We called 100 radiology practices after hours. Here is what answered.
Original research: the same 100 radiology practices, rung on Sunday evening. Not one reached a person, and eleven were answered by an AI agent.
On Friday we rang 100 radiology practices across Australia and timed how long it took to reach a person who could book a scan. Eight were answered directly by a person, and the median wait to reach someone was 90 seconds. That study ended with a promise: we would ring the same 100 numbers after hours.
On Sunday evening we did. Not one of the hundred calls reached a person. Eleven were answered by a machine that could take a booking. Everyone else got a closed sign.
This is the after-hours data in full, and the most interesting finding is not who was closed. It is who answered.
Methodology
Research led by Nathan Sri, with the Triagents Research Team.
Between 7:25pm and 10:05pm AEST on Sunday 23 August 2026, one caller rang the same 100 Australian radiology practices sampled in the Friday study, one call per practice, and recorded what answered and where it led: a person, an automated booking path, a recorded message, voicemail, or nothing. Whenever a real person answered, we identified the call as a test call once the outcome was recorded. All 100 calls produced a categorised outcome. This is Triagents first-party data; limitations are stated at the end.
What answered on a Sunday evening
- Closed message, pointed to a website or SMS 55.0% 55
- A recorded message and nothing else 15.0% 15
- Answered by an AI booking agent 11.0% 11
- Voicemail 7.0% 7
- Rang unanswered or never connected 8.0% 8
- Closed, with an emergency extension 4.0% 4
Triagents first-party research. The same 100 practices as the Friday study, Sunday 23 August 2026, 7:25pm to 10:05pm AEST.
The most common outcome, at more than half the sample, was a recorded message saying the practice was closed and pointing the caller to a website or an SMS link. It is worth being precise about what that pointer buys. On most of these sites the destination is a request form, not a booking: the patient types their details into a queue that a person will work through on Monday morning. It is the contact-form experience, relocated to the phone. A door sign is not a doorway.
Fifteen more practices played a recorded message and offered nothing at all. Seven invited a voicemail. Eight numbers rang unanswered or never connected, and a published number that rings out does not read as “closed” to a patient. It reads as a wrong number, and nobody tries a wrong number twice.
Four practices deserve a quiet nod: their closed message carried an extension for emergencies and urgent referrers. That is the after-hours line doing its clinical duty, even if it takes no bookings.
Two calls got close
Two practices’ phones genuinely tried to produce a person. One after-hours menu offered to connect us to reception. Another, at a site apparently open for limited services, asked its questions and moved the call towards reception staff. At both, the reception phone simply rang out.
Those two calls are the study in miniature. The systems were willing. The seat was empty. After hours, a transfer button with nobody behind it is just a longer route to the same silence, and it is why the count of calls that reached a person on Sunday evening is zero.
The eleven machines
Here is the finding that matters most for where this industry is going. Eleven sites, spread across six operators in four states, answered after hours with an AI agent on the call itself, asking questions and working towards a booking. Two more practices pointed their closed message at an AI-assisted booking on their website; we count those with the redirects, because a pointer is not an answer.
A year ago that number would have been close to zero. One site in nine is now answering its after-hours phone with software, and those eleven sites were collecting bookings on a Sunday evening while the other 87 phones told patients to try again tomorrow.
Because that is what the Sunday caller does. A patient who rings at 7:30pm is often holding a referral and comparing options in real time. After-hours calls are a small share of volume, roughly 4% by our earlier measurement, but they are disproportionately valuable: every one is a patient who could not wait for business hours, booking with whoever answers first, tonight or at 8:31 tomorrow morning.
Friday and Sunday, together
Put the two studies side by side and the shape of the problem changes. On Friday, 99 of 100 calls eventually connected to something, and the question was how long you waited for a person. On Sunday, the question is whether anything can take your booking at all, and the answer was yes at eleven practices, every one of them a machine.
The Friday study found that a network’s phone reputation is set by its slowest queue. The Sunday study adds the second half: for roughly two-thirds of every week, evenings and weekends, most radiology phone reputations are set by a recording. The practices that answer during that time are competing in a market of eleven, and every seat in that market is currently held by software.
Were your sites among the 100?
Every practice in both studies is anonymous in print unless we are praising it, but we kept the data. If you run or manage an Australian radiology practice and would like to know how your sites went, on Friday and on Sunday, ask and we will send your result privately. It is your data as much as ours, no strings. Use your practice email address so we know where to send it.
What this study is not
The same honesty as part one. This is one Sunday evening, a single pass through the list between 7:25pm and 10:05pm, one call per practice. We recorded what answered and where it led; we did not push every AI agent or web form through to a completed booking, so this study measures whether a door was open, not how good the room is. Sunday is the quietest night of the week, and a practice’s Tuesday 7pm behaviour may differ. And a closed message is not negligence: most of these practices never promised after-hours cover. The finding is not that 89 practices are failing. It is that eleven have decided the after-hours patient is worth answering, and the other 89 are sending that patient to them.
Run the test on your own practice
Ring your own main line tonight after 7pm and listen to what a patient holding a referral hears. If it is a closed message pointing at a request form, count the hours until a person works that queue, and remember that eleven practices in this study could still have taken the same patient’s booking on the spot.
Nobody answered on Sunday night. Your network can.
TriageVoice answers within two rings at any hour, works through the booking under your site’s rules, and hands anything urgent to your escalation path with the context attached. The eleven are already collecting the bookings. Talk to us about joining them.
© 2026 Triagents.AI. This study is Triagents first-party data. You are welcome to cite the findings with attribution and a link to this page.