AI VOICE EXPERIENCE
AIGENT
-25-40% After-Call Work (ACW)
Redesigning Aigent's configuration flow so teams can get their AI agents up and running without the usual setup headaches. I wanted to make the post-call data incredibly simple to look at and understand.
I led the the UX and UI redesign for Aigent's platform. The big challenge here was making it way less painful for people to build and manage AI voice tools. I completely rebuilt how the real-time 'Coaches' were structured, so supervisors could set up live agent guidance without having to jump through hoops or guess what step came next. Along the way, I cleaned up the visuals and packaged everything into a strict, unified design system to ensure there was a uniform experience across the entire platform.
While Aigent’s platform worked, setting up the 'brain' of the AI - the Real-Time Coaches - was a massive bottleneck. The setup process was dense, packed with steps, and forced people to have a deep technical background just to get through it. Because of this, supervisors struggled to build helpful guidance for their agents quickly. It took forever to get the system running, which meant teams were losing interest and giving up on using it.
BUSINESS NEED
We needed to speed up how fast teams could adopt and actually use the tool. The goal was to let supervisors deploy new AI coaching scenarios completely on their own, without needing to call in technical support every time.
USER NEED
Supervisors needed a straightforward, step-by-step workflow that takes the confusion out of building complex coaching rules, while giving them an instant, clear look at whether their bots are healthy and working right.
CALL TRANSCRIPT SCREEN
I ran a workshop with the team to break down exactly how supervisors were setting up these workflows. We mapped out the entire journey; from the moment they create a trigger to when it actually goes live on the call center floor. By looking at it step-by-step, we found exactly where things were breaking down: the system was overloading supervisors with messy, complex logic gates instead of actually helping them.
To fix this, I broke our game plan down into three clear steps:
STRATEGY
The supervisor defines what the AI should listen for, like specific keywords, voice triggers, or changes in customer sentiment.
ANALYSIS
We give them a side-by-side view that links the live call transcript directly to the AI-generated prompts, so they can see exactly how the coaching triggers in real time.
EXECUTION
The system packages all that behind-the-scenes logic into simple, prioritized 'Coaches' that agents can easily read and act on while they are on a live call.
PROBLEM TO SOLVE
Supervisors find it incredibly frustrating and time-consuming to build 'Real-Time Coaches' that use multiple triggers. The old setup flows are poorly structured so it’s hard to even figure out how the underlying logic works. Instead of helping supervisors guide their agents effectively, the system just slows them down and leaves them overwhelmed by the setup process.
Gathering Insights
Feedback from Customer Success and our own user research made one thing clear: the AI voice technology was incredibly powerful, but the setup process was killing it. The interface just wasn't intuitive enough for supervisors to easily define triggers - like specific keywords a customer might say - and link them to the exact message or guidance an agent needed to see.
Emerging Themes
CONDITIONAL LOGIC WAS TOO CONFUSING
Users struggled over the basic 'If/Then' flow. Trying to build a simple rule like 'IF a customer says X, AND the call goes over Y minutes, THEN show the agent message Z' felt like a chore rather than a straightforward setup.
NO CLEAR STATUS TRACKING
Supervisors were completely blind to what was actually running. They couldn't easily tell which coaching scenarios were just rough drafts, which ones were currently being tested, and which ones were live on the floor.
DISCONNECTED FEEDBACK
Feedback loops from the agents actually using the tool were totally separated from the creator workspace. Because agent reactions were locked away in a different place, supervisors found it nearly impossible to actually improve or iterate on the coaching quality.
BALTO (COMPETITOR) SCREENSHOT
HOW MIGHT WE
…use clean, familiar layouts to take the confusion out of building 'Real-Time Coaches,' so supervisors can launch live agent guidance without worrying about breaking something?
Testing the Solution
After a few internal design rounds, we ran unmoderated user tests with some of our current supervisors to see if the changes actually worked. The feedback validated that the new flow was way easier to follow. The biggest takeaway was that supervisors didn't just want to fill out a bunch of blank form fields; they needed to mentally visualize how the logic was connecting while they built it.
What We Learned & How We Fixed It
BREAKING UP THE MASSIVE SETUP FORM
Insight: Supervisors were treating the old setup page like one long, overwhelming document. Because of that, they completely missed the logical steps required to make the AI work.
Action: Ditched the long form and split it into a clean, multi-step tabbed layout. By separating 'Triggers' from 'Messages,' we forced a clear, natural sequence.
MAKING THE LIVE CALL DATA SCANNABLE
Insight: The legacy call monitor table was dense and monochromatic. Supervisors found it incredibly hard to scan the screen quickly and spot when a call was going awry.
Action: Introduced high-contrast visual icons to represent bot sentiment, agent feedback, and live trigger statuses. Now, a supervisor can glance at the screen and instantly spot an anomaly in less than a second.
FIXING THE STATUS CONFUSION
Insight: It was very easy to confuse a rough draft with a live coaching scenario because the labels looked exactly the same.
Action: Added a bold, color-coded status bar to the top of the workspace. This completely killed the guesswork; supervisors now know exactly whether a bot is just being tested or actively running live on the floor.
BEFRORE - REAL-TIME COACHES SCREEN
AFTER - REAL-TIME COACHES SCEEN
The Results
The redesign completely changed how fast we could deploy live coaching. By making the workflow intuitive, we cleared the technical bottleneck and gave supervisors the ability to launch AI support instantly. Plus, by building a strict component library, we laid down a solid foundation that will let the team scale new features without breaking the app's visual rules down the road.
Once the new system went live on the floor, the impact on metrics was massive:
-10-20%
AVERAGE HANDLE TIME (AHT)
-25-40%
AFTER-CALL WORK (ACW)
+8-15%
CUSTOMER SATISFACTION (CSAT)
-20-30%
MANAGER ESCALATIONS
-30-50%
TIME-TO-PROFICIENCY
What I Learned From This Project
INFORMATION ARCHITECTURE IS THE BOTTLENECK
When you're dealing with complex systems, the biggest challenge isn't making things look prettyit’s figuring out how to translate messy, dense business logic into step-by-step workflows that actually make sense to a human being. If the underlying structure is confusing, even the best-looking UI won't save it.
SYSTEMIC DESIGN SETS THE SCALE
Designing a new feature set while simultaneously creating a component library from scratch is tough, but it's the only way to do it right. It forced me to focus heavily on the utility of every single component, making sure whatever we built wasn't just a quick fix, but a solid building block for future product updates.
DATA VISUALIZATION MAKES COMPLEXITY SCAN
In enterprise tools, dense data tables are usually unavoidable. But you don't have to settle for a wall of text. Introducing clever, high-contrast visual cues - like simple sentiment icons and color-coded status badges - can instantly transform an overwhelming screen into something a user can scan and act on in a split second.
PORTFOLIO
MORE CASE STUDIES
LIVE WEB APP
AI DESIGN AUDIT
AI SHIPPED PRODUCT
Social Feed
Social Feed
Social Feed
MOBILE PRODUCT DESIGN
INTERACTION STRATEGY
ENGAGEMENT SYSTEMS