S1E5 Ambient AI (ft. Lewis Marshall, NYC H+H)
Healthy Uptime Podcast Lewis Marshall (NYC H+H) _ Jordan Cooper (Rackspace)
September 10, 2026, 6:11PM
22m 16s
Jordan Cooper started transcription
Jordan Cooper 0:03
with Lewis Marshall, the Chief Medical Officer of New York City Health and Hospital Lincoln Hospital. He is also on the board of trustees of One Brooklyn Health. Lewis, thank you so much for joining us today.
Lewis Marshall 0:17
Jordan, thank you for having me. It's a pleasure to be here.
Jordan Cooper 0:20
So for our listeners, New York City's Health and Hospitals, or NYCH&H, is the largest municipal health system in the United States, headquartered in New York, New York, with 4,500 beds across 11 acute care hospitals, supported by 9,400 physicians. So, Louis, today we're going to be discussing
What else could we be discussing? But artificial intelligence, particularly ambient AI and the intersection of IT and clinical AI. AI is the acronym of the year. So I think a lot of organizations have been grappling with what to do in terms of governance strategies. I think
Ambient AI has been the most popular technology that I've seen health systems adopt in the last, certainly in the last decade. What is NYC H&H doing with ambient AI? And then, you know, we'll review your strategies for how you selected and implemented and scaled your solution.
on how it's improving efficiency, accuracy, and care quality. But please, tell me what's going on with ambient AI and NYCH&H.
Lewis Marshall 1:26
Sure. So we took a long time to develop a governance structure, right? Because any organization that's implementing AI needs a very strong governance structure to make sure the AI is 1 selected appropriately, 2 implemented appropriately,
three managed appropriately, and that we monitor it going forward to make sure that it's doing what we expect it to do. And so H&H has a team that has looked at developing a governance structure, which has now been in place for many years. And it's really served us very well in terms of identifying any AI tools that we may want
Jordan Cooper 1:52
Okay.
Lewis Marshall 2:09
to use, whether they are independent vendor tools or they're embedded within an electronic health record, right? So even those that are embedded within an electronic health record, we still need to make sure that we're using them correctly and then we're monitoring them and that the...
relevance of the AI tool is not changing as our patient demographics may change. And I think that that's something that we all need to consider, you know, as we move forward with this. We have developed an enterprise AI roadmap, which
gives us a roadmap of how and what we are going to look at and how and what we're going to implement, right? So that's been very telling in terms of how we've moved forward with different AI tools. As many systems, you know, Epic is a big
EHR, and as many systems have that, so do we. And so we've implemented some of the AI tools within Epic, such as the inpatient summary, outpatient summary, and a few other tools available within Epic. But looking and talking specifically about ambient AI,
And.
What I would say is the, you know, that's the new shiny object in the room and everybody has the shiny object syndrome. And it's like, I want that. So we did a long review of different vendors. This was before Epic actually came out with their ambient AI tool.
Jordan Cooper 3:40
Mhm.
Mhm.
Lewis Marshall 3:52
which was just recently. And so we had selected a vendor and we implemented it. Most recently it was implemented across the system. And so the way we decided to implement it was first we only put it in primary care offices where the attending is
the primary caregiver, right? Because we all have residency training programs. And so initially we did not want to roll it out to the residents. We wanted the attendings to get comfortable with the tool, with using the tool, with evaluating the tool, and learning how to.
Jordan Cooper 4:17
Mhm.
Lewis Marshall 4:31
to trust it, right?
Jordan Cooper 4:33
Louis, just to jump in there, this was a paid pilot limited to particular practices, or was it for all primary care provider offices and attendings in those offices, or would you not characterize it as a pilot?
Lewis Marshall 4:49
Um...
I would characterize it as...
A pilot at scale, right? So we did implement it across the system in primary care where the attending is the primary caregiver for the patient. So every institution selected initial people, they were trained, and then the tool was implemented and moved forward. And we have some
Jordan Cooper 4:57
Mhm.
Lewis Marshall 5:15
data, which I can mention in a little bit. But so this is something that we are continuing to look at now. We have started to expand it to sub-specialists, surgeons, ENT doctors, you know, and some other sub-specialists.
Neurologist.
And we're also going to be expanding to allow the residents in training as well to utilize the tool.
Jordan Cooper 5:44
So I've had, and I know a lot of our listeners will be somewhat, will be somewhat familiar with ambient AI tools and its great popularity. I've spoken to other healthcare leaders at other systems previously, and they've said, despite it being popular, another health system said, the greatest ROI is the perception that it reduces pajama time, but when you actually run analytics,
there's not a statistically significant difference at that organization in the amount of time that providers are spending after hours in their in-basket or other kind of chores. Are you finding that to be true at NYC and H&H, or are you finding real kind of measurable ROI beyond the definitive improvement in morale?
Lewis Marshall 6:32
Well, you know, hopefully you want all of those things, including improvement in morale. You want, you know, to reduce pajama time for caregivers. But one of the things that we saw was that we saw a difference between
those providers who are using an MB AI tool and those who are not. And the difference was that those who were using it were actually spending one to two more minutes in the record than those who weren't using it, which is not necessarily a bad thing, right? Because that means, hopefully what that means is that the providers are using the tool.
Jordan Cooper 6:57
Mhm.
Lewis Marshall 7:07
but then they're spending time to review the output.
Jordan Cooper 7:11
Mhm.
Lewis Marshall 7:11
right, to make sure that the output from the ambient AI is correct. There's no hallucinations and that the outcomes of that are appropriate. And so I think that that's one of the things that we've been seeing. I mean, it's not a whole big difference, but I think over
Jordan Cooper 7:31
Mm.
Lewis Marshall 7:32
10s of thousands of visits, you know, that can account for a lot of time for a provider. So I think in time, we will start to see the numbers, the minutes within the record for those who are using ambient AI come down as we learn to trust.
Jordan Cooper 7:34
Mhm.
Lewis Marshall 7:51
the tool and we realize or we monitor the outcome, right? And again, if our population starts to change, we want to make sure that that doesn't affect, you know, the outcome of the ambient AI.
Jordan Cooper 7:53
Mmh.
So can you explain to our listeners, so you have a pilot, you've scaled it across the enterprise to every primary care provider who's an attending, and then you, how do you make the decision, okay, we see that these physicians are spending one or two minutes more in the EHR
than Dr. than the control group of providers who are not using ambient listening. How do you make the decision to expand the pilot to roll it out to all the other specialty groups? Are you, because you know, if you're spending, like you said, over 10s of thousands of doctors, it could be fewer visits, fewer, you know, reimbursable CBT codes.
Lewis Marshall 8:47
So, I, I think, I think...
One of the reasons that we've done that, and this is just my thought process, not the system's thought process, is that, you know, by rolling this out to other specialties, we'll get a greater return on investment. So even though we're spending a few more minutes within a chart, you know, reviewing the outcome from Ambient AI.
we think in time that this will come down and we'll really start saving time on the provider side. And then I think about, well, what might that mean, right? And so a lot of reports that I've seen, not at H&H, but outside other organization is that
you know, we look to utilize ambient AI in hopes that we reduce work time for the providers, and then that will allow the providers to increase their templates by one or two patients a day, right, to increase their volume, which, you know, may or may not be a good thing. I mean, from a financial perspective, it might be a great thing, but
Jordan Cooper 9:43
Right.
Lewis Marshall 9:52
One of the things that I think about when we talk about expanding access and expanding A physician's template and the number of patients they may see in a session or a day is what type of support are we providing that clinician, right? So we can't, if we're going to see more people, we need more support for the doctor. We can't just
Jordan Cooper 9:58
Yeah.
Lewis Marshall 10:12
expect the doctor to pick up where the lack of support staff doesn't help them.
Jordan Cooper 10:19
So just to clarify, you are seeing an expansion of doctors' templates and the increase in the quantity of patients they're seeing per day when using ambient AI, or that's where you'd like to be?
Lewis Marshall 10:31
That's where we'd like to be. I don't think we have not, as far as I know, we have not as a system looked at that yet. We're still in the process of looking at ambient AI usage and how that affects, you know, charting, documentation.
Jordan Cooper 10:33
Mhm.
Mhm.
Lewis Marshall 10:50
coding and other aspects of the visit itself, and improving patient satisfaction and provider satisfaction.
Jordan Cooper 10:52
Mhm.
And then...
Could you speak about the patient and provider satisfaction with ambient listening?
Lewis Marshall 11:05
So I think everybody's hope is that with ambient AI, that when you're in the room with the physician, they're looking at you and they're speaking to you and they're not sitting with their back to you and on a computer trying to type in the things that you're telling them, the patient's telling the doctor. So
Jordan Cooper 11:10
Mhm.
Mhm.
Okay.
Lewis Marshall 11:24
You know, I think it's going to take some time for us to get there. I think it's going to take us some remodeling of our exam rooms and placing computers in the right place and exam tables in the right place, because it may not be the same, right, using ambient AI, because you may not necessarily need the computer, right.
Jordan Cooper 11:40
Okay.
Lewis Marshall 11:45
at the desks where you're facing in the wall. So I think from a provider standpoint.
what we're going to see is an improvement in provider satisfaction because they'll actually get to talk to the patient and spend the time with the patient, not necessarily typing. I think from a patient perspective as well, patient satisfaction as we know, you know, if you go into a room and sit down in a chair and have a conversation with a patient,
even if it's only for a few minutes, the patient's perception is that you spent more time with them. So I think that this ambient AI, and if we use it in the correct way, will definitely help us improve our patient satisfaction and give us more time with the patients, you know, to have that conversation.
Jordan Cooper 12:24
Mhm.
So even though you mentioned earlier that your AI governance committee and structure has been in place for years, it sounds like some of your goals are still aspirational. You don't yet have quantitative data on whether patients and providers do have more eye contact, less doctors behind the screens, and more improvements in
patient reported health outcomes or at least patient reported satisfaction scores. Is that true? It's still more aspirational?
Lewis Marshall 12:59
A lot of it is still aspirational, but we do have teams at central office that are looking at those metrics. But we just don't have, I haven't, they have not been shared as of yet, other than the time, you know, time in the chart based upon using it or not using it.
Jordan Cooper 13:11
Mhm.
Would you say that there's a few kind of leaders in the field of ambient AI? Would you say that there are significant differences or that they're more fungible? And what led you to selecting the vendor that you ended up, that NYCH&H ended up going with?
Lewis Marshall 13:38
So I can't comment on how we selected the vendor because I was not involved in that process directly, but I know that we did look at several major vendors in this field. And then the one that they selected was based upon
Jordan Cooper 13:43
Yeah.
Lewis Marshall 13:56
you know, quality, history, how long the system had, how long the tool has been available, right? Because that's one of the things that I think we look at in general is like, how long has this been available and out there and how many other systems are using it as opposed to.
you know, being on the bleeding edge of technology and really just being the first one out of the gate. But I think we've taken a more cautious route, which has, I think, for us, especially here at Lincoln, paid off in terms of how we've rolled it out, how we've gotten buy-in from clinicians.
Jordan Cooper 14:19
Mhm.
Mmh.
Lewis Marshall 14:37
including clinicians who weren't really sure that they would like it, right? Because not everybody necessarily likes all this new technology. So.
Jordan Cooper 14:46
So I know that we mentioned a broader AI strategy through an AI governance board, and I'm sure there are many AI initiatives and priorities. Is there any way in which ambient AI is integrated with any other AI initiative or as part of a broader AI strategy?
Lewis Marshall 15:11
I don't know the answer to that question.
Jordan Cooper 15:15
Well, okay, let's go to the next topic. As I know that we don't have too much more time on this episode, but we wanted to also discuss the intersection of IT and clinical AI. And that's kind of where I was going with that question. So ambient listening AI is clinical AI, right? It's capturing a conversation between the provider and the patient and documenting it within the electronic health record.
Lewis Marshall 15:15
Yeah.
Jordan Cooper 15:38
I'd like to ask if you could elaborate upon what intelligent infrastructure might look like and how technologists and clinicians have been collaborating at NYCH&H in such a way as to fuel innovation and accelerate that adoption and unlocking new possibilities for care delivery.
Lewis Marshall 15:57
Those are all great questions. I think a lot of it is really for us depends upon our leadership, like our system, CMIO is very much interested in making sure that whatever tools that we roll out,
have a clinical impact and not just roll it out because it looks good. So he's looking to make sure that it's one has an impact clinically and two is scalable, right? That's the other issue that we're looking at is whether something is scalable or not and if so, how. And then also looking at
Jordan Cooper 16:16
Okay.
Mhm.
Lewis Marshall 16:36
Um...
you know, how we use the IT and the ambient AI together in terms of decision support, you know, autonomous decision making by ambient AI or by AI in general, and making sure that, you know, there's always a human.
joking with somebody the other day that we're going to start seeing ads for jobs. They're going to say human needed to monitor AI tools. So, but it's, I think that, you know, overall as a system, I think we have a great plan. I think.
Jordan Cooper 17:07
Yeah.
Lewis Marshall 17:18
Like I said, we're probably a little more cautious than some other systems, probably because we are a public health system and we don't necessarily have the resources that other large health systems may have, right, to develop their own technology.
Jordan Cooper 17:37
I guess two more wrap up questions. One, is there, how, while you're considering your AI strategy and intersection of IT and clinical AI, everything we've been discussing today, is any of that affected by, or does any of that affect the direction NYCH&H is going towards
migrating to cloud managed services or your data center strategy in general. As you know, there's a lot of costs associated with AI tokens. I'm just wondering from a system perspective, kind of is, how is AI affecting, you know, where the data is being held and processed?
Lewis Marshall 18:19
So I know, I think from a systems perspective, they are looking at where this information is stored. We've recently, I believe, gone to a cloud system and using data centers to house all of this information. And that's been an ongoing process, I think, from...
Jordan Cooper 18:25
Mm.
Lewis Marshall 18:38
My perspective, as I've seen things develop over the last couple of years, we've moved things from, you know, servers to clouds. And so that's really, I think, given us a better access at the front line. So clinicians have better and easier access to a lot of the data.
So it's been very helpful for us as a hospital and I think for us as a system as well.
Jordan Cooper 19:03
And the last question I'd like to pose to you is if you could presume that I and the listeners of Healthy Uptime Podcast are the CEO of NYC H&H. And we would like to pose the following question to you. So, you know, Dr. Marshall, you've been working with Ambient AI.
at Lincoln Hospital. And, you know, it's been a great pilot. You've even seen you scale from primary care providers to specialists. Could you please justify for us why you think we should be expanding it system-wide and we should be continuing to support ambient AI
given that, you know, we don't have the data right now to say that we've actually had a hard financial, you know, it hasn't paid for itself yet. Can you justify why we should expand ambient AI across the system?
Lewis Marshall 20:01
So in my view, I think we should expand ambient AI across the system to all specialists and all types of providers. I think that the benefit, one of the major benefits is going to be improved documentation, right? Having ambient AI available allows us to...
To.
develop a visit or record a visit and then have that visit summarized with the correct medical information. And it will allow us to have better medical records, more complete medical records. And it will also allow us to
at some point in the future, allow us to improve our coding so that we can, you know, generate the revenue based upon the care that we're providing, right? I find clinicians are terrible coders, and we probably should never code a chart. So I think ambient AI, as it develops more
Jordan Cooper 20:52
Hello.
Lewis Marshall 21:04
will help us with completion of the notes. Chart hygiene will be improved. Some of the new systems coming out that will actually pick up, you know, orders and place orders, you know, during the visit for the doctor to sign off on, you know, will all help to reduce the time that the doctor spends
Jordan Cooper 21:18
Mhm.
Lewis Marshall 21:25
in the chart and give the doctor more time with the patient. And I think overall that's just going to be a benefit for all of us. And then I think in the bottom line is going to improve as well.
Jordan Cooper 21:36
Well, Lewis, I do appreciate you joining us today. For our listeners, this has been Dr. Lewis Marshall, and I'll just add his, there's an incredible array of qualifications. So he is a medical doctor. He's also a JD, MBA, Ms.
health policy and management. So Dr. Marshall, we've covered a lot of ground today on AI and how it's expanded and scaled across the organization. It's ROI, how it's improved patient provider satisfaction. I'd like to thank you for joining us today.
Lewis Marshall 22:12
Oh, thank you. It's a pleasure being here. Thank you, Jordan.
Jordan Cooper stopped transcription
