Case study: UK Utilities - From 1% to 25% containment

Customers met rigid, menu-driven flows with no natural language understanding. Only 1% of contacts were resolved without an agent. Almost everything went to live queues, and frustrated customers dropped off and came back the next day.

Led the end-to-end redesign of the customer experience, from discovery to launch, and built the AI agent together with a LivePerson solution engineer. I also did the prompt engineering for Copilot Assist, the generative AI features that support human agents: Rewrite, Summarisation and in-conversation recommendations.

Journey mapping → intent design → RAG and Knowledge AI implementation → building the agent in LivePerson AI Studio → prompt engineering for the agent and Copilot Assist → testing and optimisation

I mapped the real reasons customers got in touch and replaced static decision trees with an intent-driven routing agent. It answers common questions from a governed knowledge base and sends everything else to the right team the first time. When a conversation reaches a human agent, Copilot Assist summarises it, recommends responses as it happens and rewrites replies in the brand’s tone, so human agents can resolve issues faster without the customer repeating themselves.

• True containment from 1% to over 25%, a 25x increase in self-serve capacity
• 8% fewer repeat contacts within a day
• 14% fewer ghost conversations and 4% fewer non-interactive ones, freeing agents for high-priority emergencies
• Lower cost per contact, reduced outsourced agent headcount and savings in the thousands of pounds