The same order goes to four imaging providers. Capitol answers first.

The referral is not exclusive

In a market with two or three outpatient imaging options, a referring office does not choose one. It sends the same order to all of them and lets the patient land wherever gets booked first. Capitol works across six states and sixty-plus locations, grown by acquisition, so every region arrived with its own habits and its own paperwork. The thing they all had in common was that a faxed order sat until a person opened it.

• The same order sent to four organizations at once
• Whoever schedules and completes the scan keeps it
• Orders arriving by fax, keyed by hand before anything could move
• Agents working eight to nine hours a day, and overtime the next morning

The order reads itself, and the patient hears immediately

Fax AI reads the incoming fax, matches it to the patient and creates the pending appointment before anyone touches it. That pending appointment goes into a bucket, and the bucket sends the text: Capitol has your order, call us to schedule. When Voice AI makes the call, a faxed order is booked in 2.6 days, against 5.9 days when it waits for a scheduler (17 of Capitol’s locations, August 2026).

• Faxed orders read, matched and turned into pending appointments automatically
• The patient texted from the queue, not from somebody’s call list
• Messages written per modality and per location, down to arrival times and directions
• The robocall removed, because nobody answers a number they do not know

Nobody was replaced. The job moved.

The workload came off the agents rather than the agents coming off the payroll. Twenty to twenty-five percent of the work left the queue, overtime stopped, and the people who were doing it moved to roles Capitol was previously hiring for. New agents reach the point of working on their own inside a week, because there is less to learn.

• 20–25% of the workload lifted off the existing agents
• Overtime on next-day order catch-up eliminated
• Agents redeployed instead of additional agents hired
• New agents up and working on their own in one week

What the queue looks like once the machine has it

The first hour was the thing worth fixing, and it is now measured rather than estimated. An order that waits for a person sits about fifteen hours before anyone opens it. The same order in the automated queue is handled in under thirteen minutes. That gap is the whole argument, and it holds across a thirty-day window rather than a good week.

Roughly two in five faxed orders now file themselves end to end, and the patient match on them is accurate to better than ninety-nine and a half percent. What the AI cannot index it logs with a reason, so the exceptions are a worklist rather than a mystery.

• About 15 hours waiting in the staff queue against 12.7 minutes in the AI queue
• 505 orders sorted a day by AI; Capitol’s best human sorter manages 67
• 39.4% of faxed orders filed with no human touch, peaking at 40.5%
• 99.59% patient match accuracy on automated orders
• 2.6 minutes of handling per order, measured on the audit log
• 1,628 staff hours taken out of order processing in thirty days
• Every order the AI cannot index logged with the reason it stopped

Abby does not get tired

Scheduling calls run through the same program. The voice agent places them, answers them and books them, and it finishes the great majority on its own: fewer than one answered call in six reaches a person, and the most common reason for a handover is simply that the patient asked for one. Patients rate the experience at ninety-three percent satisfaction across two separate measurement windows.

Capitol’s teams stopped calling the tools by their product names a while ago. The voice agent is Abby, the engine that reads incoming faxes is Mike, and the one that reads finished reports for follow-up recommendations is Frank. It is a small thing that says something real: the staff talk about them the way they talk about colleagues, because that is how the work is now divided.

The point is not that the calls are cheaper. It is that they happen at hours when nobody is in the building, which is often exactly when a patient who has just been handed a referral is free to deal with it.

• 15.2% of answered calls transfer to a human, so the AI completes about 85%
• The single biggest transfer reason is the patient asking for a person
• 93% patient satisfaction on AI-handled scheduling, measured twice
• 357 hours of patient calling absorbed in one month
• Appointments booked overnight, with no one on site to take the call