Personalization was the problem, not writing speed.

Recruiters told us the same three things: researching each candidate took too long, staying on brand was hard, and personalizing at scale was nearly impossible. Before designing anything, we ran foundational research to find where AI would actually earn its place.

Data first, then design.

The quality of the output depended on what we fed it. I partnered with Data Science, Product, and Marketing to decide which data should shape a message, starting from what recruiters already trusted: their own saved templates, the candidate's profile, job details, and company information. We validated output quality against sandbox data before committing to the final approach.

Because this was the first generative AI experience in Recruiter, we designed it as a pattern that could scale to later releases, not a one-off feature.

An operating rhythm built for a high-visibility launch.

Every design decision had a clear path to a decision-maker. I owned the quarterly design planning cycle with cross-functional partners and set the review cadence for the team.

Two hard calls.

A pattern for AI in Recruiter, not just a feature.

The launch proved recruiters would adopt AI when it personalized their own work rather than replacing it. The data foundation, the onboarding pattern, and the cross-functional review model laid the groundwork for how the experience expanded.

Organize the enterprise around roles, not applications.

Every acquisition had added its own tools. Field workers lost time hunting across systems, several apps had never been brought up to our UX standards, and onboarding meant requesting access one application at a time. The fix wasn't another app. It was a different model: start from what a person in a given role needs to do their job, and assemble the experience around that.

One place to work

A dashboard by role that pulls data from several systems into one view, so workers stop switching between apps.

Onboarding by role

Reusable access profiles, so a new or reassigned worker gets the right tools on day one.

A framework, not a project

Built so other roles could follow without starting over.

Do no harm

No one could lose access to an application or feature they relied on during the transition.

Access as a data foundation.

The role-based access model did more than simplify sign-on. Working with Security and the Chief Architect, we designed how authentication tokens and user profile data would be governed by role. That gave the data science team a trustworthy foundation for using machine learning to personalize the experience later.

COVID compressed the timeline. The plan held.

Federal rail rules required physical distancing, and crews needed information without crowding into shared workspaces. We shortened the design cycle and launched the first release on tablets for train and engine crews in 2020.

A process for every kind of request.

To keep the platform from fragmenting again, we set up a single intake with a defined path for each type of work.

One hard call.

A framework other roles asked for.

• Centralized application and document access for field crews
• Reusable access profiles that removed repetitive onboarding steps
• Fewer applications, by combining use cases by role
• Field crews working from tablets outside crowded workspaces during COVID
• More roles requesting the framework after launch, all built in-house by the UX and technical teams

The work also shaped what came next: a scalable design system for future needs, and a move toward smaller, composable micro-apps instead of large monolithic applications.

Every acquisition brought its own interface.

Applications inherited through mergers looked and behaved differently, from search to navigation to labels. Rail employees who moved between several apps a day carried the cost in confusion and slower work. Developers carried it too, copying outdated code or inventing patterns because there was no standard to build on.

Consistency

One look and behavior across applications.

Onboarding

Easier for employees moving between apps, and for new engineers joining teams.

Speed

Faster development from approved, reusable components.

Quality

Better code by building on tested patterns instead of copies.

Audit, standardize, document, govern.

Goals and audit

Aligned stakeholders, designers, developers, and product managers on objectives, then audited every component, type style, color, and icon in use.

Guidelines and library

Set principles for accessibility, scalability, and brand, and built a reusable component library with states, interactions, and usage rules.

Documentation

Centralized guidelines in one wiki connected to the CSX Starter Showcase, so designers and engineers worked from the same source.

Governance and adoption

A contribution and review process for changes, scheduled reviews, and training through UX University to drive adoption.

A hierarchy teams could build with.

Components as the basic building blocks, modules that combine them into working sections, and page templates for common layouts. The UX team also built custom modules for needs unique to rail, like a train route visualization.

Speed to market, measured in people.

A senior project manager who had always budgeted two weeks of startup time for new team members reported they could start immediately. Engineers saw the same effect in their own work.

“Using the Showcase drastically lowered the development time.”Software Engineer, CSX

The site made customers work to become customers.

Before recommending anything, we mapped business objectives against customer goals: stakeholder interviews to set the objectives, customer interviews to understand expectations, an expert review for usability problems, and a competitive analysis to see where CSX stood.

First contact

Prospects were sent to several endpoints and email addresses to request information, and the main form asked for 25 pieces of information in the very first interaction.

Account setup

New users at companies with existing accounts had no easy path to create their own ShipCSX login.

Dead ends

Links on mobile and desktop, including calculator pages, led to the wrong pages or to empty ones.

Help content

FAQs were scattered across pages, PDFs, and spreadsheets, each in a different format.

Fewer doors, clearer paths, one place for help.

Simplify first contact

One clear engagement route for new and existing customers, collecting only what's needed up front and gathering details after the inquiry.

Fix the journey

Find and correct every misdirected or broken link.

Registration

A redesigned ShipCSX sign-up so new users at existing customer companies can create accounts on their own.

Help Center

A reusable, searchable help hub organized by persona and goal, replacing scattered FAQs with a clear hierarchy of help paths.

A front door built on what customers came to do.

• A customer-first case for the redesign, grounded in research and analytics rather than opinion
• One connected experience from CSX.com into ShipCSX
• Better discoverability for prospects who didn't yet know what CSX could do for them
• A clearer story about the value of CSX's services before a customer ever makes contact

Members didn't know when to follow and when to connect.

Connection requests frustrated both senders and recipients, and many members didn't see the value of simply following someone. The strategy was to make Follow an easy, valuable first step: a way to learn from people before, or instead of, connecting with them.

Follow as the primary action

Let members make Follow the main button on their profile.

Next best actions

Nudge members toward useful follows at the right moments.

Jobs to be done

Focus on what members came to do and the quality of what they see.

Community

Showcase members, products, and business solutions more prominently to build community through Follow.

Coordinating 12 to 15 teams without slowing them down.

Follow became a real way to grow a network.

• Following creators before sending a connection request led to about a 10% net increase in total follows and more engaged sessions
• Larger follow recommendation cards increased the visibility of followable members by about 8% on feed and profile
• Bundled together, the initiatives drove about an 11% lift in member-to-member follows and about 5% more engaged feed sessions

Programs that outlast any one project.

UX University and Dev University

Multiple training sessions that taught hundreds of technical and non-technical employees at CSX how to work with UX, the design system, and modern development practices.

Company-wide hackathon

Co-led CSX's first company-wide technology hackathon, sponsored by AWS and Microsoft, and partnered with business leaders so winning ideas could become funded initiatives.

Code Bytes and UX Bytes

Short, recurring sessions where teams shared techniques, plus community meetups, mini hackathons, and team outings.

IBM Buddy Program

Paired every designer with two teammates for three months to build rapport, share client experiences, and trade practices. Other teams asked to join.

Remote team building

At LinkedIn, team-building workshops for a fully remote team and in-person campus experiences when we could gather.

The team writes it, and the team keeps it.

At IBM, my team wrote a working agreement the same way we'd run a session with a client. Three commitments held it together.

Understand motivations

Learn what drives each person and where they want to grow, then match work to strengths.

Empower and mentor

Give people the support and autonomy to do their best work, and connect them with mentors across the organization.

People first, anywhere

Remote or onsite, keep communication open and make it safe to give and receive feedback.

“People will forget what you said, people will forget what you did, but people will never forget how you made them feel.”Maya Angelou