Andrew Adams

Work / Case study

One platform, one system,
for 13,000 consultants.

Slalom's people were juggling five tools to answer one question: what's next for my career here? I led the design of me@slalom, which replaced all five with a single platform built on a design system I built and govern.

Role
Manager, UX Design: research, information architecture, design system, delivery
Team
5 designers & researchers, partnered with 8 engineers
Timeline
Feb 2021 – present · launched to 45 markets
Scope
13,000+ consultants · replaced Workday, Confluence, SharePoint, Slack and Outlook workflows
41% → 85%employee satisfaction within 12 months
−60%HR support ticket volume
22 → 3 minaverage time on the core career task
The me@slalom home page in 2024: utilization, contribution and experience gauges, top learning picks, career next steps, personal actions and quick links.
me@slalom in 2024. One home for utilization, learning, career growth and the everyday actions that used to live in five tools.

The problem

Slalom is a people-first consulting firm with 13,000+ employees across 45 markets. Internal tools for career growth matter to retention, but they hadn't kept up with the company's growth. Finding a new role, tracking development and getting HR answers meant moving between Workday, Confluence, SharePoint, Slack and Outlook.

The cost wasn't only time. Every hop between tools lost context, and every lost answer became an HR ticket.

What research showed

I ran a discovery program of 12 in-depth interviews across 10 functions, with people leaders and team members in the US, Canada and Japan, plus a 200-person survey on tool use and pain points. Two findings set the direction:

Employees had stopped expecting improvement. They'd built workarounds and accepted the friction as permanent.

HR operations was answering the same questions every week. Deflecting those with a well-designed AI layer would free the team for higher-value work.

The bet

Seven modules, one system. Each module replaced a legacy tool while feeding a unified data layer, so information flows between features instead of living in silos. An embedded AI assistant answers the questions HR used to field by hand.

The career-management workflow, rebuilt on that data layer, went from an average of 22 minutes to about 3.

The earlier career-framework dashboard: a profile header and three cards for Learn More, My Roles and My Profile.
Before: the career framework. Useful, but one narrow job among many tools.
The 2024 me@slalom home page bringing performance, learning, career and actions into one view.
After: one platform where each module replaced a tool.
The Learning module: personal greeting, search, top learning picks and a library of in-progress, required and completed courses.
Learning: required, assigned and chosen courses in one library.
The Staffing module: a three-step path (update profile, complete staffing input, find opportunities) with empty states that teach the next action.
Staffing: a three-step path, and empty states that teach the next action.

The design system

A single platform only stays coherent if everything is built from shared parts. I built the me@slalom design system and govern it: foundations, design tokens, a component library and pattern documentation.

Foundations color · type · space · grid Tokens named by role, themed by mode Components the shared library Patterns documented flows Product 7 modules, 1 data layer Governance runs across every layer: contribution, review, documentation, adoption.
  1. Foundationscolor · type · space · grid
  2. Tokensnamed by role, themed by mode
  3. Componentsthe shared library
  4. Patternsdocumented flows
  5. Product7 modules, 1 data layer

Governance runs across every layer: contribution, review, documentation, adoption.

The system's anatomy. Each layer only builds from the one before it.

Tokens named by role, not by color

Every color is named for the job it does: brand-primary; semantic families for error, warn, success and info, each split into text, foreground, border and background; neutrals for containers, content and inputs; and a graphic scale for charts that runs from graphic-great to graphic-poor.

Because components reference roles instead of hex values, the same token names resolve to different values in light, dark and brand modes. A theme becomes a data change, not a redesign.

Color token sheet in light mode: brand, semantic error, warn, success and info roles, neutral container, content and input roles, and a graphic chart scale.
Light mode token sheet.
The same color token names resolved for dark mode.
Same names, dark mode values.

AI in the workflow

I piloted AI-enabled design-system workflows with the team, using Figma AI, Claude and ChatGPT to generate component variants, draft pattern documentation and surface token inconsistencies, and coached the team as they adopted them.

The product has AI in it too. Ask me@slalom answers employee questions in place, and every answer carries a "Helpful / Didn't help" control, so the team can see where the assistant earns trust and where it doesn't.

The Ask me@slalom assistant answering a question with a link, then showing Helpful and Didn't Help feedback controls and a thank-you message.
Ask me@slalom: an answer, then a feedback loop on every response.

The lesson carried into my own projects, where AI agents implement the design system in code. When agents consume a design system, tokens and documentation stop being reference material and become instructions.

Results

Within 12 months of launch, employee satisfaction rose from 41% to 85% and HR ticket volume fell 60%. The core career task went from about 22 minutes to about 3.