• Legacy keyword-based search was brittle, manually configured, and largely ineffective

• Customers described problems in emotional, unstructured language that search could not interpret

• High friction between “What’s wrong?” and “Who do I call?”

• Opportunity to introduce AI, but with high risk around misinformation, safety, and trust

• Plumbing emergencies introduce liability, urgency, and brand risk if guidance is incorrect

Rather than treating this as a simple search upgrade, the work reframed search as a moment of customer vulnerability—often occurring during stressful or emergency situations. Discovery focused on how customers actually describe plumbing problems (“my basement is full of water”), how urgency and emotion shape intent, and where legacy search failed to respond meaningfully.

The solution used AI as an augmentation layer, not a replacement. A unified retrieval-augmented generation (RAG) and vector search index, natural-language understanding, and context awareness (location, intent, emergencies) allowed the system to interpret messy, human queries while guardrails ensured safety, brand alignment, and liability control.

Special attention was given to AI trust mechanics: hallucination prevention, moderation rules, confidence thresholds, disclaimers, and user feedback loops. This ensured the experience felt helpful and authoritative without overreaching—escalating to booking or human support when appropriate.

• Discovery and reframing of search as a trust-critical experience

• Natural-language interaction models for problem diagnosis and guidance

• Context-aware logic (location, intent classification, emergency detection)

• AI behavior definition, tone guidelines, and safety guardrails

• High-fidelity prototype of AI-augmented search experience

• Delivery guidance for production rollout and monitoring