Redesigning Joyz AI’s agent creation and analytics experience across voice and text to make agents easier to build, test, and improve.
ROLE
STATUS
IMAPCT
Setup time reduced to
Setup questions dropped by
Agents tested pre-launch
TEAM
1 Product Designer (me)
1 Product Manager
1 Tech lead and 3 developers
Joyz AI uses AI Agents to automate customer interactions across voice and text—from answering questions and qualifying leads to booking appointments.
Creating an agent required configuring tasks, knowledge, workflows, behaviour, and AI models. While powerful, users had to understand how these pieces worked together before launching an agent.
Once live, businesses had limited visibility into agent performance. They could run agents, but couldn't easily understand conversations, outcomes, or areas for improvement.
Creating an agent meant setting up tasks, workflows, knowledge, behaviour, and AI models separately.
Users had to understand how the system worked before defining what the agent should actually do.
Agents could only be properly tested after configuration, making iteration slower.
Businesses had no single place to understand conversations, outcomes, or agent performance.
Joyz AI used hands-on onboarding, so the support team regularly helped businesses configure their agents. Across these conversations, the same questions kept appearing around setup, behaviour, knowledge, workflows, and testing.
Three goals shaped the redesign. Together, they aimed to make AI Agents easier to create, test, and understand after launch.
Help users move from a business goal to a working agent without needing to understand how every underlying capability works.
Let users test and refine their agents while configuring them, instead of treating testing as a separate step.
Create a single analytics experience to understand conversations, outcomes, and agent performance.
The redesign focused on reducing setup complexity while keeping the flexibility businesses needed to build powerful AI agents.
01
Start with the agent's job
We started with “What should this agent do?” so users could define the outcome first, making it easier to configure the right capabilities without learning the platform upfront.
02
Bring configuration into one workspace
We brought tasks, access, behaviour, AI models, and analytics into one agent experience, helping users manage the agent without jumping between disconnected configuration points.
03
Make testing part of creation
We placed Test Call alongside the configuration so users could test and refine the agent while building it, shortening the feedback loop before going live.
04
Design analytics around outcomes
We designed analytics around conversations, outcomes, and agent performance, giving businesses the visibility they needed to understand whether their agent was actually working.
The redesigned AI agent experience guides users through the complete journey of creating and launching an agent. From defining its purpose and configuring what it can access to shaping its behaviour, enabling calls, and testing the experience, each step builds toward a ready-to-use AI agent.
Start by defining what the agent needs to do before getting into the setup.
The decision: Start with the outcome, so users understand what they’re building before dealing with configuration.
Let teams choose what data the agent can read and update while doing its job.
The decision: Give the agent only the access it needs, while keeping teams in control of their data.
Set the tone, language, response length, and style without changing what the agent does.
The decision: Separate what the agent does from how it communicates, giving teams flexibility without adding complexity.
Set up the voice, caller name, number, and call settings alongside the agent.
The decision: Keep calling within the same agent, so teams can extend its capabilities without creating another setup.
Try the agent in Chat or Call to make sure it behaves as expected before publishing.
The decision: Let users test the agent before launch, so they can catch issues before customers experience them.
Once published, the agent moves from configuration to real conversations. Teams can review interactions, track outcomes, and use performance data to improve how it works.
Every interaction gives sales teams more than contact information. The lead view brings together the AI summary, qualification details, call information, and transcript so teams can understand the conversation before deciding what to do next.
What moved after people started using it.
81%
Faster to launch
Reduced agent setup and publishing time from 26 min to just 5 min.
45%
Fewer setup questions
Drop in "how do I set this up" support conversations after launch.
90%
Tested before going live
In-app testing helped teams validate agent performance before publishing.
This project pushed me to rethink what simplicity means in an AI-powered product.
01
Complexity should be structured, not simply removed.
02
Human control becomes more important as automation increases.
03
People trust what they can understand and verify.
04
Every interface decision affects the larger system.
