Resonate
Cortex
Led UX from concept through general availability for an AI-driven audience intelligence platform that did not exist when I arrived.
- Role
- UX Design Manager
- Timeline
- 2024 to 2026
- Team
- 1 direct report
- Type
- SaaS, 0 to 1, audience intelligence
This case study is protected.
Enter your work email or the password you were given.
Your address never leaves this page. The check runs in your browser, nothing is saved or sent, and I won't see what you enter. The gate is only here so the work isn't sitting in public.
The situation
When I joined Resonate, Cortex didn't exist.
I was hired to help figure out the future of their SaaS business. Their existing product, Ignite, was built around a taxonomy of 14,000+ audience attributes. Powerful on paper. But to get anything useful out of it, you had to already know what you were looking for. That assumption was quietly costing them clients.
My job wasn't to redesign Ignite. It was to figure out what should come next.
I spent the first phase working directly with product leadership on discovery. We built over a dozen prototypes to pressure-test different product directions before landing on the concept that became Cortex. The shift we were designing toward: instead of "here are 14,000 attributes, go find your audience," Cortex would say "here are the audiences that already fit your goal, here's why, here's how to reach them."
My role
I led UX for Cortex from that first concept through general availability. I also hired and managed the only other designer on the team, as part of building a UX practice that didn't really exist before I arrived.
The way we split the work was intentional and changed over time.
Early on, I owned discovery: defining concepts, aligning with product stakeholders on requirements, and building initial prototypes directly in the engineering codebase using Claude Code. She took those prototypes through Figma finalization and engineering handoff. That division wasn't just efficient. It was designed to position her as the engineering team's point of contact and build her credibility with them directly. I wanted engineers to see her as the expert, not as the person passing along my decisions.
As we got ahead of the roadmap, the split changed. She moved to feature-level ownership end to end. I shifted into a review and strategy role, stepping in when the backlog grew or a feature needed backup.
Beyond the product work, I took on scope that wasn't asked of me. I identified early that Resonate lacked a coherent multi-product strategy and brand architecture. Without one, anything we built would be harder to sell and position long-term. I put together a proposal, got executive buy-in, and led a company-wide shift from a "house of brands" model to a "branded house" model with Cortex at the center. The first product built under that new strategy was Custom Engaged Audiences (CEA), Cortex's flagship feature. It was designed from the start to work both as a feature inside Cortex and as a standalone commercial product for specific sales scenarios.
I built my career doing this kind of work. When I see the gap, I build the plan and drive it.
The design challenge
The core problem wasn't visual. It was a trust problem.
We were asking users to let an AI build an audience for them, in a product category where the mental model was "I control every attribute." The shift we were designing for was from builder to reviewer. That required every decision to earn trust quickly.
A few things that mattered most:
- Words. Terminology had to be exact. CEA audiences don't behave like attribute-based audiences. They're not composable the same way. Calling things by the wrong name creates wrong expectations. We landed on "topics" (not "audiences") when building a CEA. Small call, real consequence.
- The processing screen. This wasn't a loading state. It was the product demonstrating its reasoning in real time, step by step, so users felt like participants in the process rather than passengers. Getting this right took multiple research iterations.
- Modular architecture. The multi-product strategy meant CEA had to work in two different product contexts from day one. That added complexity upfront. It was the right call.
How we worked
I built a two-layer process to keep thinking separate from execution.
Strategy and reasoning happened in a dedicated Claude project workspace. I'd bring in raw inputs (product tickets, stakeholder Looms, meeting transcripts, research sessions) and work through the design problem before writing a single line of code. That workspace is where user stories got written, assumptions got pressure-tested, and the prompts that drove prototype development got drafted.
Execution happened in Claude Code, inside the live engineering codebase. Prototypes weren't disconnected mockups. They were built in the real repo, on real branches, using the same component library and design system as production. Any stakeholder could pull the branch and run it locally. It looked and behaved exactly like the real platform.
Research ran in parallel throughout. I ran 15+ sessions with participants across internal staff and client-facing colleagues, walking people through the working prototype and asking open questions. One practice I held to throughout: I kept what research participants said strictly separate from what the product team said in the same recordings. Product-side commentary doesn't count as external validation.
What was hard
Cortex launched to Early Access with five client accounts. The first few weeks were encouraging. Clients described it as "incredibly smart," and one account activated a Cortex-built audience in a live campaign within the first week.
Then the honest feedback arrived. One of the more active accounts hit friction during audience refinement and went back to Ignite to deliver. Geographic constraints that held for the first few audiences silently dropped for the rest. Edit actions that should have updated a batch only applied to part of it.
The pattern was clear. Ideation worked. Refinement didn't. And users needed audience insights (demographics, psychographics, passion points) surfaced inside Cortex before they could finish their work without leaving the platform.
"Great for ideation, breaks down at refinement" is not a finished product. Defending the work wasn't the right move. We took the feedback at face value and built the next roadmap from it: audience insights first, editing reliability second, microsegmentation third.
The harder part was doing that while the designer I'd hired was still growing into the role. The weeks we were iterating fastest were the weeks she needed me most present. Keeping her development on track while moving quickly was as much a leadership problem as a design one.
What happened
General availability launched in late April 2026.
The week before GA, seven accounts were active in Cortex. The first week of GA, 104 accounts were active. The product went from a closed beta to the full client base in a single week and held.
What happened after that is the part I find more meaningful. Average daily active time per user roughly doubled from the Early Access baseline (around five minutes) to where it sits now (around ten minutes), and the trend has been climbing steadily rather than spiking and dropping. Events per week grew 8x while active accounts grew roughly 10x. That means existing users were doing more per session, not just more accounts logging in once.
- 7 to 104
- weekly active accounts, first GA week
- 261
- unique accounts, 90 days post GA
- 100,061
- total events, 90 days post GA
- 80/100
- UMUX usability score
UMUX sample size was small, so I read it as directionally meaningful rather than conclusive. The industry benchmark for "good" is 70.
What clients said
I've been very impressed with the logic it's used in surfacing these audiences. I wouldn't have never thought of that.
Bonneville
It's honestly incredibly smart. Like it's way smarter than I was expecting.
ICON Media Direct
I love having access to it, it's super easy, it's fun to play with. It's an amazing story for us to talk to our partners about using AI-enabled planning.
Swellshark
Recognition
Resonate gives out a small number of 49er Awards each year across a company of 100+ people. I received one for the Cortex work, specifically for uncovering strategic value. It wasn't a participation award. It was the company saying UX could be trusted in ambiguous situations to find the path forward and bring people along with it.
What I'd do differently
Ship the audience insights layer earlier.
Every piece of Early Access feedback pointed to the same gap. Users needed to stay in Cortex end to end to complete their work, and the missing insights were the reason they couldn't. We knew that layer was important. I underestimated how load-bearing it was for the full workflow.
I'd also push harder, earlier, for written requirements from product leadership. The absence of documentation created ambiguity that slowed us down at multiple points and put more pressure on the design layer to fill gaps that product should have owned. I know how to have that conversation now, and I know it needs to happen at the start of the engagement, not after the first sprint.