In this webinar, EPR Fireworks experts and featured guest Dr. Banerjee of Polk County Fire Rescue explore how artificial intelligence can support EMS agencies today. The discussion covers AI in ePCR documentation, QA/QI and clinical decision support, operational intelligence, and practical considerations for adopting AI responsibly while keeping providers at the center of patient care.

Read the full webinar transcript below.

Speakers

• Neil Arsenault, Chief Revenue Officer, EPR Fireworks
• Dr. Banerjee, Chief Medical Director, Polk County Fire Rescue
• Shane Moss, Chief Experience Officer, EPR Fireworks

Welcome & Introductions

Neil Arsenault, Chief Revenue Officer, EPR Fireworks:

We’ll go ahead and get rolling. We see folks trickling in, but welcome officially to today’s webinar. It’s August 25, and this is Artificial Intelligence in EMS, brought to you and presented by EPR Fireworks.

We’re glad to have you here and look forward to a valuable session over the next hour.

Speaking of the next hour, this does count toward continuing education credit, or CEU, as long as three things happen. First, register for the webinar. Second, attend the full session. It is date- and time-stamped, so we want to make sure you stay through the full hour to receive credit. Third, we’ll have a form for you to complete after the webinar where you can submit your EMT license number and certification information.

Be sure to complete all three steps to receive full credit. Either way, we’re glad you’re here.

I’m your host, Neil Arsenault, Chief Revenue Officer at EPR Fireworks.

I’m joined today by Dr. Banerjee, Medical Director for Polk County Fire Rescue in Florida. We’re very excited to have him join us as our keynote speaker and deliver most of the shared content on AI and EMS.

I’m also joined by Shane Moss, our Chief Experience Officer here at EPR Fireworks. He’s going to give us a tour and bring AI to life within the Fireworks platform toward the end of our session.

Our agenda for today is to lead with the future of AI in EMS. That will be a 30- to 40-minute presentation. Then we’ll go into a preview of AI powered by Fireworks, and we’ll close with a short discussion of what to look for as you think about introducing, embracing, or using AI to its fullest extent within your department.

If time allows, we’ll also go through some of the curated questions. Thank you to everyone who submitted questions ahead of time. We may also have questions in real time. If we don’t get to your specific question today, we’ll answer those after the webinar.

For those joining us today who aren’t customers and may be less familiar with EPR Fireworks, we’re a software-as-a-service provider in the records management space. We were founded in Florida in 2016. We have more than 650 departments that we’re proud to call customers today, and we operate in 35 states.

We deliver one unified platform experience for incident reporting, patient care reporting, prevention and inspections, workforce management, and equipment management. We support state and federal reporting requirements and provide turnkey reporting for ISO and accreditation standards as well.

A lot of what we build comes from your suggestions and innovation. That really drives what we do within our platform.

Thank you for being here, and thank you to our customers for being customers.

With that, we’ll transition over to Dr. Banerjee. Doctor, thank you for being here. We’re excited to hear your take on AI.

The Future of AI in EMS

Dr. Banerjee, Medical Director, Polk County Fire Rescue:

Welcome, everybody.

Polk County Fire Rescue is a large agency. We run about 135,000 to 140,000 calls a year, and we transport more than 100,000 patients a year.

We’ve been with Fireworks since 2022. It was one of the quickest implementations this agency has ever had, and it has been a wonderful experience for us.

Now we’re starting to move into AI. I’ll give you a little bit of our journey, share some of my thoughts, and talk about what the future may hold for everybody.

Understanding the entire situation is important.

What do we know about AI?

Understanding AI can be difficult. It’s neither purely good nor purely bad. It’s a powerful tool, and its impact depends on how we build it, regulate it, and use it.

It has also introduced ethical, social, and environmental risks that we need to be careful about.

What is AI?

AI is a set of technologies that allows computers and machines to perform tasks that would normally require human thinking and work. As humans, we tend to follow step-by-step rules and procedures. AI can bypass some of those steps, solve certain problems, and make decision-making much quicker.

There are different types of AI.

The first is narrow AI. A lot of you may be attending this Zoom on your phone. If you have an iPhone, you have Siri. Other platforms have their own assistants.

Then there’s generative AI, which can generate or transform information and content.

There’s also artificial general intelligence, which is essentially a theoretical future level of AI that could more broadly match human intelligence and skills across many tasks.

With AI, have fun and be careful, because there are three sides to everything: the good, the bad, and the ugly.

The Good

From an office standpoint, AI can improve efficiency and productivity. It can improve decision-making. It can enhance employee engagement through learning and development.

It can also produce a better work-life balance because it can take some time-consuming tasks away from you and give that time back to your family.

It can enhance customer service and communication as well. People can use AI tools to help write a polished message or email that they may not have been able to produce as easily on their own.

The Risks

There’s also an ugly side that we need to be aware of.

Remember, AI is built by humans. Garbage in, garbage out.

Systems may make up false information. Models can use pattern matching instead of true logical understanding.

Prompt injection is another concern, where instructions can potentially be hijacked or overridden by a malicious user or input.

There are concerns around human creative works being used without permission or compensation.

AI can also erode human skills if we begin expecting computers to do work that humans previously had to understand and perform themselves.

There’s misinformation. Automated tools can be hijacked or modified.

There are cybersecurity risks. Hostile groups can use AI to code malware or manipulate what we see.

And there is the concentration of power, where a relatively small number of technology companies control infrastructure that a large portion of us rely on.

EMS at a Crossroads

We’re at a crossroads in EMS.

We’re beginning to use AI in EMS and trying to make our lives better, more effective, and more efficient.

That can be good and bad.

A lot of agencies are using technology around documentation for both billing and patient care.

Nobody really likes doing documentation. I don’t like doing patient records as a physician.

But newer generations of providers are comfortable with technology. They want technology that makes things easy and quick to use.

Technology can allow providers to complete tasks quickly, auto-populate parts of a chart, and finish documentation much faster than if they manually entered everything.

However, it can also overlook things, and those omissions can potentially decrease reimbursement.

So there’s always a balance.

One thing many organizations are using or trying to use is an AI-generated narrative.

The idea is to take information that has already been entered into the system and turn it into a story that explains the patient care provided.

That narrative needs to provide the information that people like me, the medical director, need, while also giving the billing department what it needs.

We want an optimal record that protects the agency, protects the provider, saves time, and reduces repetitive information entry.

We also need to be careful because we’re serving a population that is increasingly critical and informed about patient care. Patients and families can search for information themselves, and that changes the environment in which we’re providing care.

Everything has to have balance.

Benefits of AI in EMS

What are some of the benefits of AI in EMS?

It can streamline documentation.

Voice-to-dictation narratives can be extremely helpful.

AI can potentially optimize dispatch and response times by using analytics and historical data to determine where ambulances should be positioned.

It can support automated chart review for QA/QI and help us understand where we can improve and what we may be missing from the bigger picture.

It can reduce administrative fatigue.

If people working in an office are responsible for reviewing 1,000 charts, perhaps they can focus their time on 100 higher-priority charts while AI assists with reviewing the other 900 lower-risk cases.

It also keeps humans and clinicians in control of patient care.

Even if AI generates a narrative or recommendation, it may not fit the particular system you’re working within.

Remember: If you’ve seen one EMS system, you’ve seen one EMS system.

Everybody is different.

Five Major Goals for AI and Data in EMS

Dr. Corey Slovis has been someone I’ve looked up to for a long time. One of his ideas was that there are five reasons and five causes for everything.

So what are the top five data analysis goals in EMS?

First, improve dispatch and response times.

Second, improve analytics for resource allocation.

Third, improve EMS training and education.

Fourth, improve telemedicine. This may be something many of you are moving toward now with remote consultations.

Fifth, improve decision-making and support for field EMS providers and QA/QI teams.

Ultimately, the goal is to provide and improve patient care.

1. Dispatch and Triage

If you look at dispatch and triage, AI-powered dispatch systems can analyze calls, the history of calls, and other information to help prioritize where ambulances should be located.

Natural language processing can detect words and patterns in emergency calls that may indicate a life-threatening situation and help ensure those cases receive immediate attention.

It can also potentially assist in determining whether an ambulance or a telehealth consultation would be beneficial.

That could allow your medics to work more effectively.

2. Resource Allocation

What are we struggling with on a day-to-day basis?

A lot of agencies have staffing shortages.

We’re dealing with weather and real-time traffic patterns.

AI can allow you to predict where high-volume emergency calls are likely to occur and assign your resources accordingly.

Predictive capabilities can enable EMS teams to position ambulances strategically, reduce response times, improve overall efficiency, and hopefully improve patient care.

As a result, it can also reduce workload for EMS administration.

3. EMS Training and Education

A lot of agencies are now using virtual tools, including virtual reality. We are too.

These technologies can provide interactive learning experiences for situations and procedures that may be difficult to recreate in normal training.

You can simulate going into a fire or performing certain medical procedures and see how your skills perform in a computer-generated environment.

Technology can also tailor training to your weaknesses and help improve deficiencies so you perform better in real-life situations.

4. Telemedicine

Telemedicine is becoming an important part of EMS.

We’re stretched thin. We’re doing so many calls.

There are many patients who may need a prescription or another type of medical assistance but may not necessarily need to be transported to a hospital.

When we use an ambulance to transport a lower-acuity patient, that resource isn’t available for the next emergency call, which could be a cardiac arrest.

AI-supported tools and telemedicine can potentially support decision-making and provide instant recommendations to help make sure patients receive the right care.

In some situations, a physician can connect remotely, talk to the patient or family, prescribe appropriate treatment, and allow the ambulance to return to service instead of transporting the patient to the hospital.

That can be a benefit for both crews and patients.

5. QA/QI and Clinical Decision Support

QA/QI is a very difficult issue.

When we’re transporting around 100,000 patients a year and trying to QA those records with limited staff, it makes life difficult.

How can we optimize that?

Only a portion of our calls involve truly life-threatening situations. Many other calls are lower acuity.

AI can potentially help providers make more informed decisions in high-pressure situations.

Imagine a provider is on scene and isn’t quite sure what’s happening. AI could potentially say, “The heart rate is really high and the respiratory rate is really high. Could this be sepsis?”

It gives the provider a little nudge and helps make sure they’re aware of potential differential diagnoses.

It can also provide algorithms that help QA staff analyze lower-risk cases and identify QA/QI opportunities that may otherwise be missed because of manpower limitations.

That gives us the opportunity to perform meaningful analytics across all of the patients we care for.

Even patients who initially appear lower acuity could actually be sicker than we recognize.

AI can give us additional awareness of things we may be missing.

Decision-support tools can potentially suggest diagnoses, treatments, prognosis considerations, medication information, medication interactions, and other details that may help optimize patient care.

Maximizing Accuracy in High-Risk Calls

We’ve been with Fireworks for several years now, and we’ve been looking at what we can do with EPR and AI.

We recognize that roughly 10% of our calls are truly life-threatening emergencies.

We need to maximize accuracy in those cases.

Accuracy means several things.

Knowing that someone has abnormal vital signs may help you realize that the person is sicker than they appear.

They may be talking to you and acting normally, but they may be tachycardic or hypotensive.

How can we use technology to recognize people who are sick but may not yet look critically ill?

We want preventative awareness before we reach the point where it becomes obvious that somebody is dying.

Pediatrics is also difficult because normal vital signs change as children age.

A heart rate of 140 in a two-month-old is very different from a heart rate of 140 in a 12-year-old.

We need ways to maximize the accuracy of patient care.

Simple things like recognizing abnormal vital signs can trigger clinical awareness.

Why is the patient tachycardic?

Why are they bradycardic?

Why are they hypotensive?

Why are they breathing too fast?

Why are they hypoxic?

Do they have a fever?

Even low temperatures can sometimes be associated with sepsis.

AI can potentially give you those nudges and make you more clinically aware of situations you may not recognize in real time because the person in front of you doesn’t look as sick as they actually are.

EMS Reimbursement and Documentation

We also want to maximize reimbursement, because reimbursement helps fund and improve our agencies.

Why is EMS reimbursement so important?

EMS agencies rely on a combination of public funding, billing, fees, mileage reimbursement, and other sources.

Public payers like Medicare and Medicaid can reimburse at rates that don’t always keep pace with the actual cost of providing care.

That creates a financial gap agencies have to overcome.

We’ve talked about wanting electronic patient care reports that are quick and easy to complete, allow providers to dictate narratives, and capture the information needed.

But agencies are operating in a difficult financial environment.

Even when you submit information, insurers can deny claims because of coding issues, documentation of medical necessity, or other documentation failures.

Rejected or reduced claims increase the financial gap.

ALS1 and ALS2 classifications are examples of why accurate documentation matters.

ALS1 generally involves medically necessary ground transport with an ALS assessment or qualifying ALS intervention.

ALS2 represents a higher level of care and higher reimbursement when specific qualifying criteria are met.

Many agencies may already be providing the care required to qualify, but if that care isn’t accurately documented, the agency may not receive the appropriate reimbursement.

Qualifying procedures can include interventions such as intubation, defibrillation, cardioversion, pacing, needle chest decompression, intraosseous placement, and blood products.

If you’re doing qualifying procedures and they aren’t reflected correctly in the documentation and billing, you may be leaving money on the table.

AI can potentially read the chart, identify procedures, and help ensure the documentation accurately reflects the care provided.

That’s important because ultimately agency administrators need to understand the financial impact of the care being delivered.

Researching AI Accuracy

I’ve dabbled in research, and before we fully embraced AI, we wanted to understand how accurate it actually is.

AI is not perfect.

In one study, we looked at commonly used AI tools and evaluated AI-generated answers to stroke treatment questions.

The results suggested differences in speed and accuracy among AI systems.

The broader point of the research was that AI has potential in stroke and neurology decision support and may be useful as a teaching tool, but more evidence and more work are needed before we can assume it is always accurate enough for clinical use.

We also looked at AI in emergency medicine education and board-style review questions.

One lesson from that work was that using more than one system to validate answers can potentially improve confidence in the output.

The key is validation.

AI can be very useful, but you should not assume the first answer it gives you is automatically correct.

Responsibility and the Future of AI

We don’t know exactly what the future holds.

Hopefully, we do this the right way.

Remember, AI is human-controlled.

Garbage in, garbage out.

It’s our responsibility to be as accurate as possible with the information we have.

We have a big responsibility.

Lives and finances can be affected by what we do.

With that, I’ll turn it back over to Neil.

Keeping People at the Center of AI

Neil Arsenault:

Great. Thank you, Dr. Banerjee, for the presentation.

When you laid out those five goals, the thing that struck me is that four of them, outside of response-time improvement, are really about people.

Resource planning, decision support, training, and extending access to clinicians.

To me, that reinforces that people remain at the center of this.

These are tools that can make us better at our day-to-day work.

Now we’re going to have Shane show a little bit of how we bring AI to life within our products and platform.

Shane, I’ll turn it over to you to give everyone a tutorial of some of the new things we’ve been working on in partnership with Dr. Banerjee, Polk County, and others.

EPR Fireworks AI Demonstration

Shane Moss, Chief Experience Officer, EPR Fireworks:

Thank you.

We’re going to dive into some ePCR-specific AI tools we’ve developed.

The first two are very similar in form and function: Refine with AI and Generate with AI.

Both of these tools are geared specifically around the narrative.

Refine with AI

With Refine with AI, the paramedic comes in and writes their own narrative.

Everything is documented the way they want it.

Now we have the ability to select Refine with AI.

When I select that, I get several options.

I have my existing narrative, and I can choose the style or enhancement I want.

I can select a professional style, a concise style, or a detailed style.

It’s similar to pasting your narrative into an AI tool and saying, “Make this more professional,” or “Make this narrative more precise.”

You can also select no style changes at all.

The second thing we can do is change the format.

We can keep the current story format the original narrative was written in.

Or, if our agency dictates a specific format such as SOAP or CHART, we can take the existing narrative and put it into that format.

We also have the ability to select specific focus areas.

We can ask AI to look at grammar, medical terminology, clarity, structure, and completeness.

Once I select Refine Narrative, the software reads the narrative and reprocesses it based on those selections.

It gives me the AI suggestion alongside the current narrative.

If I like it, I can apply it.

Generate with AI

The other option is Generate with AI.

If I start with a blank narrative and select Generate with AI, the system follows a similar process.

I can decide how I want the narrative formatted, such as a regular narrative, SOAP, or CHART.

Then it analyzes the ePCR and creates a narrative specific to that ePCR.

You can think of it as an older auto-generated narrative capability, but significantly enhanced.

It’s looking at the available data points.

It’s writing the narrative as professionally as it can.

It’s looking at medical terminology.

It’s looking at grammar.

The narrative comes back organized into the selected format, such as CHART, with sections including chief complaint, history, assessment, and other relevant information.

It also gives you information about what it did and processing details.

We also have an attestation at the bottom indicating that the information was generated by AI.

As Dr. Banerjee alluded to, there needs to be human oversight.

The provider reviews the information and, if they agree with it, applies it to the narrative.

Those are the two narrative-focused AI tools.

Quality Check with AI

There are a couple of other things we’re currently working on.

These aren’t fully released yet, but they’re in their final stages as we QA the processes.

The first is Quality Check with AI.

The system can take your ePCR and run quality checks against it.

I have a QC with AI button.

The system goes through and provides an overview of the ePCR.

It looks for completeness, consistency, and clinical appropriateness.

For example, if you documented that you administered a medication, is that reflected in your narrative?

Or is there something in your narrative that you didn’t document in the procedure section?

It’s looking for those types of inconsistencies.

It can return AI recommendations and identify potential issues.

Quality Review Against Agency Protocols

The second QA option takes that further.

The system can allow an agency to upload its protocols.

Instead of validating only against standard NEMSIS 3.5 requirements, it can also validate against your current local protocols.

That takes longer to run, but it can return a much more detailed assessment.

It also gives you the ability to validate each finding.

This comes back to Dr. Banerjee’s point: there needs to be human oversight to make sure the information coming out of the system is accurate.

From a clinical standpoint, your QA staff can confirm findings, adjust their severity, or mark something as a false positive.

You get an overview of what was flagged within the ePCR, whether it’s documentation-based or clinically based.

The system can classify findings by severity and populate QI remarks that can be added back to the ePCR.

It can also analyze those findings based on the agency’s protocols.

That’s where we’re heading with AI-supported QA/QI for ePCRs.

Generate ePCR with AI

The final ePCR-specific feature is Generate ePCR with AI.

This feature allows you to dictate into the ePCR what you’re doing and have AI help populate the report.

For demonstration purposes, I’ve pasted in a narrative as if I had dictated it.

The AI reads that information and returns a field-by-field comparison against NEMSIS 3.5 data elements.

It identifies the fields it believes it can populate inside the ePCR.

If you disagree with something, or AI picked up something incorrectly, you can edit those fields before pushing the information into the ePCR.

It can identify medications, vital signs, patient information, complaints, exam information, and other available information.

The goal is to allow providers to dictate and then auto-populate as much of the ePCR as possible from that dictation.

Ask AI

Moving beyond the ePCR, we’re working on some other AI tools.

One of those is Ask AI.

Within the Ask AI environment, it functions similarly to conversational AI tools you may already be familiar with.

You ask the system questions, and it returns information for you.

One of the interesting things about this tool is that it can work across modules.

The system can use information across different areas of the Fireworks platform.

For example, you could ask how many hydrant inspections, incident reports, or training hours a particular person completed.

Today, you might need to run three or four separate reports to answer that question.

Ask AI can bring that information together in response to a single question.

As a simple example, I can ask how many incidents we had last year.

Then I can ask for the incident-type breakdown.

The system can return that breakdown and create a chart to help visualize the information.

This is one of the tools we’re working on to make it easier to interact with agency data.

AI-Powered Dashboards

Another tool we’re developing involves dashboards.

Any report or query you run within Ask AI can potentially be pinned to your dashboard.

That means you don’t have to repeatedly return to Ask AI and recreate the same query every time you want the information.

Staff Advisor

Another area we’re developing is Staff Advisor.

Staff Advisor allows you to analyze and compare people across your agency.

For example, imagine you need someone to temporarily move into a supervisor role and you have three or four candidates.

You can compare those individuals side by side across the scope of their careers within your department.

Wellness Analytics

Another area, and one I think is particularly important from an EMS and fire response standpoint, is wellness.

The system can look at analytics coming from your emergency response data, whether those are structure fires, medical calls, or other incidents.

From a wellness standpoint, it can analyze how often someone has been exposed to potentially traumatic events.

How many pediatric calls have they had?

How many cardiac calls?

How many trauma events?

It can look across the person’s total call responses.

If you’re connected to our staffing model, it can also look at hours worked.

The goal is to help identify people who may be struggling and allow supervisors to get those people appropriate support.

The system looks at information over a historical time frame.

It considers heavy call volume, high call volume, traumatic events, and other patterns.

Each time the report runs, it can look for changes from the previous period.

The question is: Do I have somebody on my crew or within my service who may need help?

This tool can potentially help supervisors identify those patterns.

Firefighter Exposure Analytics

On the fire side, along the same lines, we can also look at occupational exposure risks.

That includes structure fires, hazardous materials incidents, vehicle fires, outside fires, and other exposure events.

It uses similar analytics to the wellness model, but instead of looking primarily at traumatic events, we’re looking at physical exposure events.

We can historically track how many structure fires, outside fires, vehicle fires, hazardous materials incidents, and other events someone has been exposed to.

These are some of the tools we’ve begun working on from an AI standpoint.

The Two Approaches to AI

Broadly, we’re looking at a couple of different things.

How do we create an environment where staff are comfortable leveraging AI?

How can AI reduce administrative burden?

And then we’re looking at the analytics component.

Once all of this data is in the system, what is the most efficient way to get intelligence back out of it?

Going back to some of Dr. Banerjee’s points:

Where should we position ambulances?

How do we improve response times?

How do we improve hospital wall times?

Those are the types of questions we’re thinking about when it comes to ePCR, AI, and EMS.

Neil Arsenault:

Great. That’s awesome. Thank you, Shane.

We do have time for questions today.

Before we get to those, I want to ask a thought-provoking question.

Dr. Banerjee, then Shane, and then I’ll finish.

What is one piece of advice you would give agencies and departments starting their AI journey today?

Advice for Agencies Starting with AI

Dr. Banerjee:

I would make sure everything is HIPAA compliant.

That’s the most important thing.

Make sure you have barriers in place and oversight around what you’re doing with AI.

AI is great because it can make your life easier and make us look smarter than we are.

But the truth is that we have to be honest about what we’re doing and how we’re doing it.

Remember, what we put into the system affects what we get back.

Be HIPAA compliant.

Be very heavily involved with AI initially until you understand it.

And don’t ever completely take your foot off the gas when it comes to oversight.

Things are always changing, sometimes for the better and sometimes not.

As long as you’re overseeing it and aware of it, you can optimize your resources, provide better patient care, and create a better work environment for your crews and staff.

Neil Arsenault:

Perfect. Thank you.

Shane, what would you say?

Shane Moss:

I would say: Start small. Start boring.

Don’t try to take on a major project right away.

Don’t start by saying, “I want AI to analyze all of my cardiac arrests and tell me whether medications were administered within a certain amount of time,” and then build this massive project around it.

You can go down a rabbit hole with AI, and that’s when it can start producing bad data.

You have to start small.

You need to learn the intricacies of AI, how AI works, and how it returns data.

To Dr. Banerjee’s point, you need guardrails.

You need human oversight as you begin to work with it.

It’s like everything else: You have to test the system before you trust the system.

A lot of what you get out of AI is directly related to the prompt you put into it and the questions you ask.

Start small.

Become comfortable with how you ask questions in AI and how the system returns information and data points.

Then work your way up to some of the larger and more complicated tasks.

Neil Arsenault:

I think that simplicity makes a lot of sense.

These projects can become very large in scope, but before you get there, start small and allow them to mature within the department.

My advice isn’t too far from that.

From a technology perspective, we have a point of view at Fireworks that experimentation with the technology before commitment is important.

Think about your own personal journey using AI tools.

When you first start prompting, you don’t necessarily know exactly what you’re doing or everything the technology can do to help you.

You learn over time.

Experimentation matters.

Q&A: Protecting Patient Data and Maintaining Trust

Neil Arsenault:

One of the questions we received is: How can agencies use AI while protecting patient data and maintaining trust?

Dr. Banerjee, you talked about HIPAA compliance. What would you add?

Dr. Banerjee:

Most agencies have oversight structures that can help make sure you’re doing things with the right intentions.

AI may make your life easier, but you need legal and organizational oversight to make sure you’re staying HIPAA compliant.

We’re dealing with very sensitive information.

Get that under control first.

Start slow.

Start small.

Define what you’re doing.

Then expand from there.

Neil Arsenault:

That’s important.

These decisions shouldn’t be made in a vacuum.

Use the structures within your agency and make these decisions collectively.

Q&A: Efficiency vs. Accuracy

Neil Arsenault:

Another question is: How do we make sure AI improves not just efficiency, but also accuracy?

A lot of the conversation around AI today focuses on doing more work faster.

But quality matters just as much.

From the vendor side, we have more ideas for applying AI to our platform than we have time in the day to execute.

My recommendation to agencies is to join that conversation and tell your technology vendors what you’re trying to accomplish.

But also be selective about the projects you take on.

If quality is what you want to improve, focus first on tools designed to improve quality.

Don’t race toward efficiency just because the ROI sounds attractive.

Particularly when we’re talking about patient care, quality may be the more important outcome.

Technology can support both efficiency and quality.

Q&A: What Should Agencies Be Cautious About?

Neil Arsenault:

What should agencies be cautious about as they adopt AI?

Shane Moss:

First, don’t stick your head in the sand.

AI is here.

Whether your agency chooses to embrace it right now or not, you still need to address it.

There may already be people in agencies taking information, putting it into publicly available AI tools, asking AI to correct a narrative or check something, and then pasting that information back into another system.

Don’t assume it isn’t happening just because you haven’t formally adopted AI.

Get in front of it.

Put policies in place.

Start having conversations as an agency.

And if you embrace AI, you also need to think carefully about accountability.

If AI generates a patient care narrative and you’re later asked whether you wrote it, you need to understand what that means.

That’s why attestation and human review matter.

AI Is a Tool, Not a Replacement for Thinking

Dr. Banerjee:

AI-generated content can absolutely be challenged.

That’s why AI should be a tool to make your life easier.

It is not a tool to replace your thinking.

You’re responsible for the AI content you approve.

You still need to look at what you’re writing and what you’re submitting.

The technology is supposed to make your life easier, but the documentation still needs to be accurate.

If you document the wrong information, you’re still accountable for it.

The agency also has responsibility because QA/QI teams are involved, but ultimately you approved the narrative.

Neil Arsenault:

I wrote that one down:

“This is a tool to make our lives easier, not a tool to replace our thinking.”

That’s probably the quote of the session.

The technology curve is forever changing.

We’ve seen that in software for a long time, and AI is accelerating what computers can accomplish once we set them forward on a path.

But you still have to monitor the technology and make sure the outcomes are good outcomes.

That’s really the key.

AI can help us drive both quality and efficiency in what we’re doing day to day.

The part that excites me is that many departments don’t get the opportunity to review every call.

AI could create an opportunity to expand what gets reviewed so we can provide that feedback loop, add to training, add to education, and ultimately get better at patient care.

Sometimes removing the administrative burden frees us to be more present in the actual clinical setting.

There are a lot of possibilities there.

But getting it deployed effectively is going to require experimentation before we unlock that additional time.

Using a Sandbox to Experiment with AI

Dr. Banerjee:

We were very fortunate because EPR gave us a sandbox to play with.

We volunteered our training department to go into the sandbox and experiment with different scenarios based on vital signs and everything else going on, so we could see how accurate it was.

Having that sandbox gives you an environment where you can experiment, manage, manipulate, organize, and optimize the components to fit your department in the best way possible.

Closing Remarks

Neil Arsenault:

Great.

Those are the questions I had.

I want to thank our special keynote speaker one last time. Dr. Banerjee, thank you for being here. We really appreciate your contribution to today’s program.

A reminder to everyone: there will be a form to complete if you’re seeking CEU credit.

Shane, thank you for showing the platform.

We’d love to carry on the conversation with customers and prospects alike.

If you’d like to experiment with AI in your department, feel free to connect with us. We’ll send out follow-up information with ways to connect and join us in the conversation.

Bye for now, everybody.

Have a great afternoon and a great week.