The Patient Who Never Called Back: Where Healthcare AI Falls Short
Picture a health system that has done nearly everything right with AI this year. Ambient documentation tools are running in its exam rooms, and physicians are finishing their notes before they leave the building.
Now picture a patient of that same health system calling on a Tuesday morning to book the follow-up appointment that the AI tool documented so well. She listens to two minutes of hold music, works through a phone menu with nine options, selects scheduling, and waits another two minutes before she hangs up and tells herself that she will call back later. She does not call back. As an industry, we renovated the kitchen and left the doorbell disconnected.
Ambient documentation works well and earned its adoption. My point is that health systems have invested far more in AI for the exam room than in AI for the phone line, and this article explains why that gap exists and how leaders can measure it.
Where health systems have invested in AI
Health systems have settled the question of whether to adopt AI. In February 2026, Eliciting Insights surveyed executives at 120 U.S. health systems found that 75% are using at least one AI application, up from 59% a year earlier. Clinical note-taking and ambient listening lead every category at 68% adoption, while administrative AI chatbots reached 25%.
I read that gap as a reflection of who sponsors each purchase. Ambient documentation has an obvious champion in the physician who gets time back at the end of the day, while patient access rarely has a sponsor with the same urgency. Deloitte’s 2026 Global Health Care Outlook, which surveyed health system executives in the U.S. and five other countries, found that about 30% operate generative AI at scale in select areas and only 2% have deployed AI across the entire enterprise.
Phone Access Connects to Every Other Patient Access Priority
When MGMA asked 236 medical practice leaders in December 2025 to name their top patient access focus for 2026, the answers split almost evenly among no-shows at 27%, online scheduling at 24%, phone access at 22%, and the time that new patients wait for an appointment at 21%. I take that spread to mean that patient access is one connected problem, and the phone sits near the center of it.
A patient who cannot reach the practice to reschedule often becomes a no-show, and a slot that goes unbooked because a call went unanswered is one more reason that new patients wait longer for care. Online scheduling helps, but MGMA reports that its adoption remains uneven, so the phone still carries much of the scheduling load. MGMA put the stakes plainly when it wrote that “your phone number is your brand.”
Why the front door was skipped
Health systems skipped the front door because building it is harder. An ambient documentation tool listens to a visit and produces a note, and it never needs to know whether Dr. Patel has a twenty-minute opening on Thursday. An AI voice agent that schedules appointments must know that, which means integrating with the scheduling rules inside EHR systems such as Epic, Oracle Health, athenahealth, eClinicalWorks, and Meditech. Reading from the EHR is the easier half of the work, and writing appointments back correctly is the real project. That difference explains why phone AI is evaluated as an IT integration project while ambient documentation is funded as a physician retention initiative.
How Parlance builds for the front door
Parlance has worked on this integration problem with health systems for more than 30 years. Callers use physician names and medication terms that general-purpose AI models were not trained to recognize, so name recognition must be built for healthcare before the system can route the call to the right person. When the voice assistant fails to understand a request twice, the call transfers to a live agent, so automation does not turn into a dropped call.
At Virtua Health, self-service grew from 51.5% to 68.2% of all calls and the net promoter score rose from 20 to 55 after the health system deployed Parlance, results that Virtua discussed in a conversation recorded at HCCT 2025.
Three phone metrics that make the case for phone AI
Most organizations can report documentation time savings to the minute but cannot report their call abandonment rate. MGMA recommends starting with three phone measures, which are average speed to answer, abandonment rate, and call transfer rate. Most patient access teams already have all three in a phone system dashboard, and I would encourage any leader to pull those numbers before the next AI steering committee meeting. They are also the first step toward finding the wasted steps in how the organization answers the phone.
A better definition of AI maturity
The industry measures AI maturity by counting applications, pilots, and use cases, and Deloitte’s 2% enterprise-wide figure suggests that counting is a weak indicator. A better test is one that the patient can run herself. Can she reach your organization on the first attempt at seven in the evening, without being routed into a queue that nobody is staffing? If she can, the AI strategy is mature. If she cannot, the health system has built a beautifully documented visit that a patient never made it to.