PwC and AWS just published a blueprint for agentic AI in healthcare. They’re working with providers to redesign intake, prior authorization, and billing workflows using Anthropic’s Claude. The idea is simple: stop copying data between systems. Let an agent complete the loop.

If you run a medical, dental, or veterinary practice, this matters. Not because you need enterprise-scale infrastructure, but because the same pattern applies at your scale. Every time your front desk manually enters a patient’s insurance details into three different screens, or chases down a prior auth by phone, or copies notes from your EMS into a billing template, you’re paying for work a well-designed agent can do autonomously.

The dollar impact is real. Practices in the $1M to $25M revenue band typically leak $70K to $220K annually on administrative tasks that could run unattended. That’s not a technology problem. It’s a workflow design problem. The question isn’t whether AI can help. It’s which tasks you audit first and how you pilot the right agent without ripping out your entire stack.

The administrative bottleneck you already know

Walk into any practice at 9 a.m. and watch the front desk. One person is on hold with an insurance company. Another is rebooking a cancellation while the phone rings. A third is trying to pull a recall list from the practice management system so they can start calling patients who missed their six-month cleaning.

This isn’t a staffing issue. It’s a design issue. Every task that requires a human to look at one system, remember something, and type it into another system is a candidate for an agent. The PwC-AWS work focuses on hospital-scale workflows, intake that touches a dozen departments, prior authorizations that involve multiple payers and clinical reviewers. But the underlying pattern is identical at the practice level.

Your front desk spends hours each week on tasks that follow a script. A patient calls to book an appointment. Your team checks the schedule, confirms insurance, asks about the reason for the visit, and enters everything into your PMS. If the patient needs to reschedule, the loop repeats. If they don’t show, someone has to call them back and start over. If they go dormant after one missed appointment, they fall off your radar until you manually pull a list and start dialing.

Each of these loops has a decision tree. If the patient says X, do Y. If the insurance requires Z, route to a human. If the appointment is within 48 hours, send a confirmation. These are the exact workflows an agentic AI system is built to handle. The agent doesn’t just answer questions. It completes the task, writes back to your system, and escalates only when it hits something outside its scope.

What agentic AI actually means in a practice context

Agentic AI isn’t a chatbot. It’s a system that can take an action, check the result, and decide what to do next without waiting for a human to click a button. In the PwC-AWS model, an agent might pull a patient’s chart, check their insurance eligibility, draft a prior authorization request, submit it to the payer, and follow up if it’s denied. All of that happens in the background. The clinician sees a note in the EHR that says “prior auth approved” or “needs additional documentation.”

At the practice level, the same logic applies. A front desk voice agent answers the phone, listens to the patient’s request, checks your schedule, books the appointment, sends a confirmation text, and logs the interaction in your PMS. If the patient asks a clinical question, the agent routes the call to a nurse. If they want to reschedule, the agent handles it. If they mention a new insurance plan, the agent flags it for verification.

The agent isn’t replacing your front desk. It’s handling the 60 to 70 percent of calls that follow a predictable pattern so your team can focus on the 30 percent that require judgment. One practice we work with ran a pilot with a voice agent handling inbound appointment requests. Call abandonment dropped from 18 percent to under 5 percent in the first month. The front desk didn’t shrink, they just stopped spending half their day on hold or repeating the same script.

The difference between this and the chatbots you’ve seen is autonomy. A chatbot waits for you to ask a question and gives you an answer. An agent completes a workflow. It books the appointment, updates the record, sends the reminder, and checks back in if the patient doesn’t confirm. That’s what “agentic” means. The system has a goal and the authority to take the steps required to reach it.

Three agents that redesign your admin workflows

We build agents for practices around three core bottlenecks. Each one targets a specific loop where manual work creates delay, error, or lost revenue.

The Front Desk Voice Agent handles inbound calls. It books, reschedules, and confirms appointments. It answers the top 20 routine questions, things like office hours, parking, new patient paperwork, insurance accepted. It routes anything clinical to the right human and logs every interaction in your PMS. The agent doesn’t need training on your specific vocabulary. It learns your schedule rules, your insurance list, and your routing logic during setup. After that, it runs 24/7. Patients who call at 7 p.m. or on Saturday get the same experience as someone who calls at 10 a.m. on Tuesday.

The Recall and Reactivation Agent watches your recall list and reaches out at the right interval. If a patient is due for a six-month cleaning, the agent sends a text or makes a call. If they don’t respond, it follows up through a different channel. If they book, it confirms. If they ignore three attempts, it flags the record for your team to review. The agent doesn’t just send reminders. It rebooks dormant patients. One dental practice we advise reactivated 140 patients in 90 days using an ops agent that worked their recall list every morning. That’s $60K in production that would have stayed on a spreadsheet.

The No-Show Agent protects your daily schedule. It identifies high-risk appointments based on history, sends smart reminders at the right cadence, and fills last-minute cancellations from a waitlist. If a patient cancels two hours before their slot, the agent texts three people on the waitlist and books the first one who confirms. The average no-show rate for practices without an agent runs between 8 and 15 percent. With an agent managing reminders and waitlist fills, that typically drops to 4 to 6 percent. For a practice with $2M in annual production, that’s $80K to $180K in recovered revenue.

These aren’t hypothetical tools. They’re live in practices today. The setup takes weeks, not months. The agents integrate with your existing PMS, they don’t replace it. And the ROI shows up in the first 60 days because the work they’re doing is work you’re already paying for, just inefficiently.

Auditing which tasks should run autonomously

The PwC-AWS blueprint starts with a workflow audit. Which tasks require a human to copy data between systems? Which decisions follow a rule set that can be encoded? Which handoffs create delay because someone has to remember to follow up?

You should run the same audit. Walk through a typical day at your front desk. Count how many times someone picks up the phone, listens to a request, opens your PMS, checks the schedule, asks a follow-up question, enters the information, and hangs up. That’s a loop an agent can close. Now count how many times someone pulls a list of overdue recalls, starts calling patients, leaves voicemails, logs the outcome, and repeats. That’s another loop.

The tasks that make sense for an agent share three characteristics. First, they follow a decision tree. If the patient says X, do Y. Second, they require access to a system of record. The agent needs to read your schedule, check insurance, or update a note. Third, they happen frequently enough that automation saves meaningful time. Booking appointments, sending reminders, managing recalls, these happen dozens of times a day. Ordering new office supplies happens once a month. Start with the high-frequency loops.

We built a worksheet that maps the most common front desk tasks in medical, dental, and veterinary practices to the agent type that can handle them. It’s called the Front Desk Automation Map for Clinics, and it walks you through the same audit framework we use in the Omni Advisory process. You can download it, print it, and use it in your next team meeting to identify which workflows are costing you the most time.

The goal isn’t to automate everything. It’s to automate the loops that don’t require clinical judgment or complex negotiation. Your front desk should spend their time on the patients who need help navigating a complicated insurance issue or the family member who’s anxious about a procedure. They shouldn’t spend their time reading the same script 40 times a day.

Piloting an agent without ripping out your stack

The PwC-AWS model is built for health systems with legacy EHRs, multiple payer contracts, and compliance requirements that span state and federal regulations. Your practice has the same constraints at a smaller scale. You can’t replace your PMS. You can’t break HIPAA. You can’t afford a six-month integration project.

That’s why the right approach is a pilot. Pick one workflow, typically inbound appointment booking, and run an agent alongside your existing process for 30 days. Route 20 percent of calls to the agent. Let your front desk handle the rest. Compare call abandonment, booking accuracy, and patient satisfaction. If the agent performs, expand the scope. If it doesn’t, adjust the routing logic or the script.

The agent doesn’t live inside your PMS. It connects to your PMS through an API or a secure integration layer. When a patient books an appointment, the agent writes the record to your system in real time. When your front desk updates a record, the agent sees the change. The two systems stay in sync without manual reconciliation.

The same pattern applies to recall and no-show agents. They read your schedule, identify the patients who need outreach, and execute the task. They don’t replace your workflow. They automate the repetitive parts so your team can focus on the exceptions.

One multi-location dental group we advise started with a voice agent at their busiest location. Call volume was up 30 percent year-over-year, and they were missing calls during lunch and at the end of the day. The agent handled after-hours calls and overflow during peak times. In the first 60 days, they booked 180 appointments that would have gone to voicemail. They rolled the agent out to three more locations in month three.

The pilot model works because it’s low-risk. You’re not changing your PMS. You’re not retraining your entire team. You’re adding a layer that handles a specific task and proving the ROI before you expand. If you want to see what that pilot looks like for your practice, book a 60-min Omni Audit. We’ll map your current workflows, identify the highest-value automation opportunities, and build a 90-day roadmap. No deck, no generic recommendations. Just three outputs you can act on.

The compliance and patient experience questions you’re already asking

Every practice owner asks the same two questions when we talk about voice agents. First, is it HIPAA compliant? Second, will patients hate it?

On compliance, the answer is yes, if you build it correctly. The agent doesn’t store protected health information in an external system. It accesses your PMS through a secure, encrypted connection, completes the task, and logs the interaction. The conversation is recorded and stored in the same way your current phone system stores calls. The agent follows the same data handling rules as your front desk. If your PMS is HIPAA compliant and your integration layer is secure, the agent is compliant. We build every Omni agent with that architecture from day one. You can read more about how we approach security and compliance in the Omni Advisory documentation.

On patient experience, the data is clear. Patients don’t hate agents. They hate waiting on hold. They hate calling during business hours and getting voicemail. They hate leaving a message and not knowing when someone will call back. A well-designed voice agent answers immediately, completes the task, and confirms the outcome. Patients who’ve interacted with the agents we’ve deployed report higher satisfaction than patients who waited on hold for five minutes to talk to a human.

The key is designing the agent to sound helpful, not robotic. It should acknowledge when it doesn’t understand something and route to a human without making the patient repeat themselves. It should confirm the booking, send a text with the details, and give the patient a way to reach a person if they have a follow-up question. That’s not hard to build. It’s just a different design philosophy than the IVR systems everyone hates.

What the $70K to $220K in annual leakage actually represents

Practices in the $1M to $25M revenue band lose $70K to $220K each year on administrative inefficiency. That number comes from three sources. First, missed appointments. A practice with 15 to 20 operatories or exam rooms typically sees 8 to 15 percent no-show rates. Each missed slot is $200 to $1,500 in lost production, depending on the procedure. Over a year, that’s $40K to $120K.

Second, abandoned calls. If 10 to 20 percent of inbound appointment requests go to voicemail or hang up after holding, you’re losing new patient bookings and reschedules. A practice that handles 200 inbound calls a week and loses 15 percent is missing 30 booking opportunities. If half of those would have converted, that’s 15 appointments a week, or 780 appointments a year. At an average production of $300 per visit, that’s $234K in lost revenue.

Third, dormant patients. The typical practice has 10 to 20 percent of its patient base go inactive each year. These are patients who missed one recall appointment and never rebooked. Reactivating 100 of them is worth $30K to $50K in production, and it costs almost nothing if an agent is doing the outreach. But if your recall process is a manual list that someone calls through once a quarter, you’ll reactivate 10 or 20 at best.

Add those three together and you’re in the $70K to $220K range. The practices at the lower end of that band are smaller, single-location, with tighter operations. The practices at the higher end are multi-location or high-volume, where the inefficiency scales with size. Either way, the leakage is real, and it’s measurable.

The Omni Audit quantifies it for your specific practice. We pull your no-show rate, your call abandonment rate, and your recall reactivation rate. We calculate the dollar impact of each one. Then we show you which agent closes which gap and what the ROI looks like over 12 months. You can see the full audit process and what practices in your vertical typically discover at the AI audit for medical and dental practices.

Why the workflow redesign starts with an audit, not a vendor demo

The PwC-AWS blueprint doesn’t start with technology. It starts with a process map. Which workflows create delay? Which handoffs require a human to copy data? Which tasks happen frequently enough that automation saves meaningful time?

You should start the same way. Don’t call a vendor and ask for a demo of their AI scheduling tool. Pull your front desk team into a room and map the workflows that cost you the most time. Inbound calls. Recall outreach. No-show follow-up. Insurance verification. Billing questions. Rank them by frequency and impact. Then ask which ones follow a decision tree that an agent could execute.

That’s what the Omni Audit does in 60 minutes. We don’t show you a demo. We map your workflows, identify the automation opportunities, and build a 90-day pilot plan. You walk out with three things. First, a workflow map that shows where you’re losing time and money. Second, a prioritized list of agents that close those gaps. Third, a roadmap that tells you what to build first, what to pilot second, and what to defer until you’ve proven ROI on the first two.

The audit isn’t a sales call. It’s a working session. We’ve done this with hundreds of practices, and the pattern is consistent. The highest-ROI agent is almost always the one that handles the task your team complains about most. For some practices, that’s inbound calls. For others, it’s recall. For others, it’s no-shows. The audit tells you which one to start with and why.

If you want to see what agentic AI looks like in your practice, book your Omni Audit here. We’ll map your workflows, calculate your leakage, and show you the fastest path to ROI. No deck, no generic advice. Just a plan you can execute.

The next 90 days

The healthcare industry is redesigning workflows around agentic AI. PwC and AWS are doing it at the health system level. You can do it at the practice level. The technology is ready. The integrations are proven. The ROI is measurable.

The question isn’t whether this works. It’s whether you’re going to audit your workflows now or wait until your competitor does it first. The practices that move in the next 90 days will recover $70K to $220K in leakage this year. The practices that wait will keep paying for manual work that could run autonomously.

Start with the audit. Map the workflows. Pilot the agent. Measure the ROI. Then scale. That’s the pattern that works, and it’s the same pattern the enterprise players are following. You just get to move faster because you don’t have a legacy EHR with 40 years of technical debt.

You can explore more about how Omni agents integrate into practice workflows at the Omni platform overview, or dive into the specific capabilities of Omni Voice and Omni Ops. If you want to see case studies and implementation guides from other practices, check out the EDNA insights library and the Omni learning center.

The workflow redesign starts with one question: which task are you still doing manually that an agent could complete autonomously? Answer that, and you’ve found your first pilot. See Omni for medical and dental practices and let’s map it.