The Implementation-Success Gap: Why Most Deployments Aren't Working in Healthcare AI
With healthcare embracing AI, are deployments delivering measurable success?
The question isn't just about whether the technology works—it's a deeper diagnosis of why promising solutions are struggling to deliver meaningful results for clinician and patient outcomes.
Your hosts Dr. Junaid Kalia (Neurologist & Founder) and Edward Marx (Healthcare CIO) bring perspectives from hospital floors and C-suite boardrooms to this critical analysis. They're joined by Dr. Sara Rana (AI Policymaker, Texas HHSC), whose regulatory expertise reveals the often-overlooked policy dynamics shaping healthcare AI adoption.
Together, they dissect survey data from 43 health systems to expose the complex reality behind the AI hype.
"And that's the problem… that we look for quick results, and that's not something that I believe can ever culminate until and unless there's a proper adoption and implementation strategy set up."- Junaid Kalia, MD
Loading...
What You'll Discover:
[1:19] Priorities for AI Adoption
[2:57] Current AI Use Cases
[5:34] The ROI Disconnect
[9:00] Challenge to Widespread Adoption
[11:11] The Role of Leadership
[14:25] Regulatory Barriers and Solutions
Resources
Journal Article: Adoption of artificial intelligence in healthcare: survey of health system priorities, successes, and challenges (2025)
Transcript
Junaid:
So the main topic of today's podcast is going to be this particular article, which is essentially adoption of Artificial Intelligence in healthcare, survey of healthcare priorities, success and challenges.
Junaid:
So I'm going to start with Dr. Rana. Dr. Rana, what did you think about the study quality itself? Do you think that us making assumptions through this study is a good idea? Because it does seem like that they were able to get a lot of responses.
Dr. Rana
Yeah, I think one of the biggest challenge surveys is just getting the great, a good number of responses. So that's great to see. I think it was pretty great how, you know, everything was divided in different sections. I think sometimes with surveys, like you have to make sure there isn't any bias or like who exactly you're kind of looking at. And I think the other thing to kind of notice too is sometimes with surveys, you don't have all the right answers. You can have branching questions, all these different kinds of responses, but sometimes it doesn't. know, give enough ability for people to give their full kind of comprehensive responses. So that's kind of my thoughts on surveys.
Junaid:
As a Neurocare AI director of neuro ICU and a director of stroke as a clinician, for me, the major goal has always been actually patient safety and quality, which was very interesting that it's from when it was applied to like the general purpose and then caregiver burden and satisfaction, which I didn't realize was such an important goal and such an important goal, to be honest with you. And as a matter of fact, it is the perfect application of AI as we go along. Your thoughts as far as workflow efficiency and productivity, patient safety and caregiver as the top three goals. Is there any other goals comes in mind to you guys?
Ed:
Well, I that was super interesting because I do travel a fair amount around the country and I meet with executives almost every night. so last night, I don't want to name the health systems, but there are probably health systems that are captured in this survey. And absolutely, it was no surprise on the caregiver side. that's why we'll probably talk about it. The largest deployment right now with AI is going to be ambient listening.
And so that really speaks to the caregiver satisfaction. And so there's a lot of attention being paid for that because what they were all saying and we're seeing is that through COVID, there was so much burden put on our caregivers. And now it's time that we do something to bring life and career satisfaction back to them. So I think that's why you're seeing that as the number one in the survey. And we're seeing that as number one in actual adoption for AI.
Junaid:
Now we come to this one. Respondent use case deployed in limited areas or fully deployed. And then the first one is imaging and radiology, early detection, ambient nodes, risk of deterioration, et cetera. So I'm going to let you go first, Sara. What are your thoughts as far as was this a surprise or did you, this was expected?
Dr. Rana:
Well, I think for imaging and radiology, I've seen so many great innovations, technologies in that space. So that wasn't as surprising for me. I'd say medical coding, I was pretty surprised by that being the lowest since I've seen so many companies in that space. I think the risk detection and prediction analysis. So I was in that space, particularly for transplant surgery and built a tool in that space. So I found it to be super exciting and interesting just given that we learn all these risks and risk scores, but do we utilize them each day? AI is just perfect example of how that's able to be used and adopted in such a great way and really help with diagnosing and predicting different complications.
Ed:
Yeah, I wasn't surprised either on the radiology. Radiology has always been the leader when it comes to tech adoption through the ages, right? So that wasn't very surprising. I was just thinking how interesting studies like this would be is if we didn't include just for the analysis anything radiology, because I think it skews AI when you talk about adoption of AI, a lot of it is all radiology.
Ed:
And so I don't know that we're getting a true picture, right? Cause I think it's these other areas that give us a better picture. And it's same almost with ambient voice right now. It would be fun to take those two out and then really see the state of adoption. And then another thing and goes back to what Sara was saying, you know, on the, on the coding side, again, I just wonder how many, how many respondents are just being a little bit conservative when it comes to the aspect of coding and anything to do with reimbursement because of the reason we already talked about the competitive nature of the way our healthcare is set up. So I think that would be interesting analysis, right? Cut out those, the two main areas and see just how much AI are we really doing. And I suspect it's a very small percentage of organizations that are truly, you know, going at this boldly right now.
Junaid:
So now this is interesting, the degree of success. this is again, I was actually surprised, but I would like to know your thoughts before I speak.
Ed:
What I heard, again, this is just fresh and it's not different than what I've already heard from others previous to this week alone. But while there's success, it depends, right? What's the definition of success? There's success, let's talk clinical documentation. There's success, but there's no ROI. So, and it depends what the ROI is. But if we're talking hard dollar ROI, what was reported to me, all these leaders that are doing this is that there really isn't a hard dollar ROI. And they're getting an ROI in terms of yes, clinician satisfaction, less burnout, those sort of things, which is great and super, super important. We all know that. But when it comes to the promises that may have been bantered about regarding financial impact, positive impact, it just isn't happening. So that's the feedback that I had. So when we see percent reported degree of success, I wonder if some of that is like, yeah, we're hitting some of the metrics that we had hoped for, but they're all soft. We're not hitting the hard dollar metric.
Dr. Rana:
Yeah, I totally want to echo that. think even just working in the payer side and thinking about how to adopt all these tools in, they all come back and ask, what is the ROI? And it is really soft at this point. I think the metric of engaging patients in care and communication, I find on the clinician front that that is the number one priority. So it was interesting that I was kind of lower on the list, just given that I think that's what's really ingrained in providers' minds. I think clinical documentation definitely makes sense, or stratification, revenue cycle management. I thought even clinical research being super low is interesting too, given that there's a lot of use cases for AI in that space, given that a lot of tasks can just be done with AI now. That was interesting.
Junaid:
There two perspectives from my end. Number one, I started my clinical practice when there were paper charts. Yeah, I'm a dinosaur. I look very young, just kidding. So what I'm saying is that when I was actually going out to get the job, it was very important that I chose a place where there's a proper EHR. Because what I realized at some point in time, there is no real ally. It is an expense. Epic, certainly expensive, expensive. For me to do my job effectively, at some point in time, as a physician, I said, I'm sorry, dude, I don't care how big you are as a system, how much you're paying me, but if I'm gonna hunt for information and you want me to take care of 44 ICU patients, dude, I need a tool. And that's why one of the ways that I look at AI adoption in the future is that this is not gonna come in only for a hard ROI, it is actually those soft ROIs that are gonna become more and more important.
Junaid:
That's number one thought. Number two is that the problem is the data silos and implementation. The idea is that to say, and people do this all the time and I hate that, is that the ROI is a curve of adoption. You can get to 800 % of ROI, but it takes you five years. If you're saying that, you're gonna get the ROI today and people do this. I mean, that's why we are very ethical and the problem is that you need to make sure that all imaging centers are on board. You need to make sure that all technicians are on board. You need to make sure all radiologists are on board. You need to make sure that everyone's using the system and trained on it. And then you get the ROI and it takes phase implementation and it takes time. And that's the problem that we look for quick results. And that's not something that I believe can ever actually, you know, culminate until and unless you guys there's a proper adoption and implementation strategy set up because the minute you have that proper implementation and strategy set up then and then only all the time like this like if you look at Epic and Serna integration they have implementation plans up to 18 months. Why do we expect that clinical documentation is going to start doing day one? sort of the mismatch between expectations in my opinion and actually driving.
Dr. Rana:
Yeah, I definitely agree with the lack of AI tool maturity. think, you know, when I was in the founder building space, I found there were a lot of issues and hiccups that need to be resolved. The biggest one being hallucinations, the LLM space for me. I think just, you know, again, even in the healthcare space, given that it's so nuanced, given that there's lot of, you know, evidence and facts that are driven, that drive this field, it's really important to be accurate. And I think that was one of kind of the biggest challenges I noticed. I think on the regulatory compliance side of things, to be honest, the best AI policy I've seen in the space currently is the EU policy. And so as I go, you know, trying to write policies for our agency and the state and beyond, I have no reference points just given how do we adopt AI? What are ethical frameworks? What are ways that we're making sure that privacy is being insured?
Dr. Rana:
Of course there are like guidelines and safeguards and policies we've written, but is there's not an actual comprehensive policy or framework that we actually use on a day-to-day basis. That's something that I've struggled with and been working on for a while now, but I think those two are super interesting to me.
Ed:
Yeah, the one that's not on here that I think maybe overrides all of them is sort of a bias for action and just strong leadership. Because with new things like this, everyone is scared. I'm scared of my job. I don't want to lose my job. What if I deploy some awesome AI and something fails? Something goes wrong.
Ed:
And, or we didn't achieve that ROI in 12 months, you know, because we didn't have the patience to wait for the maturity. And so I think that's typically what's the biggest thing lacking is, you you say we're seeing, we're seeing a few organizations, right? That are sort of leading the way and everyone else is not even being a fast follower. They're being, you know, the majority in the middle or whatever the term is used and they're all waiting and not enough people are pioneering.
Ed:
So I think that one, and no one's gonna say that on a survey. One, it's not gonna be asked. I mean, there's the leadership support, but what they're asking there is, is your CEO or board or whatever, that's sort of high level. No, you can't ask in a survey, what are you doing? Are you the failure point to the person who's responding? Of course they can't ask that. And of course we would always think that we're not and that we're like that leader. But I can tell you that a lot of it is just people are scared and they don't wanna take that risk even though you can manage that risk and that's what leaders are supposed to do is take risk but manage those risks accordingly.
Junaid:
So that's another final point that I want to ask you about is that I feel personally that for me in terms of deployment of these systems internationally, is fairly easier, interestingly, as compared to US. the reason behind it is that, first of all, FDA is an amazing organization. I love it. mean, the amount of work they do for the amount of people is insane.
Junaid:
Number one. Number two, they have officially moved over. The internal systems actually have a new Large Language Model and they are improving their efficiency. I'm not going to talk about how some of these politically issues will decrease the staff and have an impact on FDA. But overall, just to give you a little bit of background, we have the drugs platform, FDA for drugs, FDA for medical devices like pacemakers, pens, all of that they did not really have a true pathway. It's called software as medical device. So they took the medical device pathway, converted into a software as medical device, literally, and then move forward rather than building a whole sort of tangent that we are gonna see. So I think, what are your thoughts on that as far as regulation? we need, FDA needs to really sort of deal with this because otherwise we are stopping the innovation to go forward. Again, keeping in mind that this is profile again,
Dr. Rana:
Every time I prescribe a medication, hey, Topiramate can cause one in 10,000 glaucoma. Hey, aspirin is going to cause you bleeding in your knees if you fall. I mean, we know. I mean, we have been doing risk management on an everyday basis.
Yeah, I think adoption is a lot longer in the US and there are a lot of talks about shortening the process of getting through FDA approval. I think as you said, like being able to combine different groups is a great, great solution. I think also the biggest challenge in the biotech space is that it takes even like 10 to 15 years to even adopt, you know, a certain drug or biologic. I think in the AI space, what's really important is getting the tool in as many users and as much feedback as possible. And the barrier is getting through that FDA approval. But I think that's the really one way to see the acceleration and the fine tuning of a product to the point where it's actually deliverable to consumers. So I think like being able to really find a way that as you said, like abroad, that it's a lot easier to adopt and then you're able to get more pilots in and get more feedback and finding the product a lot faster. And I think that's really the point where you see the most acceleration. So being able to find ways to make that process easier is really important.
Ed:
Yeah, and I recall that the bench to bedside metric, you know, back in the day was like 18 years or something. And I can't remember where it is today, but it was reduced significantly, but it's still pretty long. And if I'm a sick person or I'm an overworked, overburdened clinician of some sort, you know, that's way too long. And so whatever we can do to adopt it, would be kind of fun if with the government agencies, again, I respect the agencies and what they're set up to do. But it has been a barrier to quick adoption of these different things at the point already made. And I just would think it'd be so cool if they themselves leveraged AI to make this process faster. Right. And so if they hired and they may have, I don't know, but if they were to hire a couple of, you know, AI gurus to be in that office and then really automate that whole process, because I know it's very manually intensive, and leverage AI themselves. I always said, you know, to be innovative, you must be innovative. So if I'm going to be responsible for something AI, I need to be AI. I need to be leveraging AI in everything that I do. And so, you know, that's the thing. If you have someone involved who's just looking at it from a pragmatic point of view, it's gonna be 18 years bench to bedside. But if it's someone that is AI enabled person, who's super curious, I think we can actually change that viewpoint and be 18 days bench to bedside, not 18 years.
© 2025 Signal & Symptoms. All rights reserved.