Patient calls answered, screened, and booked across 120 practices
The Situation
Practices lose revenue at the front desk. Every missed call is a missed appointment, and reception peaks exactly when the practice is busiest, so the most valuable calls arrive when nobody is free to answer them.
At 120 practices this stops being a staffing gap and becomes a structural one. No amount of front desk hiring covers 8,000 inbound calls a week, and outbound follow-up never happens at all, because there is nobody left to make those calls.
The practices served a bilingual Dutch and English patient base, which doubles the staffing problem: coverage has to speak both.
Why it was hard.
Every caller is told they are speaking to an AI agent before anything else happens. The EU AI Act and GDPR require that disclosure, so passing as human was never on the table, and building for it would have been building something illegal.
That inverts the usual bar. The agent has to hold a patient who already knows it is a machine, screen what they need, book into a real calendar, and recognize when a call is urgent and hand it to a human immediately. It earns the call rather than borrowing it.
What Was Built
Inbound: the agent answers, screens the request, books into the practice calendar, and escalates urgent cases to staff. One system across 120 practices rather than a separate deployment per site.
Outbound: campaign calling at 25,000 calls a week across 100 practices, for recall, follow-up, and rebooking. Work no front desk had the capacity to do at all.
Voice quality was treated as a product requirement, not a nice to have. The agent switches between Dutch and English based on the caller.
The Results
What This Means For You
If missed calls cost you revenue and your team cannot catch them all, a voice agent is one of the fastest paybacks in AI. In Europe the real question is whether it can be built inside the AI Act and GDPR rather than around them. This one was, disclosed on every call, and it still handles four in five end to end. The audit tells you in two weeks whether your systems can support one.
Not sure where you stand? Take the free AI Readiness Calculator