Half is a believable number, because the pattern is real. At My Smile Miami, a dental practice where we run the phones, the agent answered 3,402 calls in 12 months and more than 1,600 of those calls arrived outside business hours, both measured from the call logs. Month one produced 93 counted bookings and roughly $27K in estimated booked appointment value, calculated using the practice’s average appointment value. The estimate is not collected revenue. Roughly half that practice's phone demand was landing when nobody was at the desk. Read that as evidence the pattern exists, not as proof it is yours.
On cost, Benian publishes no price for anything, including the AI Audit: it is a four-week, fixed-scope diagnosis paid half at kickoff and half on delivery. A voice build is quoted after a free call, because the number moves with your call volume, what languages you need, what the agent has to do beyond taking a message, and which systems it has to read and write. Two of the fixes below, the phone configuration and the coverage schedule, cost nothing and do not involve us at all.
Prove the number before you spend anything
Start with the log. Your carrier, or whoever administers the phone system, can give you 90 days of call detail records, the raw record of every inbound call with its time, its duration, and how it ended, and it costs nothing. Then count five things separately, because they have different fixes: total inbound calls, calls that rang out to voicemail, calls where the caller hung up while on hold or inside the phone menu, calls that arrived outside your opening hours, and calls that hit a busy signal because every line was in use. Half is rarely one problem. Expect the count to split across several of those five, in proportions you had guessed wrong.
Then clean the count. One patient who calls three times in ten minutes is one lost patient, not three missed calls, so group repeats by number before you total anything. And add the category the call log cannot see: calls you answered and did not book. If your front desk picked up, took a name, promised a callback, and nobody called back, that call is lost too, and no phone product on earth fixes it.
Next, find out who was calling. Existing patients and new patients behave completely differently when they hit voicemail. An existing patient rescheduling a cleaning will usually try again; it costs you goodwill and staff time. A new patient with a toothache at 6:40pm is working down a list of clinics and will simply call the next one, and you will never see them in any report. Sample a week of voicemails and, if you record calls, a week of transcripts. Ask your front desk which hours feel worst. Their answer is usually right and usually specific: the lunch hour, the first hour after opening, Monday morning.
Finally, put a number on it. Missed calls per month, times the share that would realistically have booked, times what a booked appointment is worth to you. We publish a free missed-call calculator that does this arithmetic on your own inputs, and it deliberately invents no benchmarks, because a vendor's industry average is not evidence about your clinic. If that number turns out to be small, you have just saved yourself a purchase, and the count cost you nothing.
The cheapest fix is usually inside the phone system
Before anyone sells you software, have someone competent open your phone configuration and check the boring things. How many rings before a call goes to voicemail, and does it roll to a second handset before it does. Whether all your lines can carry calls at once or a second caller gets a busy tone while the first is still talking, which is a miss that never shows up in your missed-call count. Whether the appointments option is the first thing the phone menu offers or the fourth. Whether the number on your Google listing, your website, and your appointment reminders actually rings the line you think it rings. None of that costs anything to check, and a misrouted number or a maxed out line is exactly the kind of miss that never appears in a missed-call report at all.
Look at your after-hours greeting too. Many of them are a recording that tells the caller to call back during business hours, which is the audible version of a closed door. At minimum it should say when you open, and offer one thing the caller can do right now: an online booking address, a text-back number, an emergency line for the thing that genuinely cannot wait until morning.
Then look at the clock. If your peak of missed calls sits in a single predictable window, it is a coverage problem before it is a technology problem. Staggering lunch so the desk is never empty, or moving one person to phones during the morning rush, is a scheduling decision you can make this week and reverse next week if it does not help. It is also the honest control group: if that alone drops your miss rate, you learned something worth more than a demo.
One more free move: give callbacks an owner. A voicemail returned within the hour and a voicemail returned tomorrow are not the same offer to a patient who is working down a list, and in plenty of clinics nobody is specifically accountable for the queue. If that is your situation, understand what buying software does. It turns unanswered voicemails into well organized unanswered messages. The queue is tidier and the outcome is the same.
Staffing, an answering service, or an AI agent
If routing and coverage do not close the gap, you have three real options and they are not ranked by sophistication. Hiring or reassigning is the most flexible: a human handles judgment, apologizes convincingly, and can be trained in a morning. It costs a known amount every month forever, it does not cover nights and weekends unless you pay for nights and weekends, and it stops helping the moment two phones ring at once. For a clinic whose misses are concentrated in a two hour daytime window, this is frequently the correct answer, and it is the answer no software vendor will give you.
An answering service puts a person on the line for far less than a full hire, and it wins when your after-hours volume is low and the calls need judgment more than they need a booking. The structural weaknesses are worth knowing before you sign. Most services read a script and take a message, so somebody at your clinic still calls the patient back in the morning to actually book, which means you have bought a nicer voicemail. Billing is usually per call or per minute, so your invoice spikes in exactly the weeks your volume spikes, which are the same weeks their operators are busiest. Ask any service what their average speed to answer was during their busiest week last quarter, and ask what share of their calls end in a booking rather than a message.
An AI voice agent answers instantly at the same cost on a chaotic Monday as on a quiet Thursday, and the good ones book rather than take messages. It wins when your volume is spiky, when a real share of demand arrives after hours, and when you want the caller to hang up holding an appointment time. The honest weakness is that it is software: a badly built one is worse than voicemail, because it wastes the caller's time before losing them. Which is why the design of the handoff, what happens when the agent is not sure, matters far more than how good the demo voice sounds.
What we ship, plainly, so you can compare it against anything else you are shown. The agent runs on your existing number and in accounts you own, answers in English and Spanish with more than thirty other languages on request, checks real calendar availability before it books, writes the call into your CRM, and hands the caller to a named person on your team when it is unsure instead of guessing. Every call goes to the office as a structured summary. At My Smile Miami it has held a 100% pickup rate since launch across 60 hours of phone time, measured from the call logs. The practice's office manager wrote in December 2025 that Monday voicemail triage went from 45 minutes of waiting on the voicemail system to a summary with transcripts first thing in the morning, and that weekend appointment requests get scheduled without a backlog. That last part is client-reported, not something we measured, and we label it that way on purpose. A build ships in 14 to 21 business days once it is scoped.
Judge every option against your own baseline
Check each option above against the count you built in the first section, not against a demo. Take your five way split and make the provider mark it: which of those rows does this actually move? A coverage change moves the daytime window and nothing else. An answering service moves after hours and overflow, and does nothing at all for the calls you answered and never booked. Nothing on the list moves a schedule that is already full. A provider who cannot mark up your own log in front of you is selling a category rather than a fix for your clinic.
Then make them prove their numbers the way we try to prove ours. Ask for every figure to be labeled as measured, client-reported, estimated, or a projection, and treat a page of round unlabeled numbers as marketing. Ask for the unflattering ones: how many callers asked for a human, how many calls were transferred, how many bookings the system claimed that the schedule does not show. Ask them to demonstrate the urgent path live, the caller whose question the agent cannot answer, rather than the happy path booking. Ask whether you get raw call logs and recordings yourself or only a monthly summary, because a summary is a vendor grading its own homework. Then pull the same 90 days of records a quarter after launch and compare the two counts yourself.
Be careful with patient data, because a clinic phone line hears names, symptoms, and insurance details. Treat the phrase HIPAA compliant as marketing rather than a status: compliance is a property of an entire setup and of how it is run, never of a product on its own, and we will not claim a certification we do not hold. Ask which agreement the provider will sign covering patient data, the business associate agreement being the one that matters. Ask how long call audio and transcripts are kept and whether you can shorten that. Ask which other companies touch the audio on the way through, because the telephony provider and the model provider sit inside your perimeter whether or not the sales deck mentions them.
Then be precise about systems, and about the exit. What we publish for voice work is a connection to your calendar and your CRM, with Google Calendar and HubSpot as the named examples. Anything past that, your practice management system or EHR included, is scoped against the system and the version you actually run, and a provider who answers yes without asking which version that is has not done it, us included. On the way out, ask what leaving looks like: cancel in any month, and walk away still holding the phone number, the recordings, and the configuration. Ours run in accounts you own and log into, so there is nothing for us to hand back.
When not to buy anything
Do not buy if the count came back small. Plenty of clinics who say half discover, once repeats are grouped and after-hours calls are separated out, that they answer most of what matters and the real gap is thirty minutes a day. That is a coverage tweak, not a purchase, and your money belongs on whatever the actual leak is.
Do not buy if your problem is capacity rather than pickup. If your first available new patient appointment is five weeks out, answering every call faster does not produce revenue, it produces a longer waitlist and a worse first impression. Answering more calls only pays when there is somewhere to put the person. Look at your open slots for the next two weeks before you look at any demo.
Do not buy if the calls you lose are calls you already answered. A front desk that picks up, takes a name, and never calls back is a process problem, and an AI agent added on top of it books the after-hours callers while the daytime leak keeps running. Same story if the losses cluster around one script, one insurance question your team cannot answer, or one price conversation that ends the call.
And do not buy if you are not willing to let software touch your calendar. An agent restricted to taking messages is expensive voicemail with better manners. The value only shows up when the caller hangs up holding a time window and your desk finds the appointment already on the schedule. If your clinic is not ready for that, keep the money, fix the routing, and revisit it next year. If you genuinely cannot tell where the leak is, that whole-operation diagnosis is what the audit exists for, and its output is a plan you keep whether we build any of it or not.