Role
Product Designer 2 (owned: selection process, citation system, AI chat)
Timeline
3 months, 0→1
Generate Brand Research
From 2 weeks to 30 minutes: a 98% cut in strategist research time.
Problem
Pitching a new campaign is high-stakes for creative agencies, and the prep behind it is tedious. We interviewed 10 creative strategists, and every one of them pointed to research as the most time-consuming part of the job: 2-3 weeks for a junior strategist to sift through brand history, past campaigns, social media trends, industry trends, and competitor analysis, before the team even starts distilling it into a "golden nugget" idea. The catch: strategists didn't fully trust existing AI tools. Misinformation and hallucination were dealbreakers in a field where credibility is key to winning the client. We had 3 months to ship a new product that solved both problems at once: speed and trust.
Process
With a tight deadline, we ran user interviews and design sketches in parallel rather than sequentially. Tools like Notebook LM helped us synthesize interview transcripts into themes fast, and we benchmarked competitors (Mintel, Perplexity, Waldo) for patterns worth borrowing or avoiding.
We brainstormed layouts with AI and shipped weekly stakeholder reviews to keep the compressed timeline honest. Engineering prototyped in parallel, using AI to move from design to working code faster than a typical handoff.
I owned the selection process (how a strategist's research methods get set up and run), the citation system, and the AI chat interface: the three pieces most tied to solving the trust problem. We landed on a 2-step flow: enter basic client info, then let AI run the research. It was deliberately simple so non-technical strategists could self-serve with no onboarding.
Solution
The insight that shaped the product: strategists are usually asked to pitch research built on their own agency's proprietary methods, but no internal tool existed to save and reuse those methods at scale.
Generate Brand Research lets a strategist enter basic client info (name, industry, competitors) and run their agency's own saved research methods via AI, executing 20-30 research threads in parallel. Every result ships with citations back to the original source, so strategists can verify instead of taking the AI's word for it. That directly answers the trust concern we heard in research.
Citation
Since citation was ranked the most important for the user, we made it easy to find in the paragraphs for fact checking. User also mentioned they like to read related publications in full for holistic understanding of the topic, so we added a full list view for easy reference.
Ai Chat
Since the AI Chat design pattern across the Omni platform had many variations, we created a design system for AI components in this chat. From heading styling to animations, we made sure the experience felt seamless for users.
We saw 1k users onboard and average usage of 30min per session in the first 3 months post-launch. What used to take 40+ hours now takes under 30 minutes, a ~90% reduction in research time.
Result
Adoption was fast and strong in the first 3 months post-launch. What used to take 40+ hours now takes under 30 minutes, a ~90% reduction in research time. Demand is now pulling us into EMEA and Asian markets.
The most requested fix: filtering by timeframe, so strategists aren't served outdated research. That's in development now.
"This feels like cheating for us strategists."
Karen O., Head of Strategy, Hearts & Science
Reflection
Looking back, running research and design in parallel under a 3-month deadline meant making calls with partial data more than once. We assumed users wanted an easy point of entry, based on complaints that existing tools like Waldo were hard to use. With more time, I'd want to validate that with user testing, to see if there are harder-to-learn features worth the tradeoff that we missed. What I'd do differently: I pushed to build the timeframe filter into v1, but got stakeholder pushback given the deadline. Looking back, I'd design it in parallel and use that time to gather more user feedback and build engineering buy-in, so it could ship as a fast follow right after v1 launched.