The cost of the calls nobody makes
Every sales team carries a pile of calls that should happen and never reach the top of anyone's list: people called twice and never reached, people quoted three weeks ago who went quiet, tomorrow's appointments nobody confirmed.
They go unmade because each is low-probability, and the consultant is right: talking to one warm patient beats failing to reach ten cold ones. But in aggregate the pile is a real loss — unused stock sitting inside advertising money you already spent.
AI voice agents exist for that pile. Not to take the warm patient, but to work the cold stack and find the warm one.
What it can do
The agent holds a spoken conversation: it calls, introduces itself, asks a few questions, understands the answers and keeps the conversation going. When the call ends, what lands in the CRM is not an activity tick — it is the recording, the full transcript, a short summary, and whether the other person came across positive, negative or neutral.
That last part is the point. The agent's value is not that it made the call but that it left the outcome readable. Looking at the list in the morning, a consultant sees who was interested and builds the day around it.
Automations can place the calls too: "call a new lead within five minutes", "call everyone with an appointment tomorrow", "call anyone silent for seven days". Given how much first-response time decides in this sector, the first one earns its keep alone.
Calls you can hand over
- First-contact verification. Is the person who filled the form genuinely interested, which treatment, when are they available. Collecting that needs no human.
- Appointment and date confirmation. Calling tomorrow's patient costs far less than the slot lost when they do not show.
- Working the unreachable list. Finding the ten people in a hundred who actually want to talk.
- Scheduling callbacks. "When would suit you?" and a note on the calendar.
What they share: they carry information, need no judgement, and quietly cost money when they never happen.
Calls you should not hand over
The line is clear, and because this is healthcare it is firm:
- Clinical questions. A wrong answer to "can I have the procedure while on this medication?" costs more than every minute the agent saved. It should hand that question to a person, not answer it.
- Price negotiation. A discount is a commercial decision; automating the negotiation costs margin.
- Objection handling. Convincing a hesitant patient takes intuition and flexibility. That is human work.
- Bad news and complaints. These are relationship moments. A complaint met by an AI becomes a bigger complaint.
Honesty: the patient should know what they are talking to
Passing the agent off as human may lift engagement in the short run. Over time it is a reputational risk, and in medical tourism reputation is worth more than a conversion rate.
The right call is for the agent to identify itself. What we see in practice: most patients keep giving information knowing they are speaking to an assistant — as long as it knows what it is doing and hands over to a person when needed. What irritates people is not the AI, it is a conversation that hits a wall.
Announcing that the call is recorded remains a legal requirement, agent calls included.
How to start
Pick one narrow job and measure it. The most reliable starting point is the unreachable list: it is a pile nobody touches, so there is nothing to lose, and the result is measurable — how many were reached, how many came back positive, how many were handed to a consultant.
Two weeks later you have a number. If it is good, take the second job (appointment confirmation); if not, fix the conversation flow. That is how a voice AI investment should be judged — not by how natural it sounded in the demo.