Case study: A Dubai clinic cut patient intake time by 85% in six weeks
From eleven minutes on the phone to ninety seconds of WhatsApp. Here is exactly what we built and what it cost.
The clinic is a multi-specialty group in Dubai Healthcare City with roughly 600 outpatient bookings a week across five locations. Before this project, new-patient intake happened over the phone and took an average of eleven minutes per caller: nine minutes of demographic and insurance data capture, two minutes to find a slot. Roughly one in four callers dropped before the booking was confirmed.
The ask was blunt: stop losing that twenty-five percent.
We built a WhatsApp-first intake flow powered by an AI agent that can code-switch between Arabic and English. The agent gathers patient data, verifies insurance eligibility against the two major UAE networks in near-real-time, offers available slots at the caller's preferred branch, and sends the confirmation. If insurance fails the automated check, the conversation hands off to a human within sixty seconds.
The integration work was the unglamorous part. Three systems had to talk to each other: the clinic's booking platform (Bookings by Dr.), their insurance verification portal, and the WhatsApp Business API. The insurance portal had no API — we built a small scraper behind a firewall that mimics a clerk clicking through the form. Not beautiful, works every time.
What actually changed the number, though, was not the automation. It was the channel shift. Patients who would have abandoned a phone call do not abandon a WhatsApp thread. They reply when they get a moment. The median time-to-completion is ninety seconds of actual interaction spread over about four minutes of elapsed time, because people respond between other things.
Six weeks from kickoff to go-live. Two of those weeks were spent arguing about which fields were actually mandatory — a healthy conversation that the clinic had never had before, because the paper form had never been audited. A third was spent tuning the Arabic voice for the Gulf dialect the clinic's patient base actually uses.
Results at the ninety-day mark: - Intake completion rate up from seventy-four to ninety-six percent. - Median intake time down from eleven minutes to ninety seconds of active interaction. - Receptionist phone load down by roughly forty percent, freed up for walk-ins and insurance escalations. - No-show rate down eight points — because patients now receive timely WhatsApp reminders rather than relying on a follow-up phone call that rarely landed.
Cost to run: about three hundred fifty US dollars a month in model and infrastructure costs, plus the one-time build. The clinic's previous effort — a third-party "patient engagement platform" — had been costing nearly four thousand a month and handling none of this.
Two lessons we took forward from this engagement. One: meet people on the channel they already use. Trying to push callers to a web form was the first version of this and it failed quietly. Two: the integration work is what you are actually buying. The AI is cheap. Making it talk to your real systems reliably is what takes the time.