Discover Audiences tool
Discover Audiences is an internal audience intelligence platform for creating, comparing, and analyzing audiences across multiple data sources. It enables strategists to define relevant target groups for client campaigns and share audience outputs with planning and activation tools - connecting audience discovery to the broader media campaign workflow.
💼 Business goals
Provide reliable, data-driven audience inputs for media campaigns.
Centralize audience knowledge and tools across the company.
Equip media planners with AI-powered solutions to enhance their work and significantly reduce the time spent discovering audiences.
👩🏭Role
I led product design across three development teams, shaping the product vision and its AI capabilities. I designed and prototyped every major iteration of the tool and established design standards for its highest-value functionalities.
🚀Impact
1.74 K active internal users
⌚Timeline
3 years (2024 -2026)
✨AI layer
The work on Discover Audiences began as generative AI was becoming widely available. During the product’s 3-year lifecycle, user expectations changed significantly. At first, users were curious about AI but concerned about trust and hallucinations. Over time, they began to expect AI to help them start tasks, guide them through the workflow, recommend next steps, and connect different capabilities.
Look for the ✨ icons to follow the evolution of the AI layer throughout the case study.
🩶 Credits
Zuzanna Pawlak, Maciej Marczak, Dominika Klimek, Agnieszka Berhent, Aleksandra Chlabicz
Discover & Define
To collect information about users and their needs I’ve run following researches for version 1st of the product
👑 User Interviews
Goals:
Map the existing workflow, from building an audience to activating it as part of a client campaign.
Identify user groups, pain points, and unmet needs.
Understand users’ expectations and concerns regarding AI-powered tools.
In numbers:
10 online sessions
45 minutes per session
Participants: audience tools superusers from across the company
📋 Questionnaire
Goal: Establish users preliminary needs from ✨ ai powered tools
In numbers: 373 respondents
Respondents: Internal system users
🚀Internal workshops
Goal: Possible tool funcionalities and behaviours, wireframing
In numbers: 15 participants
🧠 Key findings and design responses
The research revealed three major opportunities.
The workflow was fragmented
Evidence: 100% of respondents said that creating an audience required using multiple tools.
Design response: Centralize audience data, creation, comparison, and sharing in one workspace.
Preparing client-ready outputs was manual
Evidence: 80% of respondents captured screenshots from multiple platforms and combined them into a final PowerPoint presentation for clients.
Design response: Create one editable space where users could collect selected audience data and prepare a client-ready presentation.
3. ✨ Users wanted AI, but needed to trust it
Evidence: ✨ 70% of respondents were concerned about AI hallucinations.
Design response: Ground AI outputs primarily in trusted partner data, provide context for generated results, and keep users in control of the final decision.
Develop & Deliver
🕹️ Building the prototype
I translated the research findings into an end-to-end prototype that supported the following workflow:
Build an audience
Review and edit presentation-ready audience data in a single view
Send the audience to planning tools and activation platforms
The goal was to replace a fragmented, screenshot-heavy process with one centralized workspace. Audience comparison was out of scope for V1 and deferred for at least another year.
👑 Usability testing
I ran usability sessions with audience-tool power users from across the company.
Goals:
Validate the V1 concept and information architecture
Test the Build Audience flow
Evaluate prompt-based interactions
Identify issues with navigation and terminology
Assess whether AI-generated results felt clear and trustworthy
In numbers:
10 online sessions
30 minutes per session
🧠 Key findings and design responses
✨ Users were unsure what to enter in prompts
Evidence: 50% of participants struggled to understand what to enter in the prompt field.
Design response: Introduce prompt starters, examples, and informative empty states to help users get started.
Terminology differed across tools
Evidence: 60% of participants encountered different terms for similar concepts across existing tools. This made the product harder to understand and navigate.
Design response: Create a shared vocabulary for Discover Audiences and explore a language agent that translates user intent into consistent product terminology.
Users expressed the need for faster compare feature development
Evidence: 50% of participants said the workflow should include audience and segment comparison.
Design response: Explore an audience and segment comparison experience.
Improve
👑 Reccuring usability sessions
Improving Discover Audiences was not a linear process. Over 3 years, I continued to refine the product through recurring usability sessions and feedback from active internal users.
Goals:
Understand the current workflows and use cases for the tools in pitch preparation.
Identify the most critical user needs and pain points during the pitch process.
Uncover and diagnose recurring usability issues.
Gather feedback on potential new features or improvements.
Track user sentiment and satisfaction over time.
In numbers:
• 8 online sessions per quarter, repeated across three rounds
• 30 minutes per session
🧠 Key reccuring usability findings and design responses
✨ AI should guide users through the product
Users expected AI to do more than answer questions. They wanted it to help them understand what to do next.
Design response: Introduce a Next Steps AI option that recommends relevant actions within the workflow.
✨ The AI layer insights layer felt flat
Users found the AI experience useful but wanted it to surface less obvious and more actionable insights.
Team response: Explore connected agents that share context and communicate with one another rather than operating as isolated features.
✨ Users requested an audience represenating agent
A user-initiated idea led to the concept of a conversational “Persona” agent that users could interact with while exploring the audience.
Design response: Add a Persona agent to support audience exploration through natural-language conversation.
Next project
Flip the Brief tool