The AI operating system behind UAE clinics' patient operations

Context

Clinics lose revenue in the gaps of patient operations: enquiries from Google, Meta ads, Instagram DMs, and WhatsApp that nobody answers in time; no-shows that never get recovered; follow-ups and recalls that never happen. Zavis is an AI-native platform that closes those gaps: every channel lands in one inbox, an AI front desk answers and books 24/7, and everything writes back to the clinic's EMR.

I lead product across the platform: the AI agents, the omnichannel inbox, bookings with live EMR sync, and the automations that keep patients on schedule.

Why healthcare makes AI agents hard

A generic chatbot can afford to be approximately right. An agent booking a real patient into a real clinic cannot.

• Guardrails first: the agent must never invent a price, a slot, or anything resembling medical advice. We constrain what it may commit to and design explicit handoff paths to human staff for everything else.
• EMR as the source of truth: bookings, reschedules, and cancellations sync bidirectionally with the clinic's EMR. Zero double entry is a product promise, not a nice-to-have.
• Multilingual patients: UAE patients switch between Arabic and English mid-conversation. The agent keeps up in the patient's language, on chat and on calls.
• Trust is per-clinic: every clinic has its own tone, services, and policies. Agents are brand-tuned per clinic, and a failure at one clinic damages trust in the whole platform.
• Revenue is the metric that matters: the platform is judged on captured leads, reduced no-shows, and recovered revenue, not on how impressive the AI demo looks.

What I own and how we work

• Product strategy and roadmap for the AI front desk (chat + voice), omnichannel inbox, and patient-journey automations.
• Workflows across the patient lifecycle: lead capture from ads and DMs, qualification, booking, reminders, no-show recovery, reviews, and recalls.
• Working directly with engineering and ops to onboard clinics across specialties, each with different services, EMRs, and operating styles.
• Evaluation discipline: agent changes are tested against real conversation scenarios before rollout, and production conversations are reviewed to catch failure modes evals missed.

Outcomes

Zavis runs patient operations for leading UAE clinics and hospitals across dental, dermatology, aesthetics, home healthcare, wellness, and multispecialty care: capturing leads across channels, cutting no-shows, and recovering lost revenue. Clinic-specific numbers are confidential: I'm happy to walk through them, and the decisions behind them, in a conversation.

What this taught me

• Demos are easy; production is the product. An agent that wows in a demo and mishandles one real patient conversation is a failed product.
• In regulated, high-trust domains, knowing what the agent must never do matters more than what it can do.
• Vertical AI wins on workflow depth, not model quality: the EMR sync, the recall campaign, and the payment link are why clinics stay.