Optimizely: Opal AI
Designing how AI agents embed into marketer workflows.
Role
Product Designer
Team
4-person masters capstone team (including me)
Timeline
February - August 2026
Advised by
Optimizely VP of Product
Optimizely Senior Product Manager
Overview
Optimizely gave us their product suite and one instruction: design agent-assisted workflows. Choosing where to aim it was the first design decision, and it led to personalization: the capability sold to nearly every customer, and shipped by almost none.
The problem
The VP who buys personalization never has to use it. The marketer who inherits it can't get started, and can't tell whether anything worked. This led to churn rates during first-time use at 70-80%.
What I did
Found the failure, then leveraged Opal, Optimizely's AI, as a quiet agentic assistant that gently guides the user through their first campaign to give them confidence to internalize their knowledge of the product.
Where it stands
Prototype currently in testing. Optimizely asked for a PRD to take to their engineering teams.
How it's going
February
Kickoff: an open brief
Optimizely gave us their product suite and one instruction: design agent-assisted workflows. Our first design decision was where we should aim this project.
We worked backwards, looking for the widest gap between what customers say they want and what they actually do. Personalization was the extreme case: the capability customers name first, and the one almost none of them ship.
March
Desk research, and conversations inside Optimizely
The gap wasn't only on the customer's side. Personalization turned out to be one of the company's most stubborn internal problems too. It sells easily, so nothing forces anyone to fix how it works.
It's the only product in the suite without a PM.
April – May
Interviews
5
SMEs inside Optimizely
4
marketers outside it
around 80
secondary sources, tagged for confidence
May
What we found
The buyer isn't the user
A VP buys the vision and her job ends at signature. A marketer inherits a goal she never chose, with no brief and no reason to care.
It's an activation problem first, and a churn problem because of it
Most accounts never ship a single campaign. The product goes unused, then quietly lapses. Nobody is churning off personalization, they never started.
The product is genuinely hard to start
Seven stages before a campaign is live, and a developer has to install tracking code before any of it runs.
June
Validated the problem with advisors
Our Optimizely advisors confirmed the diagnosis and the persona we're aimed at.
You're brand new to the business, the old person who ran it left, you're picking it up. That's so real for us. I can't think of a better persona.
June
The direction: Opal, quiet and proposing
Optimizely already has an AI agent, but it lives in a chat box waiting for a prompt, it hands the hardest parts of the job back to a marketer who doesn't know what to ask for.
Not this
"Ask Opal anything." A blank prompt is the same empty page she already can't fill.
This
Opal reads her traffic and proposes one campaign she can run today, with an honest read on whether it worked.
July
Prototype-building
Four decisions the build is holding to:
Meet her where she works
The nudge sits on the page she already opens, not behind the personalization tab she never does.
Propose, don't ask
Named campaigns complete on sight. Opal is the escape hatch when none fit.
Only offer what she can run
Anything needing a developer or a paid add-on doesn't appear at all.
Never claim a win
No predicted lift, no green success state. The hedge is never quieter than the number.
From whiteboard to working prototype:
Sketch
Mapping the workflows on a whiteboard: where Opal should step in, and where it shouldn't.
Flow
Low-fi flow: the nudge, the starter campaigns, one setup step, and the honest read.
Prototype
The built prototype, running on Optimizely's real UI.
Now
Testing with users
Early signal on the approach has been good across four audiences.
3
professors and guest critics
2
Optimizely advisors
1
user test so far
+
more sessions booked
Next
Refinement
Understand our test findings and fold them back into our flows.
Then
Presentations and hand-off
Final capstone presentation, and a PRD handed to Optimizely, their internal route for getting work onto the engineering roadmap.
Full research corpus, decision records, and build are in a private repo. Happy to walk through it.
VIEW MORE PROJECTS
2025
Affirm
A self-service deal platform that cut routine updates from days to under 20 minutes.
Interaction Design | AI Prototyping | B2B
2026 - Present
Optimizely: Opal AI
Designing how AI agents embed into marketer workflows.
AI Workflows | Masters Capstone
2022 - 2025
SuperWorks
Brand identities, design systems, and hand-coded sites across three years of agency client work.
Brand Identity | Web Design | Visual Design
2025
InvestiGators
A social investing card game, from concept to boxed product in one month.
Product Design | Game Design | Masters Project
2026 - Present
Amigurumi Pattern Studio
A browser-based 3D editor that turns sculpted shapes into crochet patterns, validated stitch by stitch.
Product Design | 3D Tool Design | Systems Design
2022 - Present
Pink Daisy Crochet
A quarantine crochet hobby grown into a small business of characters, patterns, and an Etsy shop.
Entrepreneurship | Branding
2022
Social Pattern Library
25 UX patterns against online harassment, published publicly and cited in Congress.
Design System | Qualitative Research | Internship Project