The goal
Strategy
Impact
AI personalization is used across industries to create relevant recommendations and contextually appropriate experiences at scale. It's the layer that determines how the agent adapts to each user, team, and organization. I shape the UX vision and strategy for that layer. That means working with engineering and product to define how personalization should work, and prioritizing it on the roadmap. I ground the direction in research and analytics, so it drives the outcomes that matter, like deeper engagement and faster time-to-resolution. The goal is AI capabilities that feel seamless, natural, and cohesive across every surface of the product.
Exploring how context powers personalization, and defining the scope.
I started by exploring how context could sharpen personalization, and defining the scope. From there, I worked with PM to map out how context composes across an organization, how team-based operations interact with the model, and what "working" would even look like.
Mapping the layered context model to observability.
The taxonomy borrows from how industry AI products already handle context: identity, operational graph, indexed knowledge, declared rules, learned memory, and user-authored templates. My contribution was working through which layers matter most for observability, how they'd interact with things specific to this domain (the entity graph, the on-call schedule, alert policies, existing runbooks), and where the observability-specific decisions actually live. The deliverable was a matrix mapping these layers to where each one matters most for observability.
Personalizing at the team and org level, not just the user.
The AI experience personalizes at the user, team, and organization levels — an architecture that differs from most consumer AI. For a team-based operations tool, the team layer is load-bearing: it's where shared runbooks, ownership, and on-call defaults live. Getting that layer right is what makes personalization feel cohesive across a whole organization, not just tuned to one person.
Framing the success the team could measure.
Adapted the learning-curve idea — same user, same task, day 60 quality vs day 1 — as a way to make "personalization is working" concrete for PM and engineering. Not a novel evaluation concept in AI, but naming it clearly gave the team a shared target and moved the conversation from "add personalization" to "here's what we'd need to see to believe it's helping."
From research to a design strategy
Design framework
synthesis
More relevant, accurate AI
What I delivered:
- The UX patterns of AI personalization — Surveyed major AI products to distill personalization into the core patterns that guide our design direction: legibility, editability, scope control, and origin.
- Personalization category taxonomy — Mapped how AI products break personalization into distinct user-facing features, giving us a clear catalog of the options to draw from.
- Mental models of personalization — Identified the three mental models AI products build on, and what each one actually personalizes, so we could pick the model best suited to our product.
- Personalization design framework — Synthesized the patterns, taxonomy, and mental models into a design framework that maps personalization onto our product and points to what to build first.