AI powered intent setup

TL;dr Redesigned AI-powered Intent onboarding, reducing enterprise setup from hours to minutes for 30k+ admin users.

Role Timeline Team Impact

Lead Product Designer 2 months PM + 5 Engineers Reduced onboarding friction

Responsibilities

• Defined product requirements

• Led AI ideation and strategy

• Concept user-testing w/ Onboarding Managers

• High-fidelity UI in Figma

The Problem

Building from Scratch

There were no clear requirements, so I made my own.

I fed the roadmap and user research into a Claude project to define the onboarding strategy from the ground up.

Key Feature Challenges

• Support 50+ Offerings with intent signals

• Reduce manual setup and cognitive overload

• Create scalable onboarding for enterprise admins

• Design for complex multi-state recommendation flows
  

The feature concept

Strategic Shift
The old Intent setup experience relied on manual topic selection.

I reframed Intent from a configuration tool into a guided AI-powered onboarding system.
This shifted admins from building setup manually to reviewing optimized recommendations.

Designing a full system

Once AI recommendations became central to onboarding, the problem shifted from screen design to systems design.

Intent became a living system with complex states, edge cases, and evolving logic.

19 scenarios
across 5 trigger categories

Onboarding, updates, edge cases, partial and total failures, and multi-admin workflows.

To reduce design debt and simplify implementation, I aligned with Engineering around a centralized scenario map.
(instead of dozens of disconnected Figma flows)

🖼️ **[IMAGE — zoomed out, with zoomed in content for scale.

AI accelerated early exploration

Google Stitch helped me visualize the optimization partner V1

STRATEGIC DESIGN DECISION 1:

Two Types of Topics

The PM initially wanted one table for both offering-based and general intent topics.

As I mapped the workflows, it became clear there were two different mental models.

Trying to force both into a single table created cognitive overload and made large-scale setup harder for admins.

I proposed two optimized workflows over one bloated table.

Although it increased development scope, the simplified workflows reduced scanning effort and made onboarding easier to manage at enterprise scale.

Admin with Offerings configured
Quickly review and approve Intent topics across multiple Offerings.

Admin without Offerings configured
Easily review topics relevant to the business

STRATEGIC DESIGN DECISION 2:

Designing for clarity and trust

Early explorations used layered tabs to organize recommendation groups, but usability reviews showed the experience became visually exhausting and difficult to scan.

I shifted the experience into a stacked scrolling layout instead.

This created a clearer narrative flow:

• admins could review recommendations in context

• recommendation logic became easier to understand

• the onboarding experience felt more aligned to their business

That moment of clarity mattered.

The easier recommendations were to understand, the more confidently admins adopted them — improving onboarding completion, signal quality, and long-term platform value.

Usability Testing

Balancing simplicity with flexibility

Graphic will go here. (placeholder below)

Strategic design decision 3

Designing for cascading system states

As recommendation logic evolved, approval and removal actions became increasingly interconnected across Offerings and shared Intent topics.

Small changes could create unintended cascading effects:

• removing shared topics

• breaking recommendation relationships

• creating conflicting states across workflows

The removal experience needed to be extremely clear and predictable.

Working closely with Engineering, I mapped cascading approval and removal states to close system gaps and reduce accidental disruption.

This led to more granular interaction patterns, including chip-level approval states for Offering-based topics.

Claude-assisted workflow testing helped rapidly uncover edge cases and state conflicts that would have been difficult to validate through static Figma flows alone.

In reflection

What this project became

What started as an onboarding redesign evolved into a scalable recommendation system designed around trust, flexibility, and long-term maintainability.

The final experience balanced:

• AI-assisted onboarding

• enterprise scalability

• multi-admin workflows

• progressive complexity

• implementation-aware UX decisions