaching on their own, with a clear step-by-step flow and an instant read on whether their bots were actually working.

Make the operating state visible.

The redesigned workspace gives supervisors an immediate read on what is live, paused, or still a draft.

Redesigned Aigent Coach Builder with coach status, trigger totals, and feedback visible in one workspace. — Coach Builder · Status, trigger performance, and feedback in one workspace

What we found

The tech was powerful. The setup was killing it.

Customer Success feedback and user research pointed at the same culprit: an interface too dense for supervisors to configure.

Conditional logic was too confusing

Users struggled with the basic If/Then flow. Building a rule like ‘if a customer says X, and the call runs over Y minutes, then show message Z’ felt like a chore, not setup.

No clear status tracking

Supervisors couldn’t tell which coaching scenarios were drafts, which were in testing, and which were live on the floor.

Feedback was disconnected

Agent reactions were locked away from the creator workspace, making coaching quality difficult to evaluate and improve.

Testing

Supervisors needed to see the logic, not fill out forms.

Unmoderated tests with real supervisors validated the new flow and changed how I structured the builder. They needed to visualize how the logic connected as they built it.

Break up the massive setup form

Supervisors treated the old setup page like one long, overwhelming document and missed the logical steps needed to make the AI work.

I split the long form into a multi-step tabbed layout. Separating Triggers from Messages created a clear, natural sequence.

Aigent trigger configuration with separate Triggers and Messages tabs, trigger types, linked coaches, and a text pattern. — Coach configuration · Separate Triggers and Messages

Make live-call data scannable

The legacy call monitor was dense and monochromatic, so supervisors couldn’t quickly spot a call going awry.

I introduced high-contrast icons for bot sentiment, agent feedback, and live trigger status so an anomaly became obvious in under a second.

Legacy Aigent Call Monitor with a pale table and limited visual hierarchy. — Before · Dense and monochromatic

Aigent Call Monitor showing dates and times, agent IDs, coach trigger counts, and transcription states. — After · Trigger counts and transcription states

Fix the status confusion

A rough draft and a live coaching scenario looked identical, making it easy to confuse what was actually running.

I added a bold, color-coded status bar so supervisors could instantly distinguish draft, testing, and live coaches.

Aigent Coach Status control with Draft, Testing, and Live options and a definition of each state. — Coach status · Draft, Testing, and Live

Impact

The product context behind the redesign.

My contribution was the configuration workflow and live-call interface. Ubiquity’s current Aigent page publishes the ranges below for its wider offering. They provide product context, rather than measured outcomes of my 2022 redesign.

−25–40%: After-call workLess manual wrap-up after every call.
−10–20%: Average handle timeFaster average call resolution.
+8–15%: Customer satisfactionPublished CSAT increase range.
−20–30%: Manager escalationsFewer calls handed up to managers.
−30–50%: Time to proficiencyNew agents reached full speed sooner.

Ubiquity · Published Aigent ranges

Takeaway

Information architecture was the real bottleneck.

In complex enterprise tools, the hard part is translating dense business logic into steps that make sense to a human. Building the feature and component library together forced every component to earn its place and gave the team a stronger foundation to scale on.