GenAI customer copilot
One AI copilot, designed once and configured for different clients.
I led experience design for a global consultancy. I set the experience strategy and design direction for a white-label AI copilot, turning complex agent capabilities into one reusable customer experience.
Consumer brands in travel, retail and banking used it to explore GenAI applications through configurable prototypes.
I led. Conversational UX and AI interaction patterns · the design system behind every client skin · rapid prototyping and storyboards across industries · design partnership with AI engineering.
Outcome. Reusable copilot patterns and client configurations for enterprise teams.
Every client wanted its own GenAI copilot. Building ten would not scale.
Each client asked for its own assistant. Building ten separate copilots would have been slow, expensive and inconsistent, so we designed one platform whose conversational engine, reasoning and UI could be configured and re-skinned per client.
A bespoke build per client
One platform, configured per client
I owned the experience layer between the AI and the client.
The cross-functional team brought together engagement leadership, AI engineers responsible for orchestration and integrations, and client teams providing domain requirements. My job was to turn model capability into a product people could trust, and a system other teams could ship from.
Set the ambition for GenAI customer experience and owned client relationships.
- Conversational UX and AI patterns
- Design system for every skin
- prototypes across industries
- Storyboards for C-suite demos
Agents, reasoning, retrieval and integrations, built against the patterns we defined together.
Brand, data and rules set per client, without needing a redesign.
Standardise the behaviours. Configure the expression.
The design direction separated what should remain consistent—conversation states, confirmation, fallback, handoff and core components—from what could change for each client: visual identity, tone, enabled capabilities and connected data. A bespoke request was assessed against this boundary before becoming a new pattern, rather than automatically creating a parallel product.
Models are non-deterministic. The interface has to earn trust.
LLMs don't always follow multi-step logic perfectly. These patterns became part of the design system, so every client copilot handled uncertainty the same way.
Reasoning made visible
Show which sources were used when it builds trust; hide the workings when they only add noise.
Confidence & fallback
When required data is missing or confidence is low, the copilot asks one clear question instead of guessing.
Human handoff
Sensitive or high-stakes moments route to a person, with the conversation history passed along.
Feedback & confirmation
Nothing is booked, paid or changed without an explicit yes. Actions are reversible where possible.
Guardrails & compliance
A guardrail step checks tone, privacy and policy on every answer before it reaches the customer.
Processing states
The assistant's avatar spins at different speeds to signal how much work is happening.
White-label by parameter, not by redesign.
Each client is a set of parameters on the same components: brand tokens, tone of voice, connected data sources, enabled agents and content rules. Switch clients to see one chat component take on each configuration.
My role, plus the team, equals the outcome.
The platform was a collaboration between design, AI engineering and client teams. My contribution was the experience layer and the system that let every deployment feel considered. The work also supported executive demonstrations and internal enablement.
- Conversational UX and AI patterns
- Design system for every client skin
- Prototypes and storyboards across industries
- Design and engineering alignment
- Engagement leadership and client strategy
- Agent orchestration and integrations
- Client configuration and deployment
- Testing and evaluation
Adopted across leadership enablement, client demos and active product builds, with C-suite visibility worldwide.