Key takeaways
- AI now plays some role in patient visits at most medical groups; the front office is where many practices start.
- AI fits repetitive, rules-based work: routine questions, call capture, summaries, reminders and document sorting.
- Clinical triage, difficult conversations, disputes and anything needing judgment should stay with people.
- Most groups have no AI policy yet; write a one-page policy before the first pilot, not after.
AI has already arrived in most practices
AI is no longer a future project for most medical groups. In an August 2026 MGMA poll, AI played some role in patient visits at 83% of medical groups, up from 71% a year earlier[1]. Much of that is clinical, such as ambient note-taking, but the front office is where many practices look next, because that is where the repetitive work and the phone queues are.
Source: Medical Group Management Association (MGMA), 2026[1]
Show the data
| Item | Value |
|---|---|
| 2025 | 71% |
| 2026 | 83% |
Practices that are circling back to AI and automation to improve access have clear priorities. Scheduling came first at 31%, then calls at 27%, registration and eligibility at 23%, and prior authorization at 16%[2]. Those are the right places to look, as long as each task is matched to what the technology does well today.
A task map for the front office
| Task | Fit for AI today | Why |
|---|---|---|
| Answering routine questions: hours, directions, parking, portal help | Strong | Repetitive, answers come from an approved list |
| Capturing after-hours and overflow calls | Strong | Calls are answered instead of going to voicemail; staff get a summary |
| Summarizing calls and messages for staff | Strong | Saves reading or listening time; staff verify before acting |
| Sorting faxes and documents into queues | Good, with review | Pattern recognition; a person confirms the patient match |
| Eligibility checks and benefit summaries | Good, with review | Rules-based, but exceptions need a person |
| Booking and rescheduling in the PM or EHR | Depends on integration | Only safe with tested write access and your scheduling rules |
| Symptom triage | Keep with clinicians | Clinical judgment and safety risk |
| Billing disputes and complaints | Keep with people | Needs empathy, authority and context |
The pattern is simple. AI is strong where the task is frequent, the right answer comes from a list your practice approved, and a mistake is easy to catch and correct. It is weak, or inappropriate, where the right answer depends on clinical judgment, on reading a person's emotional state, or on authority to make an exception.
Where AI helps most today
The most immediate gains are on the phone. A large share of calls ask the same few questions, and a well-configured assistant can answer them from approved facts at any hour, then pass anything patient-specific to staff. Ava, for example, answers office and new-patient questions from the information a practice signs off and hands test results, refills, bills and appointments to the staff line the practice chooses.
- Routine questions at any hour, so staff spend phone time on patients who need a person.
- Overflow during the morning peak and lunch, so callers are answered instead of waiting.
- After-hours capture with a clear summary waiting for staff the next morning.
- Call and message summaries that cut the time spent listening to voicemail.
- Document sorting that routes referrals and records requests to the right queue.
What should stay with people
- Anything that sounds like an emergency: the caller hears an approved instruction to hang up and call 911, with no follow-up questions.
- Symptom assessment and clinical advice, which belong with nurses and providers.
- Bad news, sensitive results and conversations about cost hardship.
- Complaints, disputes and requests for exceptions to policy.
- Consent decisions involving minors, guardians and proxies.
- Any situation where the caller asks for a person.
Write this list down and test it. Every AI tool you consider should show you exactly what happens on each of these calls.
What patients think
Patients are more cautious than clinicians. Six in 10 US adults said they would feel uncomfortable if their own provider relied on AI to diagnose disease and recommend treatments[3]. Privacy is the larger concern: 77% of the public say they are concerned about the privacy of personal medical information provided to AI tools[4]. Those findings are about clinical decisions and health data, not about being told the clinic's parking options, but they shape how patients react when they realize they are talking to software.
- Tell callers they are speaking with an automated assistant.
- Offer a clear route to a person at any point.
- Keep the assistant away from health details unless your agreements and safeguards cover it.
- Use the practice's own approved wording, not generic scripts.
Write the policy first
Most practices are adopting AI without rules. In January 2026, only 20% of medical groups had AI governance or a formal AI-use policy, 22% were developing one and 56% had none[5]. A small practice does not need a committee. It needs one page that answers a few questions before any tool goes live.
- Which tasks may AI tools perform, and which are off limits?
- Which tools are approved, and who approves new ones?
- What patient information may each tool receive, and is a business associate agreement in place where needed?
- How do patients reach a person?
- Who reviews AI outputs, how often, and how are errors reported and fixed?
- May staff use public AI chat tools for work, and if so, with what restrictions?
Redesign the work, not just the tools
AI only saves time if someone decides what the time is for. In June 2026, 68% of practice leaders said their organization had not redesigned a role or adjusted staffing with the help of AI in the past year, while 26% had[6]. If an assistant takes routine calls, decide in advance whether the front desk will use that time for waitlist calls, earlier eligibility checks, or more attention to patients at the counter. Write it into the schedule.
Small practices are behind, and that is fine
Across the economy, AI use is concentrated in larger firms. Between December 2025 and May 2026, 37% of firms with at least 250 employees reported using AI, while fewer than 20% of firms with four or fewer employees did[7]. Small practices do not need to catch up all at once. They need one well-chosen task, a policy and a way to measure the result.
Questions to ask any front office AI vendor
- Which tasks is the tool live for today, and which are still planned?
- Where do its answers come from, and how do we approve or change them?
- What does it do when it does not know an answer?
- What exactly happens on a call that sounds like an emergency?
- What patient information does it collect, store or send, and for how long?
- Will you sign a business associate agreement if the tool handles protected health information?
- How do staff see what it did on every call or message?
The first question matters most. Demonstrations often mix live features with roadmap ideas. Ask the vendor to mark each feature as live or planned in writing, and test only the live ones.
Measuring whether it helped
Decide the measures before the pilot starts, and pull a baseline for at least four weeks. For phone tasks, track calls answered, voicemails left, callbacks made the next morning and staff minutes spent on routine questions. For document tasks, track time from receipt to filing and the share of items a person had to correct. Add a short staff survey at two and six weeks, and read a sample of calls or summaries every week. A tool that improves the numbers but frustrates patients or staff is not finished; adjust the approved answers and handoff rules before you expand it.
Start with one task
Choose the task
Pick something frequent, low risk and measurable, such as after-hours calls or lunch-hour overflow.
Write the approved answers
List the facts the tool may use and the wording for emergencies and handoffs. A clinician signs off.
Test with fictional calls
Staff play patients, including tricky and urgent scenarios, before any real caller reaches the tool.
Go live for a set window
Start with a few hours a week and review every summary for the first two weeks.
Measure
Compare calls answered, voicemails left and staff time on routine calls with your baseline.
Decide
Expand, adjust or stop, based on the numbers and on what staff and patients said.
Example: a fictional three-provider family practice started with lunch-hour overflow only. In the first month, most lunch calls were routine questions answered without staff, and the rest arrived as summaries routed to the right staff line. The practice extended coverage to after hours and kept clinical calls with its on-call service.
Frequently asked questions
Which front office tasks should a practice automate with AI first?
Frequent, low-risk, measurable tasks: routine phone questions, overflow and after-hours call capture, and call summaries. Leave clinical triage and sensitive conversations with people.
Do patients accept talking to an AI assistant?
Many will for simple tasks if they are told it is automated, the answers are accurate and a person is easy to reach. Surveys show strong concern about AI and health data privacy, so keep health details out unless safeguards and agreements cover them.
Does a small practice need an AI policy?
Yes, even a one-page one. It should list approved tools and tasks, what patient information each may receive, how patients reach a person and who reviews outputs.
Will AI replace front desk staff?
In most practices it changes the work rather than replacing people. Plan what staff will do with the time saved, such as waitlist calls or earlier eligibility checks.
References
Medical Group Management Association (MGMA) (2026). As AI use expands, medical practice leaders shift attention to measurement (opens in a new tab) โ
Medical Group Management Association (MGMA) (2026). Automatic for the people: AI moves for medical practices to boost the front office and access (opens in a new tab) โ
Pew Research Center (2022). 60% of Americans Would Be Uncomfortable With Provider Relying on AI in Their Own Health Care (opens in a new tab) โ
KFF (2026). KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice (opens in a new tab) โ
Medical Group Management Association (MGMA) (2026). AI governance in medical group practices: Rules for the humans in the loop (opens in a new tab) โ
Medical Group Management Association (MGMA) (2026). AI is slowly redesigning work in medical practices rather than replacing workers (opens in a new tab) โ
U.S. Census Bureau (2026). Large Firms With at Least 20 Employees Biggest AI Users (opens in a new tab) โ