Phones Ring and Nobody Picks Up

It’s six-fifteen on a Tuesday evening, and the front desk phone is still ringing. Whoever’s up front is finishing a check-in, pulling a chart for the next patient, and can’t get to it in time, so the call rolls to voicemail, the way it always seems to. “We’re so busy at the desk that a lot of our calls go to voicemail, and those patients just call the next practice,” is how one practice owner put it to me, and it’s close to the single most common thing I hear walking into a medical office for the first time.

What owners actually ask me isn’t “should I use AI.” It’s smaller and more specific than that: can something answer the phone when we can’t, can someone else finish my notes so I’m not doing it at ten at night, can we stop losing money to insurance denials we never had time to appeal, can we get through prior authorization without a dedicated staffer just for the paperwork. Nobody’s asking for a platform. They’re asking for their Tuesday back.

I’m Isaac Krupp. I spend a full working day inside a business, in person around Los Angeles and Southern California, remote anywhere else, watching how the work actually happens before I say a word about AI. I don’t sell software, build anything custom, or take a referral fee from any vendor I mention. What you get at the end is a written plan: three to five specific moves, in plain language, with honest monthly costs next to each one.

What AI for physicians can actually do right now

Here is the AI for doctors landscape as it actually stands today, one pain point at a time, in the order it tends to matter most.

Phones ring and nobody picks up

For a lot of doctors this is the pain that hurts before you even see the register. The front desk is checking in a waiting room, pulling a chart, and the phone keeps ringing in the background until it doesn’t anymore, it just goes to voicemail. Practices this size miss a real share of their calls, and most patients who hit voicemail call the next name on Google instead of leaving a message, which the research on missed calls in medical practices makes plain (https://www.patientprism.com/blog/missed-calls-medical-practice-revenue/). The fix is a healthcare phone agent that picks up overflow and after-hours calls, answers hours and directions questions, and books or reschedules routine visits. Anything that sounds like a symptom or an emergency goes straight to a person, every time, no exceptions. See our AI answering service for small business for how that piece works on its own.

Charting at home after the kids sleep

This is the pain doctors describe without being asked, the light still on in the kitchen after the family’s gone to bed, one more note to finish before tomorrow starts. An ambient scribe listens to the visit, with the patient told and agreeing, and drafts the note, the after-visit summary and suggested codes while the doctor is still in the room. One health system reported after-hours documentation time falling by close to a third once it switched a scribe on (https://www.advisory.com/daily-briefing/2026/02/04/ambient-ai-oi-ec), though a larger, more careful multi-site study found the time saved was far more modest (https://www.statnews.com/2026/04/01/ai-ambient-scribes-modest-time-savings-clinical-documentation/). Either way, the doctor reads and signs every note; the tool never finalizes a chart on its own, and no visit is recorded without the patient’s consent.

No-shows and empty appointment slots

An empty Tuesday afternoon costs the same as a full one, the room, the clinician and two staff are all paid for either way. Phone-booked visits no-show far more often than ones booked online, and switching to online self-scheduling has been shown to cut unused slots substantially in peer-reviewed research (https://pmc.ncbi.nlm.nih.gov/articles/PMC12081397/). The fix is self-scheduling paired with automated confirm-or-reschedule texts and an auto-filled waitlist tied to the scheduling system, so a cancellation gets backfilled instead of sitting empty. Reminder texts carry only the date, time and location, never a diagnosis, and every patient can opt out.

Claim denials start at the front desk

A single wrong digit on an insurance ID at check-in doesn’t show up as a problem for weeks, then it comes back as a denial and someone has to fight it. A large share of providers now see a meaningful slice of their claims denied, and a real portion trace back to intake errors made before the visit even started, according to Experian Health’s 2025 survey (https://www.experian.com/blogs/healthcare/state-of-claims-2025/). Reworking a denial, the research shows, costs far more than getting it right the first time (https://www.healthcarebusinesstoday.com/why-denial-rework-costs-small-practices-more-than-they-think/). The fix is automatic eligibility verification the night before every visit, plus a claim scrubber that flags likely denials and drafts the appeal before submission. Coders and clinicians still choose every code; no tool changes one to whatever pays.

Prior authorization paperwork fights

Every insurance carrier runs its own portal, its own forms, and its own idea of what counts as medical necessity, and keeping up can feel like a second full-time job for whoever draws the short straw. Electronic prior authorization tools read the order and the chart, check whether a payer requires sign-off, fill the payer’s form with the relevant evidence, and draft the letter of medical necessity for the clinician to sign. On a normal Monday, that means a dashboard showing what’s pending and what’s approved instead of a stack of half-finished portal tabs. The clinician still reviews and signs every letter; the tool never stretches a finding to fit a payer’s checklist.

Can’t hire or keep medical assistants

Posting a medical assistant job and getting a handful of applicants, one of whom calls out constantly, is close to the norm, not the exception, and it’s one of the hardest roles in the practice to keep filled. AI won’t produce medical assistants, but an applicant-screening assistant can text every candidate within minutes, ask the same screening questions of everyone, and book interviews straight onto the office manager’s calendar. The bigger relief comes from the phone, scheduling and eligibility moves above, which free up enough time that the practice needs one fewer person to cover the same day. Hiring and background decisions stay entirely with the owner.

Patients can’t pay their share

High-deductible plans mean the practice is now collecting from the patient directly, not the insurer, and a lot of patients simply can’t pay a surprise bill mailed weeks after the visit. A text-to-pay layer on the practice management system sends a cost estimate before the visit, offers card on file at check-in, and follows mailed balances with a simple pay-by-phone link instead of a paper statement on the counter. Financial hardship, payment plans and any decision to send an account to collections stay with the owner, one patient at a time, and no reminder ever reveals a diagnosis.

HIPAA compliance lands on nobody

In a lot of small practices, the doctor is also the owner and, by default, the privacy officer, and the annual risk analysis that’s supposed to anchor everything else has simply never been done. The most common HIPAA violation among small practices is exactly that, skipping the risk analysis, according to HIPAA Journal’s review of the landscape (https://www.hipaajournal.com/editorial-hipaacompliance-challenges-small-medical-practices/). A compliance-management tool walks the practice through that risk analysis in plain language, stores the policies, tracks staff training, and keeps every vendor’s signed business associate agreement in one place. The risk analysis stays the practice’s own legal responsibility, and no AI vendor in this plan gets turned on before that agreement is signed.

Replying to online reviews is a trap

A one-star review sitting unanswered hurts the next new-patient search, but a well-meaning reply that confirms someone was even a patient can itself be a privacy problem, and it happens more often than most offices realize, according to a large review of healthcare review replies (https://www.hipaajournal.com/study-healthcare-review-replies-hipaa-violations/). A review-monitoring tool drafts a safe reply, from a template that thanks the reviewer and invites them to call the office, without confirming a visit happened, while a human reads and posts every single one. No reply or automated message ever names a date of service or describes someone’s care, even to correct something false.

A day in a medical practice with AI turned on

Picture Dr. Elena Vasquez, who runs a four-doctor family practice in Pasadena. Before, her Monday started with the answering machine full from the weekend, a waiting room stacking up because the front desk was juggling check-ins and a ringing phone, and a stack of prior-auth forms from three insurers nobody had touched since Friday.

With the AI moves running, her Monday starts differently. The phone agent already caught the weekend’s after-hours calls, booked two of them into open slots, and left the rest as callback messages sorted by urgency. The front desk pulls up a short list of patients whose insurance didn’t verify overnight, so it gets handled before check-in instead of after a denial weeks later. Elena walks into room one, and by the time she walks out, the scribe has a draft note waiting instead of a blank template.

She sees, say, eighteen patients that day instead of fourteen, not because she’s rushing, but because the front desk and the medical assistants aren’t drowning in calls and paperwork between each one. None of this changes what Elena decides in the exam room. It changes whether she’s home for dinner, or finishing notes at the kitchen table at ten o’clock, and whether the biller spends the afternoon on the phone with a payer or gets to leave on time.

By six, the office manager has a two-line report: what came in, what got booked, what’s still pending. Nobody had to become a software expert.

What it costs, honestly

Every price here is a range, on purpose, since it depends on your EHR, your patient volume, and the vendor. Roughly: a phone AI receptionist or a scheduling and reminder tool usually runs a few hundred dollars a month. An ambient scribe usually costs a few hundred dollars per clinician a month, and some EHRs now bundle a version of one. A compliance-management tool for your HIPAA risk analysis and vendor agreements usually runs around $100 to $300 a month. Smaller pieces, an applicant screener, a text-to-pay setup, review monitoring, tend to run tens of dollars a month each. Altogether, a working plan of three to five moves usually lands at a few hundred dollars a month across the whole practice, not per tool.

The expensive mistake I see is skipping straight to something custom, a bespoke integration or a purpose-built app, before ever trying the off-the-shelf version for six months. Almost every problem in this plan already has a healthcare-specific product built for it, tested on other practices, with a business associate agreement ready to sign. Custom software costs more, takes longer, and usually duplicates something you could have turned on this month.

What NOT to hand to AI in a medical practice

Medicine is a licensed, regulated profession, and nothing here changes that. Diagnosis, treatment decisions, medication changes and reading back test results stay with the clinician, always; an AI tool can draft and organize, but it never decides, and no AI suggestion is a defense in a malpractice claim or a compliance audit. Anything touching protected patient information goes through a vendor with a signed business associate agreement first, never a general-purpose chatbot or a personal transcription app. Coding and billing integrity stay with a certified coder and the clinician. The practice’s own risk analysis, breach decisions, collections and hiring decisions are the owner’s, with a healthcare attorney where the stakes call for one. No review reply or automated message ever confirms that a specific person is a patient, even to correct something false. Your licensing board and the confidentiality obligations that already govern patient information are the real authority here; AI fits inside those rules, it doesn’t replace them.

How the free AI day works for a physician

The day itself is simple, and it costs you nothing but the day. I show up in the morning and just watch, the phones, the front desk, the back office billing and eligibility work, without getting in anyone’s way. By early afternoon I sit down with you and walk through whichever pain points above actually show up in your practice, using your own call logs and denial reports, not a generic pitch.

You walk away with a written plan naming three to five specific moves, in priority order, each with an honest monthly cost and a plain statement of what it will not touch. No software to buy that day, no contract to sign, nothing custom built. If you want the mechanics of how the day itself is structured, see how it works.

If you run a dental office or a therapy practice instead, the same day works there too, with different pain points; see the pages for dentists and therapists.

If you run a medical practice and want to see exactly where AI would help yours, without anyone trying to sell you anything, book your free AI day — the goal is an evening you actually get home for, trusting the front desk still runs the way you’d run it.