Reduced customer support workload by 65% using an AI support assistant

Lumora Health's 18 clinics handled more than 4,200 support conversations a month across phone, email and WhatsApp. Around 70% were routine: opening hours, appointment changes, preparation instructions and billing questions.

Response times often stretched beyond six hours, staff turnover was rising and patient satisfaction had fallen to 78%.

Rather than deploying a generic chatbot, we mapped every enquiry type by volume, risk and data availability. Low-risk, high-volume topics would be fully automated; appointment actions would run through secure integrations; anything clinical would always reach a person.

• Automate informational and administrative enquiries end to end
• Ground every answer in approved, versioned content
• Detect clinical and sensitive topics and escalate immediately
• Measure accuracy with real, anonymised questions before launch

We built a retrieval-augmented assistant with hybrid search over Lumora's knowledge base, connected to the scheduling system for bookings and reschedules. A classifier flags clinical topics, and escalations include a concise summary so staff never ask patients to repeat themselves.

An evaluation suite of 1,200 real questions scored answer accuracy and safety on every release, and a quality dashboard lets the operations team review conversations weekly.

Within three months the assistant was resolving 65% of all enquiries without human involvement, around the clock and in seconds. Staff time moved to complex patient needs, and satisfaction reached its highest level on record.

The assistant handles the routine so our people can focus on care. It's the most impactful technology project we've delivered.