Rethinking hiring with AI

UBS set out to change how it attracts and hires talent. I joined as Lead UX Designer on the AI platform for hiring and worked on it from discovery through concept to delivery.

Project details are described in general terms, and wireframes are redrawn with fictional data, to respect confidentiality.

The problem

Two groups were at the centre: hiring managers, who fill roles on top of a full workload, and employees looking for their next internal move. For managers, hiring meant constant context switching, missed steps and a process that felt fragmented. For employees, opportunities were hard to see, and a move could feel like it depended on knowing the right person.

The obvious route was to automate the existing workflow step by step. We took the harder one and rethought the full workflow around AI. That changed the core question: how can people explore freely with AI and still commit to decisions they are accountable for?

Why familiar patterns weren’t enough

Automating each step would have kept the old process and its problems. A chat assistant, like the ones people use privately, looked like the perfect fit: open, flexible, easy to start. But for real decisions it breaks down. Chat turns into a messy pseudo-workflow, nothing is reliably remembered, and it becomes unclear what is a suggestion and what has actually happened. In an enterprise that has to be auditable, that is not good enough.

So we designed a hybrid: chat to explore, a structured workspace to decide and act.

How I approached it

I mapped the wider employee ecosystem to see where hiring sits, then built service blueprints for three roles: hiring manager, recruiter and candidate. Each blueprint showed what people do, what AI does, where people check and decide, and what is governed and recorded. We tested the concept with prototypes, then in sessions with a pilot group of users.

Key decisions

1. Two modes, one system. Chat explores; the workspace decides and executes.
2. From conversation to commitment. Any useful chat result can become a tracked action: a draft becomes a saved role, a discussion becomes a shortlist.
3. One agent, two interfaces. The same intelligence answers in chat, triggers actions and coordinates the systems behind them.
4. Always clear what is real. People can tell a suggestion from a committed state.
5. Human authority at key moments. AI accelerates, people commit: approving a role, moving a candidate forward, making the hire.
6. Work flows across time and surfaces. Start in the workspace, continue in chat, finish later.

Wireframes redrawn for this portfolio. All names and numbers are fictional.

Outcome

• Started as a pilot and is now live.
• Became the reference model for AI products within the bank.
• Covers the manager’s journey from drafting a role to the formal hiring request; pay transparency support is in development.
• The matching skill was reused for other work, such as screening applicants.

What we expected

The platform was part of a wider shift in how hiring works: hiring managers would handle more standard cases themselves. We expected hesitance, because their responsibilities were changing.

What happened after launch

• Fast adoption where the task was familiar: managers took up AI-supported role descriptions quickly.
• Slow adoption where the behaviour was new: few managers explored internally matched candidates or reached out to them directly, which had not been the norm before.
• First fix, in the setup: employees appeared in role matches only after an explicit opt-in, so managers saw a thin pool. We changed how employees are included in matches and refined the experience further.
• Second question, still open: we ran interview sessions with hiring managers to understand why they hesitate to contact candidates directly. This work is ongoing.

What surprised me

The design changed how the work is divided, not just how fast it goes.

The AI workflows and the skills behind them became a shared intelligence layer: recruiters’ expertise, built into the software. Hiring managers can now handle standard cases themselves, and recruiters are brought in for the special and sensitive cases we defined.

What I’d take forward

• Expect late rules. Legal decided late that a final recruiter check was needed. We added it as a conversational confirmation, where the manager confirms they checked the details with the recruiter. Human checkpoints belong in the system design from the start.
• Skills are building blocks. Designing capabilities as reusable skills paid off when matching moved into screening.
• Know the limits. The hybrid does not suit dashboard- or chart-heavy work, where a workspace panel is too small and the data view has to lead.
• Speed gets adopted; new behaviour has to be designed. Making a familiar task faster spreads on its own. Asking people to act differently needs its own design work: defaults, permission and confidence.