An AI contact center in healthcare uses voice agents and automation to answer patient calls, book and change appointments, handle refills and billing questions, and route calls to staff. It can cut hold times and cover nights and weekends. The risk is that callers who struggle with automated voices, often older, hearing-impaired or anxious patients, quietly give up instead of getting through.
Every health system has a number it watches: average speed to answer. When the phones are overwhelmed, it climbs, patients complain and board members hear about it from their neighbours. An AI voice agent promises to bring that number down almost overnight, because it can answer every call at once.
It usually does. That is the problem with measuring it.
A call answered in two seconds by a voice agent registers as a success on the dashboard, whatever happens next. If the caller is an 84-year-old with hearing loss who cannot make the agent understand that she needs to talk about a new pain, and she hangs up after three attempts, the system records a short, answered call. Nobody reports her. She may just stop calling.
That is the quiet risk with AI in healthcare contact centers. The technology improves the metrics that are easy to count, like hold times and cost per call, while potentially worsening a harder one: the number of people who needed help, tried to get it and gave up.
What does an AI contact center do in healthcare?
The phone is still the front door of most health systems. Patients call to book appointments, ask about test results, request prescription refills, query bills, get directions and, often, to describe symptoms and ask what to do.
AI is entering this work in layers. The oldest is the phone tree: press 1 for appointments. Next came speech-recognition menus that let callers say what they want. The newest are conversational voice agents built on large language models, which can hold a natural conversation, understand a request phrased any way, complete the task in connected systems and hand the call to a person when needed. Alongside these, AI tools assist human agents in real time by pulling up records, suggesting answers and summarising calls.
The business case is straightforward. Staffing contact centers is hard and turnover is high. Call volumes spike at predictable times, such as Monday mornings, and unpredictable ones, such as flu season. In England, the NHS made “the 8am rush” for GP appointments an explicit target in its 2023 plan to recover access to primary care, backing a move to modern cloud-based telephony. The pressure to answer more calls with fewer people is universal.
Where does voice AI genuinely improve patient experience?
Voice AI is good at several things that matter to patients.
It eliminates hold time for simple requests. A caller who wants to confirm an appointment time or cancel does not need to wait twenty minutes behind someone with a complex billing dispute. It works at 11 p.m. and on Sundays. It does not get tired, curt or flustered at the end of a long shift. It can speak multiple languages, which matters for patients who wait longer today because an interpreter has to be found.
It can also improve the calls that still go to humans. If automation handles routine requests, staff have more time for patients who need them. If an AI assistant summarises what the caller has already said, the patient does not have to repeat their story three times.
These are real gains, and for many patients, especially working-age adults comfortable with technology, an AI agent that answers immediately is simply better than the old phone tree and the hold music.
Why could AI raise call abandonment among older patients?
Because the populations that use health services most are also those most likely to struggle with automated voices.
Hearing loss is common and rises steeply with age. According to the US National Institute on Deafness and Other Communication Disorders, based on 2015 to 2020 national survey data, 22% of people aged 65 to 74 and 55% of those aged 75 and older have disabling hearing loss. Synthetic voices over a phone line, with compressed audio and no visual cues, are harder to follow than a human voice for many of these callers, and speech recognition often struggles with older voices, accents, speech impairments and background noise.
Cognitive impairment, anxiety and pain affect conversations too. A patient who is frightened or confused is less likely to phrase a request in a way the system recognises, more likely to be misunderstood and more likely to give up. And trust matters. Many older adults have been warned, rightly, that a realistic automated voice on the phone might be a scam, which is one reason the US Federal Communications Commission ruled in February 2024 that AI-generated voices count as “artificial” voices under the Telephone Consumer Protection Act, bringing them under the same consent rules as other robocalls.
Digital alternatives do not solve this for everyone. Pew Research Center surveys show that roughly three in four adults 65 and older own a smartphone, compared with about nine in ten adults overall, and ownership is lower still among the oldest and poorest. The patients least able to use a voice agent are often the same ones least able to switch to an app or portal.
None of this means older patients cannot use voice AI. Many will find it easier than a phone tree. It means the average experience can improve while the experience of a specific, high-need group gets worse, and a dashboard built on averages will not show it. We have argued before that access designed around the typical user tends to exclude the rest, in A Ramp Is Not Inclusion: What Disability-Inclusive Health Care Actually Takes.
A worked example: the metric that hides the problem
Consider a hypothetical health system contact center that handles 100,000 calls a month. Before automation, average time to answer is six minutes and 12% of callers hang up while on hold.
The system deploys a voice agent that answers every call immediately. Within three months, average speed to answer is under ten seconds, the voice agent fully resolves 45% of calls and the on-hold abandonment rate for calls transferred to staff falls to 5%. The monthly report looks excellent.
Then an analyst looks differently at the data. She links calls to patient records and measures, for each caller, whether the call ended with the need resolved: an appointment booked, a message sent to a clinician, a transfer completed. Among callers under 65, resolution improved. Among callers over 80, the share of calls that ended inside the voice agent without resolution or transfer doubled, and repeat calls within 48 hours rose. A small number of those callers later arrived at the emergency department with problems that had started as phone calls.
The system makes three changes: a guaranteed route to a person by pressing zero or saying “person” at any time, automatic transfer after one misunderstanding rather than three, and a direct line to staff for patients who opt out of automation, flagged on their records. Automation rates drop a few points. Unresolved calls among older callers fall back to baseline.
The figures are illustrative. The measurement lesson is not: hold time and abandonment in the queue are not the same as whether the patient got what they called for.
Is the abandonment concern exaggerated?
The strongest counterargument is that the old system was already failing these patients. Long hold times hit older and sicker patients hardest, because they call more often and have less flexibility to wait on hold during working hours. A frail patient kept on hold for 25 minutes is not well served by the human-only status quo. If AI shortens every queue, the humans who remain can spend more time with the callers who need them, which may help older patients more than anyone.
It is a good argument, and it is often true. The best deployments use AI exactly this way: automate the simple calls so that people have time for the hard ones.
But it depends on two conditions that are easy to lose. First, the route to a human has to stay fast and obvious, rather than becoming a reward for callers who persist through the automated system. Second, the savings have to be kept, at least partly, as staff time. If automation is used to cut contact center staffing to the point where human queues are as long as before, older patients get the worst of both: a voice agent they struggle with, and a long wait when they escape it.
What this means for health systems and patients
For health system leaders, change what you measure. Alongside speed to answer, track resolution per caller, repeat calls within 48 hours, calls ending inside automation without resolution and opt-out requests, broken down by age group, language and, where possible, disability. Those numbers will tell you whether the system is working for the patients who call most.
Design the escape hatch first. A person should always be reachable quickly by saying so or pressing a key, and repeated misunderstandings should trigger an automatic transfer. Let patients choose to bypass automation and remember that choice. Tell callers at the start that they are speaking with an automated agent.
Keep clinical conversations with clinicians. When a caller mentions symptoms, the voice agent should route to nursing staff or advise emergency services when appropriate, not attempt triage. Language support is a genuine benefit, but it needs testing with real callers, as we discussed in The Family Member Is Not an Interpreter.
For patients and caregivers, it is reasonable to ask for a person, and to ask your clinic to note that you prefer human callers. If an automated system misunderstands you about anything clinical, hang up and call back asking for staff, or use emergency services if the problem is urgent.
Frequently asked questions
What is an AI contact center in healthcare?
It is a patient call center that uses AI, usually conversational voice agents plus tools that assist human staff, to answer calls, book and change appointments, handle refills and billing questions and route calls. The AI can resolve simple requests on its own and transfer others to staff, often with a summary of the conversation so patients do not need to repeat themselves.
Do AI voice agents reduce hold times for patients?
Usually, yes. Voice agents can answer many calls at once, so the initial wait falls sharply, and handling simple requests automatically can shorten queues for staff. But hold time is not the only measure that matters. Health systems should also track whether callers’ needs were actually resolved, particularly for older patients and those with hearing or speech difficulties.
Are AI phone agents hard for older patients to use?
Some older patients find them easier than traditional phone trees, but many face barriers. Disabling hearing loss affects 22% of US adults aged 65 to 74 and 55% of those 75 and older, according to NIDCD, and speech recognition can struggle with older voices and impairments. A fast, obvious route to a person is essential.
Is it legal for a clinic to call patients with an AI voice?
It depends on the country and the type of call. In the US, the FCC ruled in February 2024 that AI-generated voices are “artificial” voices under the Telephone Consumer Protection Act, so the consent rules for automated calls apply. Healthcare providers should check with their legal and compliance teams before using AI voices for outbound calls.
How can I reach a human when an AI answers the phone?
Most systems transfer you if you say “agent”, “representative” or “person”, or press zero, though this varies. If that fails, ask the clinic in writing or in person to note that you need to speak with staff. For anything urgent about your health, contact emergency services or your clinician’s urgent line directly.
AI can answer every call in seconds. The measure of a good health system contact center is still the same as it always was: did the person who called get what they needed, and did anyone notice when they didn’t?
Related on the GHC blog: Ambient AI Scribes Will Reduce Burnout Only If Nobody Spends the Time They Save and Telemedicine After the Boom: It Became a Mental Health Service, and Policy Hasn’t Caught Up.
Have you, or someone you care for, struggled with an automated voice at a clinic? Tell us what would have helped.
