Latest / Becker's Healthcare Behavioral Health / Ambient AI in Behavioral Health: Lessons for 2026 and Beyond

Transcript

Ambient AI in Behavioral Health: Lessons for 2026 and Beyond · Becker's Healthcare Behavioral Health

0:00
Everyone. This is Lucas Voss with Becker's Healthcare.
0:02
Thanks so much for tuning in to the
0:03
Becker's Healthcare podcast series. It's great to have
0:06
you. Today, we're talking about how ambient AI
0:08
and behavioral health is changing lessons
0:11
for 2026
0:13
and beyond. And I'm so excited to be
0:15
joined by Ken Powers. He's the director of
0:16
behavioral health services and licensed independent clinical social
0:19
worker at Yale Health Center. And Jonathan Lyon,
0:22
he's the assistant director of behavioral health services
0:25
also at Yale Health Center. Ken and John,
0:27
thanks so much for being here today. It's
0:28
great to have you. Yes. Thanks for having
0:29
us. Yes. Thanks for having us. I do
0:31
wanna start off with introductions really quickly for
0:33
those that might not know you yet, at
0:35
least. Ken, I'll start off with you. I
0:36
mentioned your title and where you're at, but
0:38
if you could share just a little bit
0:40
about your role and your work in health
0:41
care. Sure. So I, came to EA Hawes
0:44
Health Center about sixteen years ago.
0:47
At the time, there were they had no
0:49
full time, therapists.
0:51
We now have 13 full time therapists.
0:55
I started as a therapist, quickly became the
0:57
director. Hawes Health Center is a federally qualified
1:00
health care center, which means we provide primary
1:02
care,
1:04
behavioral services, chiropractic services, pharmaceutical services. It's kind
1:08
of like a one stop shop.
1:10
Jonathan?
1:10
So I am, the assistant director of Hall's
1:12
Health Center. I'm a clinician as well. I'm
1:14
also a licensed independent clinical social worker. You
1:16
know, one of the things work in rural
1:18
communities,
1:19
you wear many hats. Terrible.
1:21
All of us still carry full clinical caseloads.
1:24
I've been at Hawesville Health for
1:26
just under ten years now. I actually ended
1:29
up there through
1:30
a training program at my, graduate school meant
1:34
to improve access
1:35
and sustainability for rural behavioral health workforces.
1:39
And that's kind of been one of my
1:40
driving forces is wanting to improve patient access
1:43
and, like, provider sustainability.
1:46
Again, thank you both so much for being
1:48
here. I I wanna level set us a
1:50
little bit. Again, from our title, we're talking
1:52
about ambient AI and certainly one of the
1:54
fastest moving technologies in health care right now.
1:57
But, again, it's also different for behavior health.
2:01
How do ambient AI scribes function
2:04
in behavioral health settings today, again, especially when
2:07
we're looking at these different types of workflows
2:10
and different visit types that are different from
2:13
traditional health care settings? Ken, I'll I'll start
2:15
off with you.
2:17
When I'm working with my staff and trying
2:19
to train new staff
2:21
on how we're gonna use
2:24
an Ambient AI scribe,
2:26
I really present it to them as it
2:28
it's just what it is. It's a scribe.
2:30
Mhmm. So you're you're going to teach it
2:34
to do what you want it to do
2:36
just like you would any other scribe.
2:38
You you tell it to do something for
2:40
you, to create a document, to create a
2:43
note,
2:44
and then
2:45
it does create something for you. And if
2:48
it's what you want, yay. If it's not,
2:50
you just give it additional instructions and say,
2:53
use,
2:54
such and such name instead of client.
2:57
So so it's utilized in that way. Another
3:00
way that I utilize it is training my
3:03
staff. Mhmm. So we we do different treatment
3:06
modalities.
3:07
So cognitive behavioral therapy, person centered,
3:10
trauma focused cognitive behavioral therapy.
3:14
And with new staff, it's really nice to
3:16
be able to,
3:18
to have the note produced
3:20
for each of those different modalities.
3:22
And then they have a comparison to say,
3:24
oh, this is what that note looks like
3:27
if I'm doing trauma focused cognitive behavior therapy.
3:29
This is what it looks like if I'm
3:31
doing person centered.
3:32
And it also gives them some insight into
3:36
the techniques that they were using in the
3:37
session.
3:38
So the Ambient AI is able to capture
3:41
that dialogue,
3:42
create transcripts that then can be used,
3:46
to also train. So it's I'm using it
3:48
to train my staff, but we're also training
3:51
the ambient AI to do the things we
3:53
want it to do. Yeah.
3:55
Jonathan, I'm sure this is similar for you,
3:57
but, again, it's it's such an incredible tool
4:00
today.
4:01
How are you using it? What are you
4:03
seeing?
4:04
Well,
4:04
at at first, I was honestly very skeptical,
4:07
because
4:08
as you mentioned, behavioral health sessions are very
4:11
different than medical and primary care. You know,
4:12
sessions tend to be
4:14
longer, a bit more open form, and a
4:16
lot more emotionally dense.
4:18
So one of the biggest things is
4:21
when
4:22
you start the session, you've got all gotten
4:24
all your consents, you start it, get the
4:26
patient, bring them in, and it just runs
4:27
in the background. And you don't change anything
4:30
whether you're doing nondirective play therapy with a
4:32
five year old or you're doing EMDR with
4:34
a 75 year old.
4:36
And then at
4:37
the end, you go in and then you
4:39
adjust your prompt
4:40
based on the type of intervention you were
4:42
using
4:43
and use that to draft the note, review
4:45
it, and so on. Yeah.
4:48
You mentioned that, again, that you were there
4:50
there were some skepticism in the beginning, and
4:52
that certainly goes across health care. There's a
4:54
lot of folks who have skepticism across AI
4:57
and, specifically, ambient AI. What challenges were you
5:00
trying to solve when you were trying to
5:02
implement an AI scribe? And, again,
5:05
after you adopted it, what were some of
5:07
the things that you were seeing? Okay. This
5:08
is actually working. This is doing something incredible
5:11
for us to be able to help patients.
5:14
So initially,
5:16
the reason that we kinda looked into it
5:18
was
5:18
documentation burden and reducing documentation friction, which the
5:23
bane of every clinician that's ever existed, I'm
5:25
sure.
5:27
So
5:28
we looked into it for that reason
5:30
and kind of, hoping to get down some
5:33
of the backlog
5:34
for, you know, notes that were being delayed
5:37
and being completed and,
5:39
easing the burden on the clinicians.
5:41
What's been interesting
5:42
is that
5:44
as we've kinda moved forward, we've noticed
5:46
other benefits as well. You know, we didn't
5:48
go into it with the goal of improving
5:50
productivity, but there was a productivity increase. We
5:53
didn't go into it with the goal of
5:56
kind of improving system stability, and we did
5:59
notice you know, we've been doing it for
6:00
a year and a half now.
6:01
Over the past year, we've had to have
6:03
some clinicians out for, you know, typical family
6:05
medical leave, you know, unpredictable things that every,
6:08
you know, employer and department deals with. But
6:12
the other clinicians were able
6:13
to kinda pick up that caseload without any
6:15
sort of issue,
6:17
when previously, they probably wouldn't have had the
6:19
bandwidth to do that.
6:21
Ken, obviously, you've touched on the fact that
6:22
documentation is a big thing, but you're also
6:24
using it for training as you've mentioned. Yes.
6:26
What does this look like for you?
6:30
So in addition to what John said, just
6:32
this last month, for example,
6:35
of our 13 therapists,
6:37
all 13 therapists saw a minimum of a
6:39
100 billable hours. Wow. Some as high as
6:42
a 164,
6:43
168.
6:45
And using Ambient AI,
6:47
every single therapist had every single note done.
6:51
That includes the encounter notes
6:53
as well as psychosocial
6:55
evaluations. So that's that's a pretty tall tell
6:58
and a very big accomplishment.
7:01
Another thing for me that was kind of
7:03
unanticipated,
7:05
but
7:06
typically when you think of AI,
7:09
many of us, rightfully so, skeptical,
7:12
but also kind of think of it as
7:15
sterile and,
7:17
you know, not not very emotionally responsive.
7:21
So one of the things for me personally
7:23
that,
7:24
caught me with with, when I started using
7:27
AI is when I actually
7:29
it drafted a note for me,
7:31
and I'm reading the note. And I guess
7:34
it's maybe somewhat shameful to admit it, but
7:36
I actually got emotional.
7:38
Like, I read it and I was like,
7:41
wow.
7:42
That's what I did. But but it wasn't
7:44
just what I did. It was the impact
7:46
of, like, this reflected
7:48
in this AI note,
7:50
accurately reflecting what I did, but also
7:53
kind of somewhat like of an objective objective
7:56
third party
7:57
that You made an impact. Yes. And I
7:59
and I'm like, wow.
8:01
I I'm really doing good work.
8:03
And I think that's also been a reaction
8:05
of of many of the therapists that I
8:07
work with when they when they see the
8:10
note. And and I've had them say to
8:12
me,
8:13
wow. I knew I was doing that, but
8:15
I but I wouldn't have taken the time
8:18
to articulate it that way,
8:20
and it helped them see the impact that
8:22
they're making.
8:24
Now for folks that are listening to this,
8:26
Ken, and they're hearing the numbers that you've
8:28
just outlined, which are certainly impressive, and the
8:30
impact that this has had for you and
8:32
your colleagues,
8:34
when we're thinking about partnerships, right, what should
8:37
leaders prioritize
8:39
when they're thinking about selecting a partner to
8:41
work with when it comes to ambient AI?
8:43
And then, again, what are some of those
8:45
metrics, both quantitative but also
8:48
qualitative, right, that evaluates
8:51
success for you?
8:52
Well, I I guess for me, I look
8:54
at it very practical.
8:56
You you can, of course, do studies and
8:58
have research and all that kind of stuff.
9:00
But
9:01
what really makes the impact for me is
9:03
seeing it play out
9:05
in day to day work. Right. So I
9:08
think it's important to
9:10
work with an Ambient AI partner
9:13
that is willing to listen
9:15
and willing to adapt with you so that
9:18
so that when you say,
9:20
you know, one of the m b the
9:21
ambient AI that I, utilize,
9:24
we have the ability to create our own
9:26
templates.
9:27
And so creating those templates, creating those prompts
9:31
is very important
9:32
and also tweaking it. So even if if
9:34
there is a template
9:36
or a a format for whatever we're doing,
9:40
being able to, go in and and utilize
9:43
that and tweak it is is very important
9:45
because you can you can change it.
9:47
I also use it administratively.
9:50
I had a request for a proposal
9:52
from our CEO
9:53
related to work looking at our productivity, and
9:57
I wanted to,
9:59
implement an increase in productivity bonus.
10:01
And she said, we'll submit a proposal.
10:04
So I plugged documents
10:06
into our AI
10:08
as well as gave it content
10:10
and prompts and said, create a proposal for
10:13
me. It was able to create that proposal,
10:16
but also
10:17
attached to it,
10:19
quantifying,
10:21
informations,
10:22
studies, and other things that it cited saying,
10:24
if you do this, here's what
10:26
the likely outcome is gonna be. So I
10:29
was able to submit that to my CEO,
10:32
probably blew her socks off because she probably
10:34
wasn't expecting that,
10:35
and and relatively quickly
10:38
and was able to gain the increase in
10:40
productivity bonus
10:42
by using
10:43
the AI to help me pull in all
10:46
the supporting documentation and information
10:48
for what I was trying to accomplish. Multilevel
10:51
impact across the organization, which I think is
10:53
so crucial. Jonathan, just wanted to ask you
10:55
as well in terms of the partnership piece
10:57
to to you. Why is it so important
10:58
to consider all of those things when choosing
11:00
a partner for this?
11:02
Well, those are, all of those things that
11:04
are, you know, going to make it easier
11:06
to adopt for the clinicians.
11:08
I think
11:09
beyond those specific instance or those, specific items,
11:15
I would encourage a focus on governance, particularly
11:18
ensuring that whatever software you're going with
11:21
meets the requirements for whatever,
11:23
regulations for your localities.
11:25
If you have any clinicians who have guidance
11:28
put out by their,
11:29
you know, their boards or their governing bodies,
11:32
some of them have put out guidance on
11:33
AI, some of them have not, but making
11:35
sure that it's compliant with that. And,
11:39
honestly, the most important thing is is ease
11:40
of use for the clinicians. You know, when
11:43
we started, the intention was literally just to
11:46
help with documentation. Yeah. But, you know, you're
11:49
given this tool and people are finding out
11:51
all of these amazing uses for it. You
11:54
know, I've used it to do
11:57
treatment summaries that span,
11:59
oh, jeez, I don't know, like 60 visits.
12:02
Incredible. And things like that. Stuff that would
12:04
take me hours and hours to do before.
12:06
And it benefits the patient. It travels down
12:08
to the patient because you're able to do
12:10
that. Yeah. It's,
12:12
something that might have taken me like, oh,
12:13
I can probably have it to you in
12:15
a week. Yeah. Yeah. Is you know, I
12:17
can get it to you tomorrow. That's incredible.
12:18
Yep.
12:19
It's so great to have you both on.
12:21
Thank you so much for taking some time
12:22
for us. I wanna leave our audience with
12:24
something actionable for the future here. Right?
12:28
When we're looking at this today for organizations
12:31
that are exploring
12:32
AI use, Ken, I'll start off with you
12:34
here.
12:35
What's that one piece of advice
12:37
that you want people to walk away with
12:39
as they look to the rest of 2026
12:41
with Circles?
12:42
I think one of the things that we
12:44
did,
12:45
that
12:46
may helped make this successful
12:48
is that when we considered Ambient AI, we
12:51
presented it to our clinicians,
12:53
and we asked for their input.
12:56
And we we asked for their input,
12:59
looking at both the software,
13:02
but also
13:03
exploring ethical,
13:05
possible ethical dilemmas, ethical considerations.
13:08
So I I think when you're implementing, especially
13:11
in behavioral health, clinicians
13:13
like to have input rather than just be
13:16
told, we're gonna do this. And when they've
13:19
when they've had the opportunity to have input,
13:21
you've asked the questions, what do you think?
13:24
Here's what we're being told. We had them
13:26
a part of our presentation,
13:28
that, you know, the the when we're considering
13:31
Ambient invite, they were there to see the
13:33
presentation and participate in. And then after that
13:36
was done, we asked questions. We had them
13:38
ask questions. So I think it's important to
13:40
get buy in from the people that are
13:42
gonna be using it because you you can
13:45
get the
13:46
best AI,
13:48
and but if your clinicians are like,
13:51
either they're skeptical or they're consider they're concerned
13:54
about the ethics of it
13:56
or whatever might be holding them back,
13:59
by looking to do this collectively
14:01
and doing it with input, I think it
14:04
advances it a lot easier. Yeah. Jonathan, that's
14:06
something that you've mentioned, right, as critical is
14:08
the adoption piece and making sure that everybody's
14:10
on board. Anything else that you think is
14:11
critical for for organization is to consider here
14:14
as they move forward?
14:15
I would say when it comes to adoption
14:17
and implementation,
14:19
being
14:20
mindful that if your intent
14:23
is to help clinicians and improve patient access
14:25
and patient care,
14:27
it's gonna be very apparent.
14:28
But if it feels like your intent is
14:30
productivity,
14:32
clinicians are gonna feel that as pressure, and
14:34
you're probably gonna have a harder time getting
14:36
buy in from them. And, of course, you
14:38
know, you wanna demystify AI and explain to
14:40
them that, you know, a large language model
14:42
doesn't actually think. You know, it's basically through
14:44
your algorithms and a sweater vest. It's not,
14:46
you know, it's not some magical thing. It's
14:50
words being connected. Yeah. Absolutely.
14:52
Ken and Jonathan, it's so great to have
14:54
you. Thanks again for taking some time. Thank
14:56
you.
14:57
And we also want to thank our podcast
14:58
sponsor, Heide Health. You can tune in to
15:00
more podcasts from Becker's Healthcare by visiting our
15:02
podcast page at becker'shospitalreview.com.