The Front Office Is Finding Air: One Network’s Early Returns on Operational AI
A referring physician orders an MRI of the lumbar spine and hits send. That order lands at three, sometimes four, outpatient imaging centers simultaneously. The center that contacts the patient first books the appointment. The rest miss out on the revenue.
Tim Haley lives inside this math every day. As chief technology officer of Capitol Imaging Services, he oversees 60 imaging centers across Louisiana, Mississippi, Alabama, Texas, Georgia, and Florida. In several of those markets, the difference between capturing a patient and losing one to a competitor comes down to minutes.
That competitive pressure is what pushed Capitol Imaging to stop evaluating operational AI and start deploying it. But the path from decision to deployment was a bit messier. That honesty is the most useful thing Haley and his technology partner, AbbaDox CTO Isaac Aronov, can offer radiology leaders evaluating the same move.
This is the story of the numbers, the failures and a look inside the playbook of managing Capitol Imaging’s first year with operational AI.
Can this work in a messy operation?
Capitol Imaging did not grow according to a blueprint. The organization expanded through acquisition, absorbing centers across seven states, each with its own workflows, staff cultures, referral patterns and operational quirks.
“We would love for the entire organization to be standardized to the T, but it’s just not the way that we’re built. Every region has its own scenario, its own caveats.”
— Tim Haley, CTO, Capitol Imaging Services
AbbaDox’s approach addressed this head-on. Rather than building a single, monolithic AI system, the platform uses modular building blocks that can be configured and activated independently for each location. Aronov describes it as layering capabilities one at a time rather than flipping a switch.
“These are all building blocks that we can collect together into something more meaningful. You need to take it in some ways baby steps, one use case at a time.”
— Isaac Aronov, CTO, AbbaDox
Capitol Imaging started with the markets that had the most acute need, deployed the scheduling AI first, established guardrails specific to each region, and expanded from there.
What do the first 90 days actually feel like?
Three realities emerged quickly that neither side fully anticipated.
The tribal knowledge problem.
Before AI can automate a workflow, someone has to articulate what that workflow actually is. Aronov found that a tremendous amount of operational knowledge lived inside the heads of longtime staff members and had never been formally documented. Extracting it was not a technology challenge — it was a change management challenge, and it required more time and consultation than initial project plans accounted for.
For any imaging center considering a similar deployment, the implication is direct: If your workflows live in your staff’s heads rather than in documented processes, plan for that extraction work before you deploy anything. Add at least six weeks to your timeline.
The voice that flopped.
AbbaDox had successfully deployed Abby—their voice AI scheduling agent—in a northern market. When they brought the same voice, pace and conversational style to Capitol Imaging’s southern locations, the deployment failed. Patients were uncomfortable. Some asked outright: “Are you a robot?”
The AbbaDox team runs sentiment analysis on every call, tracking which patients respond positively and which show hesitation. The data pointed clearly to the problem. They changed Abby’s voice, adjusted the speaking pace and bedside manner, and saw success rates increase almost immediately.
“As soon as we changed it, we had an overnight increase in success.”
— Aronov
It was a pointed reminder that operational AI in healthcare is not just a technical implementation. It is a human experience that varies by geography, culture and patient expectation.
The guardrails that scaled.
Capitol Imaging’s first regional deployment required hands-on customization, namely mapping local scheduling rules, payer-specific nuances and referral patterns unique to that market. But once the guardrails were established for that first region, subsequent markets went faster. The framework was transferable even when the details were not. For multi-location networks weighing whether to pilot AI in one market or attempt a system-wide launch, Capitol Imaging’s experience points clearly toward starting narrow and scaling the playbook.
The platform’s fax processing capabilities reinforced the early momentum. Across the network, 96% of inbound faxes are now processed automatically, with more than 17,000 imaging orders handled in a single week. For front-office teams that previously spent hours sorting and manually entering fax orders every morning, the operational relief was immediate.
What happens to the people?
Staff displacement is the unspoken fear in every AI implementation conversation. Haley and his team saw it as a key topic to address directly and emphasize that patients need people.
Capitol Imaging frames automation as role evolution, not elimination. As front-office tasks shift to AI, patient care coordinators are moving into clinical support: Helping patients on and off tables, setting up scanner protocols, expanding into functions that add more value to the operation and to their own careers.
“We’re not getting rid of you. We’re just kind of modifying your role. And so far when we’ve done that, it’s exciting for them. They thought they were coming to just register a patient. Now they’re being moved into a different role.”
— Haley
Aronov adds a note of realism. Change management does not always move at the right speed. Some progress outpaces the organization’s readiness. The human factor, he says, is not always given enough attention to make sure that staff are ready and that concerns are addressed before the next phase begins. Getting that calibration right is an ongoing effort, not a one-time project plan.
Spotlight: When a Patient Calls at 11 PM Speaking in Creole
One of Abby’s most quietly powerful capabilities is language flexibility. The AI scheduling agent currently handles patient interactions in English, Spanish, and Creole — and additional languages are in development.
Aronov describes patients finishing night shifts at 11 PM, receiving a text from Abby, and completing a scheduling call in three minutes in their preferred language — with no human staff involved.
“If it wasn’t for that, Tim would have to have a staff of people on call to answer at 11 PM and also be able to speak Spanish or Creole,” says Aronov.
For networks serving multilingual patient populations, the after-hours language capability is not just a convenience. It is a direct driver of appointment capture that would otherwise require significant staffing investment.
How do you know it’s working?
Haley’s answer is disarmingly simple: “Volume fixes everything. If we can fill spots efficiently, then we’re doing a good job.”
Patients are landing on the schedule without expanded staffing. Third-party scheduling vendors that Capitol Imaging had been relying on are being scaled back as AI absorbs their function. Across AI-handled scheduling interactions, the platform maintains a 94% patient satisfaction rate.
The next metric Haley and Aronov are building toward is what they call the inverse measurement: Not just how many appointments AI booked, but how many calls the staff did not have to make. For operations leaders building the internal case for AI, that number may matter more than any conversion rate because it translates directly to labor hours recovered.
What’s coming next?
Haley’s vision extends the AI-powered workflow from first fax to front door. The full sequence: An order arrives, the appointment is set, insurance benefits are verified, the patient receives electronic forms via text, completes them before arrival, pays their estimated responsibility, and checks himself or herself on arrival. Minimal paper. Minimal manual intervention. Maximum speed from referral to scan.
“Those are the things that I’m really, really excited about.”
— Haley
Aronov sees the opportunity reaching well past the front office. After the radiologist reads a case, AI can identify incidental findings, route follow-up recommendations to referring physicians, and offer to schedule the next appointment automatically. This care coordination work used to depend on a single staff member remembering to make a call. When that person is busy or absent, patient follow-through suffers.
“AI will make sure that the care coordination and the follow-up actually happen. The goal is to close the loop so that nothing falls through the cracks and every patient gets the follow-up they need.”
— Aronov
Yaniv Dagan, CEO of AbbaDox, sees the same pattern accelerating across the company’s customer base. “What Tim is describing at Capitol Imaging is happening at networks of every size right now,” he says. “The centers that started six months ago are already asking what they can automate next. The ones that haven’t started are watching their competitors schedule patients faster, collect payments earlier, and operate leaner. The gap between those two groups is widening every quarter.”
Haley puts it in simpler terms. His team is not chasing a technology trend; they are building an operation that can grow through its next five acquisitions without proportionally growing its front-office headcount. For the imaging centers still running the math on whether to move, his advice is practical: Pick the market where the pain is sharpest, start there, and let the results make the case for the next one.
Is Your Operation Ready? Three Signals Worth Checking
Based on Capitol Imaging’s experience, these are the conditions that either accelerate or complicate an operational AI rollout:
1. Your workflows live in people’s heads, not in documented processes. Plan for an extraction phase before go-live — add at least six weeks to your timeline.
2. You serve patients across multiple markets, languages and demographics. Voice and communication style will need to be tested and tuned per region, not assumed from a successful deployment elsewhere.
3. You’re still relying on third-party scheduling vendors or manual fax processing. These are typically the fastest workflows to automate and produce the most immediate operational relief.
Watch the Full Webinar
Tim Haley and Isaac Aronov discuss the full scope of Capitol Imaging’s AI deployment in their Radiology Business webinar.
“Your Front Office Is Drowning. AI Can Help. Here’s What We’ve Learned So Far.”
Watch the replay here · Learn more at abbadox.com