Before your hospital buys an AI agent, write down your scheduling rules

On 21 September 2026, US patient access company DexCare acquired Mila Health, whose AI agents call, text and chat with patients. The headline numbers are about no-shows and revenue. The lesson for Indian hospitals is quieter: an agent is only as good as the scheduling rules underneath it.

What happened

According to DexCare’s announcement on Business Wire, Mila’s agents schedule appointments, prepare patients for visits and handle follow-up, working inside existing payer, provider and health technology workflows through APIs. DexCare says Mila achieves a 54% patient response rate on outbound outreach, 80% higher than manual call centres, with a 50% reduction in no-shows and an 18% increase in revenue.

DexCare says its customers, including Kaiser Permanente, Piedmont, Texas Health Resources and Tampa General Hospital, book 40% more appointments with existing resources and cut time to care by five days. This is its second acquisition. Tampa General’s Dr Peter Chang is quoted saying DexCare codified 260 scheduling policies and uncovered more than 50 gaps.

My take

DexCare CEO Matt Blosl put it neatly: Mila gives the data a voice, and the data gives Mila direction. Most hospitals shopping for voice bots can get the voice from any number of vendors. What they lack is the direction.

Think about how an Indian hospital call centre actually books. The experienced agent knows one cardiologist does not see new patients on Saturdays, another’s OPD always runs late so fewer slots should be offered, and a third prefers follow-ups before noon. None of this is written down. It lives in the heads of senior call centre staff and doctors’ secretaries. An AI agent dropped into that setup will either book wrongly or escalate everything.

The Tampa General number is the one I would put in front of a board: 260 scheduling policies and more than 50 gaps. Writing the rules down is the project. The agent is the reward.

What most coverage missed

This deal joins scheduling data with outreach. That is the patient funnel in one system: who needs to come in, when they can be seen and how to reach them. At most Indian hospitals those three live in three places: the HIS, the doctor’s diary and a WhatsApp number. Whoever unifies them owns the front door, whether that is the hospital, an aggregator or a vendor.

Also note what the numbers measure. Response rate, no-shows and revenue are commercial metrics. US health systems are buying these tools as growth tools, not IT projects, and Indian leaders should budget and govern them the same way.

What I would do

• Document scheduling rules for the top 20 doctors by OPD volume, including the unwritten ones, and get each doctor to sign off.
• Measure today’s baseline: call abandonment, booking conversion, no-show rate and days to first appointment.
• Start AI outreach with low-risk journeys such as follow-up reminders and health check bookings, in Indian languages.
• Keep a clear human handover for anything clinical or emotional, and review transcripts every week.
• Keep the scheduling logic and patient data in your own systems, whichever vendor supplies the agent.

What to watch

Watch how record vendors’ own patient agents compete with independent access platforms like DexCare, and whether Indian HIS vendors open scheduling APIs good enough for agents to use.

Go deeper: When automation should not answer the phone, How to increase OPD patients in a hospital: start with the diagnosis and What healthcare marketing automation actually replaces.

Source: Business Wire. Figures as reported at the time of writing.

Questions people ask