We sell an AI front office. This is the article about the practices that should not buy one, written because the four situations below are common and a vendor who never mentions them is not being straight with you.
the sales call
the call export, sorted by hour
Key takeaways
- •Misses clustered in one 45 to 60 minute window are a rota problem that costs nothing to fix.
- •A high repeat-caller ratio means congestion during staffed hours, which after-hours coverage does not touch.
- •For a practice with a recall cycle, the largest recoverable number is usually outbound, not inbound.
- •If only the optimistic conversion case clears the price, the price does not clear.
- •The shape that works is misses outside staffed hours, a repeat ratio near 1.0, and a low case that still pays.
Every vendor in this category, us included, has an incentive to describe the problem in a way that makes their product the answer. The honest position is that a meaningful share of practices who take a demo should not buy anything, and that the reason is usually visible in their own data before the first call with a salesperson.
This is that list. Four situations where automating the phone is the wrong move, what the cheaper fix is in each case, and the one shape of practice where the arithmetic genuinely works. It is written from the same measurements the rest of this blog asks you to take, because the point of taking them is that they sometimes say no.
None of this is modesty. A practice that buys the wrong thing churns in 4 months, tells other practices it did not work, and is right. That is worse for us than a sale we did not make.
Your misses are concentrated in one rota gap#
Split your missed calls by hour before anything else, and a surprising number of practices find them clustered in a single 45 to 60 minute window rather than spread across the week. That window is almost always lunch, and it is a rota problem wearing a coverage problem's clothes.
The fix costs nothing. Stagger the front desk's break by 30 minutes so the phone is never uncovered, or divert to a mobile for that hour. Practices that do this and re-measure a month later frequently find the remaining gap too small to price.
There is a diagnostic quality to the clustering that is worth noticing. Demand does not naturally arrange itself into a single hour; if your misses do, something about your own operation is producing them, and something you produce is usually something you can stop producing.
What makes this worth checking first is that the clustered pattern is invisible in a total. A phone system reporting 180 missed calls a month looks like a coverage problem at any volume, and 140 of them landing between 12:30 and 1:15 is a completely different finding that no dashboard surfaces by default.
A miss at 12:40 and a miss at 20:40 look identical in a monthly total and have nothing in common.
Practices resist this one because it sounds like a criticism of the front desk, and it is not. A single-cover desk cannot answer a phone and eat lunch at the same time, and nobody has done anything wrong. It is a rota that was set when the practice was smaller and has never been revisited, which describes most rotas.
The exercise takes about 20 minutes with a call export and a spreadsheet. It is step 2 of sizing your missed-call gap, and it is the single cheapest thing on that list.
Your callers try again, and reach you#
The second situation is a practice whose missed-call count is high and whose distinct-caller count is not. Three hundred missed calls collapsing into 180 actual people means your callers are persistent, and persistent callers mostly get through in the end, which changes what those misses are worth.
That changes what the misses cost. A caller who tried at 11am, failed, and reached you at 2pm is an annoyance and a queue problem, not lost revenue. The revenue case for coverage rests on people who tried once and went elsewhere, and if your ratio says you do not have many of those, the case is weak.
| Ratio near 1.0 | Ratio 1.7 or higher | |
|---|---|---|
| Distinct callers behind 300 misses | About 290 | About 180 |
| What those callers did | Tried once, went elsewhere | Tried again, mostly got through |
| Where the misses fall | Outside staffed hours | Inside staffed hours |
| What actually helps | Coverage | Hold time and peak staffing |
Work out your own ratio: missed calls divided by distinct numbers behind them, grouped inside a 48 hour window. A ratio near 1.0 means one-shot callers and a real coverage problem. A ratio of 1.7 or higher means fewer people, trying repeatedly, and your constraint is capacity during busy hours rather than absence outside them.
The fix for the second case is hold time and staffing peaks, not after-hours cover. Buying evening coverage for a practice whose problem is 11am congestion is a purchase that will work exactly as advertised and change nothing you cared about.
This is the failure mode that is hardest to detect afterwards, because the product does its job. It answers the evening calls, reports that it answered them, and the report is accurate. The number that mattered, people abandoning a queue at 11am, was never in scope and does not appear anywhere in the review.
Your real constraint is outbound, not inbound#
For a practice with a recall cycle, the largest recoverable number is usually not the calls arriving at all. It is the patients who are already due and have not been contacted by anyone, and no amount of answering the phone better reaches them, because they are not the ones ringing.
The dental data makes the shape of this clear. In 2025 one third of dentists reported they were not busy enough, up from one quarter in the fourth quarter of 2024, while new-patient appointment wait times stayed stable through the year [1]. Capacity is available and it is not being filled, which is not a picture of a phone that fails to get answered.
A practice in this position that buys inbound coverage has optimised the half that was working. The chair empty next Tuesday is not waiting on a stranger who could not get through, it is waiting on a patient 8 months past a 6 month recall who nobody has rung.
Count it before you decide: how many patients came due last month, and how many were contacted at all. Practices that have never computed it commonly find a contact rate under half, and that is a problem with a known and much cheaper solution than anything in this category.
Be careful about what counts as contacted while you are at it. A voicemail nobody returned and a bulk email nobody opened are activity rather than contact, and counting them is how a practice arrives at a reassuring 80% that describes what it sent rather than who it reached.
Your volume is too low for the arithmetic to work#
The fourth case is simple arithmetic and it is the one people are most reluctant to do out loud. If you miss 20 calls a month, most of them repeats from 8 people, and your patient value is modest, there is no plausible recovery rate that pays for a subscription.
Run it honestly at three conversion rates rather than one. Distinct missed callers, times a discounted conversion rate, times what a new patient is worth. If the low case does not clear the price and only the optimistic case does, you are buying an optimistic case.
There is a floor below which this category cannot work, and single-location practices with light evening demand are frequently under it. That is not a failure of the practice or the product. It is a fixed cost meeting a small number.
The uncomfortable part is that the floor moves with patient value rather than with volume alone, so two practices missing the same 20 calls a month can land on opposite sides of it. That is why the arithmetic has to be run with your own figures rather than against a benchmark for your specialty.
If only the optimistic case clears the price, the honest answer is that the price does not clear.
Cost the alternative properly too. Median pay for medical secretaries and administrative assistants runs $22.08 an hour as of 2025 [2], so a few hours a week of overtime to work a callback list is a real option that competes with a subscription and is easier to stop.
What the cheaper fix is in each case#
Three of the four situations above have a fix that costs nothing and needs no vendor, which is the whole reason to measure before shopping. The table below is the short version, ordered by how cheap the fix is rather than by how common the situation is.
| If your data says | The problem is | The fix, in order |
|---|---|---|
| Misses cluster in one hour | A rota gap | Stagger breaks, or divert for that hour |
| High repeat ratio, misses in staffed hours | Congestion and hold time | Look at peak staffing and queue behaviour |
| Recall contact rate under half | Outbound capacity | A worked recall process, before anything inbound |
| Low volume, low patient value | Nothing worth buying | Do nothing, and re-measure in 6 months |
There is a caveat on the first row. Staggering a rota only helps if the rota holds still. MGMA polled 357 practices in May 2025 and found 70% reporting turnover stable or falling, yet front-office roles, receptionists and patient service representatives, were the most frequently cited turnover hotspot even among practices whose overall numbers were improving [3]. A practice replacing its desk staff twice a year does not have a rota gap it can stagger away. It has a staffing problem wearing a rota gap as a disguise.
None of those first three rows requires a purchase order. The measurement frequently returns an answer you can act on this week.
The fourth row deserves emphasis because it is the one people find hardest to accept. Deciding not to solve a problem is a legitimate outcome, and re-measuring in 6 months is a plan rather than a shrug.
Work the rows in order rather than picking the one that appeals. They are ordered by cost, and the earlier fixes change the data the later rows depend on: closing a rota gap alters your repeat ratio, and working a recall list alters what your inbound volume even means. A practice that buys coverage first and fixes its rota second has no way to tell which of the two did anything.
When it does make sense#
The shape that works is narrow and worth stating as plainly as the four situations above, because a practice that matches three of its four conditions is still likely to be disappointed. All four have to hold at once, and each of them is checkable against data you already have rather than against a vendor's description of your problem.
- ✓Misses concentrated outside staffed hours, not inside them
- ✓Repeat ratio near 1.0, so those callers tried once and did not return
- ✓Patient value high enough that a few recoveries a month cover the cost
- ✓The low conversion case clears the price, not only the optimistic one
Misses concentrated outside staffed hours rather than inside them. A repeat ratio near 1.0, meaning those callers tried once and did not come back. Patient value high enough that a handful of recovered enquiries a month covers the cost. And a low conversion case, not merely an optimistic one, that clears the price.
One condition is doing more work than the others, and it is the repeat ratio. Volume and value are about size, and size is negotiable at the margin. The ratio is about behaviour: it tells you whether the people you missed are the kind who come back. If they are, almost nothing you buy will show up as revenue, because those callers were never actually lost.
Practices in aesthetics fit this more often than practices in dentistry, because the demand is discretionary, arrives in discretionary hours, and does not wait. The reasons that pattern holds are worked through in why a med spa phone behaves nothing like a dental phone.
There is a fifth condition that is not about numbers at all, and it belongs here rather than in the list because it is a judgement rather than a measurement. Somebody at the practice has to own the thing. A captured request that lands in a queue nobody checks is not an improvement on a voicemail nobody checks, and every practice that has been disappointed by this category can name the week the owner stopped looking.
If your own numbers say the shape fits, the next question is what to buy and how to prove it, which is a different exercise with its own discipline. If they say it does not, you have saved yourself a subscription, and the measurements you took to find that out are the same ones you would want anyway.
