Latest / AI for Founders with Ryan Estes / The Behavioral Health Crisis Is a Data Problem

Transcript

The Behavioral Health Crisis Is a Data Problem · AI for Founders with Ryan Estes

0:00
Your doctor isn't the problem. The system is blind.
0:02
That's the problem right now. The average patient gets seen
0:06
and then disappears. No data, no signal, no clue what
0:09
happens next. And the gap.
0:11
It's where people spiral. It's where diagnosis is get
0:14
delayed for years. It's where $280 billion quietly
0:18
burns every year. But now imagine this and AI
0:21
calls you not to replace your doctor, but to notice you.
0:24
It hears your voice, it sees your face.
0:26
It tracks tiny changes over time.
0:29
Not what you say, but how you say it.
0:31
The breakthrough is in diagnosis.
0:32
It's continuous context. A founder could build this into
0:36
anything. Medication adherence, employee
0:38
burnout, customer health. Because behavior is the ultimate
0:41
data set. This is in theory.
0:43
It's already happening. Systems are learning from
0:45
millions of conversations, refining models, reducing admin
0:49
times by up to 88%, and turning check insurance into predictive
0:53
intelligence. So if your product could
0:55
actually understand how your user feels, how would it change
0:58
everything you've built? Welcome to AI for founders.
1:01
I'm Ryan Estes. In this episode with Lauren
1:04
Larson, CEO of Vadera Health, you'll learn how AI is decoding
1:08
human behavior, why most founders are missing the biggest
1:11
data layer, and how this tech is reshaping healthcare, SAS, and
1:15
anything driven by human decision making.
1:18
This is for builders who want an unfair advantage in
1:20
understanding people. If you want that edge, go to
1:23
aiforfounders.co. We study what's working before
1:27
it's obvious. We give you the playbooks before
1:29
they're called best practice. If this made you think
1:31
differently, please leave a review like you're grading a
1:34
startup pitch. Brutally honest and wildly
1:36
helpful. And if you want to get booked on
1:38
podcasts that actually move markets, go to kickcaster.com.
1:42
They don't just get you interviews.
1:44
They position you as the authority in your category.
1:46
Top founders trust Kick Caster to turn conversations into
1:50
customers. That's Kick Caster by Moburst,
1:57
or and welcome to the show. I'm excited to chat today.
2:02
Yeah, I'm looking forward to it. So a big theme of this podcast
2:05
is head health tech, Med tech, pharma tech, everything tech,
2:10
but particularly in the healthcare industry, I think
2:13
there's some really amazing opportunities that we're just
2:15
coming into now. So I'm really excited to to
2:19
learn a little bit more about your company, Videra Health.
2:22
So give us an introduction to to your company.
2:26
Yeah. So maybe I'll start with the
2:28
company I was at before called Hireview, where we did video
2:31
based job interviewing and you would interview, you could
2:34
interview a company interview to 100,000 people in a week.
2:37
You could take an interview where you might be rejected
2:40
based on your resume, but you'd have a chance to sort of tell
2:42
your story and be heard where you'd never have a chance
2:45
otherwise. And that was a, it was a really
2:47
amazing company to build. At some point we added AI to
2:51
that. This is like 2012, 2013 that
2:55
would listen to the interview and help surface who the best
2:58
candidates were and say, look, you had 1000 people interview
3:01
this week. Here are the top 50 that you
3:03
need to meet or the top ten you need to meet for the company.
3:06
And let people who might be biased against or discriminated
3:10
against get sort of seen for who their what their actual
3:14
abilities are versus what hiring managers or recruiters might
3:18
think about them. And then that really LED us into
3:21
sort of saying, can we apply this into healthcare?
3:23
Could we start to use this ability to see and hear a person
3:27
and understand who they are in healthcare?
3:31
And I think there were sort of two ways we thought about that.
3:33
The first one was when you leave a doctor, you're on your own.
3:38
Like you go, you're seeing, you get probably pretty good care.
3:42
And then when you leave, no one thinks about you again.
3:44
And we wanted to use AI to sort of say, can we just check in on
3:47
people all the time with AI and ask them how they're doing and
3:51
then use algorithms to measure how they're doing and then
3:54
surface people who are in crisis or need to need care back to
3:58
their providers who don't have time to check in on everybody,
4:00
but they need to know about the five people today who really
4:03
need help. And then the second use case was
4:06
kind of around, could we diagnose diseases and see them
4:09
earlier? Because, you know, you'll hear
4:11
what they're like. It took me 7 years to get
4:12
diagnosed with that. Like maybe that's a solvable
4:15
problem. So that's where we started with
4:17
the DARE. But obviously there's a lot
4:18
beyond that. But that's that's kind of what
4:21
we started out to build in 2019 and I've been building ever
4:23
since. Incredible.
4:25
So with your first company the the automation for video
4:30
interviewing, did Vidara come out of that or what was the the
4:34
outcome for that company? We sold that company in 2019.
4:37
It's still around. They just did an acquisition
4:39
today that they announced. So that's continued to grow.
4:42
They've interviewed 100 million people for jobs since then.
4:46
And there's no relation between the companies other than we had
4:49
sort of learned a lot about how to understand someone from a
4:54
video. And you know, we took that
4:56
learning and expertise and, and took it into healthcare.
4:59
Great. So is Videra Health, is this, is
5:02
it a video platform? Yeah, fundamentally, we want to
5:06
see people. You'll often hear doctors say,
5:08
like, if I can just see my patient for like 15 seconds, I'd
5:10
know how they're doing. And so we really believe in the
5:13
power of video. Now, not everyone wants to
5:15
engage in video. So, you know, people can just
5:18
talk. They can text in a lot of cases.
5:20
But yeah, the idea is ask people questions, see them, hear them,
5:24
quantify their responses, and then get them the care that they
5:27
need. Cool.
5:29
OK. So the the customers then are
5:31
actually the clinicians themself?
5:33
Yeah, I mean, we're usually selling to to bigger health
5:35
systems, the behavioral health space, you know, people treating
5:39
people for substance use, eating disorders, mental health, those
5:43
kinds of things. Great.
5:45
OK. So you you sell into the big
5:47
hospitals, they provide this as an opportunity to their
5:52
physicians. How does it kind of integrate
5:55
into kind of the the hospital's tech stack that they have
5:58
already? Yeah.
6:00
I mean, one of the really interesting things about
6:02
Healthcare is how, which I haven't seen in other areas, is
6:05
the reluctance of the system of record to let anybody else
6:09
integrate. So there are a lot of challenges
6:11
to integration and plugging into those workflows so that our
6:17
perfect world is the clinician doesn't have to think about
6:19
this, right? Like if a clinician has to make
6:21
a decision and think, oh, maybe I should put this person in a
6:23
follow up program to see how they're doing, chances are
6:26
they're busy. They're going to forget how to
6:27
forget how to do it. There's some friction there.
6:30
So ideally we are looking at the record and say, OK, this person
6:33
missed an appointment, let's let's do a check in.
6:35
They got this diagnosis, they got this prescription, let's
6:39
follow up and see how they're doing.
6:40
A week later, 2 weeks later, 3 weeks later, this person just
6:43
got discharged. Let's do a reach out and follow
6:45
up and see how they're doing. So hopefully the trigger is
6:48
automatic. Then we manage the connection
6:50
with the patient, collect all of their data and then push that
6:53
back into the health record so that hopefully an alert, you
6:58
know, I'm trigger an alert so that if there's something that
7:01
needs attention and it will be seen by the right people.
7:03
But hopefully it you just get notified of things you would
7:07
have wanted to know but didn't think about having, wanting to
7:09
know or figure out how to get it or you just see better patients.
7:14
In other words, you know, a lot of times our healthcare system
7:16
sees people who are good at making appointments, not the
7:20
people who actually need to be seen.
7:23
But with this, you kind of reach out, engage people who might not
7:25
proactively reach out and then go, hey, you know what?
7:27
We're going to have someone call you and make an appointment and
7:30
get you back in. And so providers should have the
7:32
experience of just going, well, I'm just seeing a lot of people
7:34
who really, really needed me today versus or just, yeah, I'm
7:37
a little, I'm a little, I have a little bit of anxiety, which is
7:41
perfectly valid. But there's people who are
7:43
sitting on waiting lists and sitting at home suffering who
7:46
aren't getting in. That's great.
7:49
Doctors are really good about learning about people who are
7:52
good at making appointments. It's.
7:56
So that's the real estate thing I've heard all day.
7:58
It's so true. Yeah.
8:00
You know that. That's that's why the moms make
8:02
all the the doctor's appointments for the family,
8:05
because it was up to dads. We it wouldn't get done.
8:08
No one would go. Fascinating.
8:11
OK, so let's say that I'm the patient.
8:13
Now it seems like there's a certain kind of like diagnosis
8:16
or conditions that, that this makes a lot of sense, especially
8:19
video, right? So like mental health, like
8:21
literally at a glance, you can see how somebody's doing, right?
8:24
Are there? Are there use cases that are
8:27
maybe more beneficial than others?
8:29
That's a great question. I think, you know, we focused on
8:31
behavioral health upfront. We started on something called
8:36
tardive dyskinesia, which is a movement disorder.
8:39
And you know, it looks like someone kind of has involuntary
8:42
movement usually around their lips or their mouth and you
8:46
might see. So once you've seen it, you kind
8:47
of start to see it in public. And it, there was never a good
8:51
way to diagnose it. Only 15 to 20% of the people
8:54
even today have been diagnosed. And so we started with that
8:56
because it was inherently visual.
8:58
And we've started working on other movement disorders like
9:01
Parkinson's disease, etcetera. So things that are like the
9:04
clinician has to see you to diagnose is really important.
9:09
We do a lot of depression and anxiety, PTSD, which you can
9:12
measure from voice or language if you have enough sort of turns
9:15
of the conversation with people. But we've really focused on
9:18
mental health and behavioral health for now.
9:21
But we're building things in Alzheimer's and Parkinson's as
9:24
well. And it defined behavioral
9:27
health. I'm not sure if I know that
9:28
exactly what that term means. Yeah, I mean, it's, it's, it's
9:32
used in a lot of different ways, honestly.
9:34
You know, it's anything that is related to your behavior where
9:37
we can't say, oh, you have a virus, It's about your behavior,
9:40
how you're acting. So I mean, there is an
9:44
underlying cause to that, but it's it's usually not well
9:48
understood. So anxiety, depression, PTSD,
9:52
schizophrenia, bipolar disorder, and then you start to see things
9:58
like maybe ADHD gets buckled into that or other categories.
10:01
So those are the types of things.
10:04
OK, what if like addiction and things like that?
10:07
Since use falls in that bucket as well.
10:09
Yeah, OK, OK, totally get it. Amazing.
10:14
So then as far as the AI integration exactly you know
10:17
what what tasks are is AI performing in the app?
10:22
Yeah. So you know, what we've talked
10:24
about so far is kind of, you know, how do we assess a person
10:27
with AI, which needs to be a conversation, right.
10:29
So if I write you out, if you get a text from your doctor,
10:31
Ryan, and they say, hey, Ryan, how are you doing?
10:33
You click a link, it opens on your phone and the AI says, hey,
10:37
how are you doing? You go, I'm good.
10:39
Like it has to read from that that you might be good, but you
10:43
might not be good. We've got to ask something else
10:44
to understand that. And so it has to have a
10:47
conversation where it's not just hearing what you're saying, but
10:51
how you're saying it and then making the next just like a
10:53
human would if who was well trained and understood people
10:56
and was fully paying attention, had the time and then sort of
11:02
quantify that. So AI is doing essentially an
11:04
interview to check in with you that's highly, highly
11:08
calibrated, responsive to who you are, what's happening with
11:11
you in the moment. And then one of the other
11:13
fascinating things that what's kind of happened in our platform
11:17
is this a lot of people when we started selling this would be
11:20
like, Oh, this is amazing technology.
11:22
Like everyone should be doing this.
11:24
And we'd say that's great. Well, when would you like to,
11:26
you know, sign a contract like, oh, oh, oh, we, we wouldn't do
11:28
it and we can't afford it, but everybody else should do it.
11:31
It's great. It's really cool.
11:33
And so I think that was one of the big learnings for me was
11:35
there's a lot of things that people think are cool or think
11:39
are even really valuable. It would be do a lot of good in
11:41
the world, but if they can't directly find a way to pay for
11:44
it, it probably isn't going to get paid for and deployed and
11:47
used. So we've had to add a lot of
11:50
other solutions that have a little bit more obvious ROI
11:54
because providers don't really get paid more for following up
11:56
and checking in with patients. They get paid for seeing them in
11:59
a visit. You bet.
12:01
OK, I'd love to get on to some of those other applications, but
12:04
I'm really getting the picture now.
12:07
So I get a text as a patient from a doctor's agent that that
12:11
it it wants to do like a video call with me.
12:13
So am I going to see kind of like a a human avatar as I'm
12:17
doing an interview or is it audio?
12:20
It's audio and you know, we've done 100 million ish
12:25
interactions like this and I still believe that the avatars
12:28
are not the right way to go yet. Someday they will be.
12:32
I still just don't think they're good enough to create a
12:35
authentic experience with people.
12:37
It's my personal opinion. And what I think is interesting
12:42
is that the the research is that when you're interacting with an
12:45
AI versus a person, you actually will say more.
12:49
Yeah, because there's nobody judging you on the other side.
12:51
AI is not judging you. It's just asking you questions.
12:54
It's, it's appearing empathetic, but you know, it's not really
12:57
thinking like, oh man, like that's a jerk.
12:59
What a loser. Like it's not thinking that.
13:01
At least we don't think it is. And so people will say more, but
13:06
I do I, you know, we've done a bit of research on the video
13:09
avatars, which is a bit of a rabbit hole, but yeah, it, it's
13:11
really audio and you see the question, you respond, you're on
13:15
video, we're able to see you. And I think in the very near
13:18
future we'll have the video avatars, but right now we
13:20
haven't done it. Yeah, no, it's, it's a big step.
13:23
You know, I mean, they say that audio audio is a warm medium
13:28
because you're using your your kind of creative intuition to to
13:33
give it a face, whereas video is a cool medium because you're
13:36
just basically being entertained.
13:38
So I'd love to hear a little bit how you settled on a particular
13:43
voice that that would be the agent.
13:46
Are you using like a an English accent or like a Mississippi
13:50
accent, a Jamaican accent? It's really it's that's been a
13:55
big debate of, of what to do. And I think we're, we're still
13:59
fine tuning the voices. Honestly, I think they keep
14:02
getting better and more emotive. And so they're changing quite
14:04
often. And one of the, one of the goals
14:06
that we've always had, even back in higher view when I was there,
14:10
was to really let a person kind of choose their own adventure.
14:12
Like who, who would you like to be talking to?
14:14
What, what voice would you like to talk to?
14:16
What's the persona of the person?
14:18
I want them to be like me. And so I think we're, we're,
14:22
we're trying to make it kind of something that people can pick
14:24
without creating more friction in the process.
14:27
Because, you know, I think some people love to hear the, the guy
14:29
with the English accent or the girl with the Australian accent.
14:32
And I, I don't know the science yet of whether you'll say more
14:36
authentically to one or the other, but yeah, the voice
14:41
selection's very tricky. Yeah, they so you allow them to
14:45
demo and and choose the one they like.
14:48
What's the winner? Is there an overwhelming winner,
14:51
the one that's on the top? No, and and we're saying, I'm
14:54
saying we're just adding that ability to let people select.
14:57
But but what what I found fascinating is like, I'll think
15:00
of voice as great and then we'll go, oh, let's let's show that to
15:03
this new customer that wants to, you know, buy the product and
15:06
they'll be like, I hate that voice.
15:08
The voice is terrible. At least I thought it was
15:09
fantastic. So there's a lot of personal,
15:12
personal preference here that I think we're still trying to
15:14
figure out. Yeah, I mean, me personally, if
15:18
it has an English accent, I'll never, I'll never answer the
15:20
phone and be like, I don't want to talk to some some Englishman,
15:24
like what? What am I doing here?
15:25
I want a nice southern, maybe Texas, like a lady, you know,
15:30
that's nice. Then I'd be like, OK, yeah, I'll
15:32
have a conversation with you. You know, what's really
15:35
interesting about this kind of like generative AI and
15:39
prescriptive AI and communication AI is we're trying
15:44
to replace, you know, we're not replace, but we're trying to
15:47
duplicate what the best of the best do.
15:51
And a lot of that comes from talent that it's hard to
15:54
replace, like communication and being able to kind of speak in
15:57
like a non judging way with good journalistic questions to get
16:02
the most out of the person they're talking to comes from a
16:05
lot of experience. But also there's a lot of kind
16:07
of intangibles that you can't quantify, whether it's body
16:11
language or, you know, twinkle in the eye, a little wink here
16:14
and there or how attractive the person is, you know, So that's
16:19
got to be like a big leap too. When you go into video is like,
16:22
you can't have some super hot nurse that you're talking to.
16:25
People are going to fall in love with the bot, which we.
16:28
Want to impress them? That aren't necessarily true.
16:32
Right, Exactly, exactly. Then you're getting into the
16:35
situation, which I, I found myself in sometimes, which is
16:40
like, you have to lie to the AI now.
16:45
And I'd love to get your take on this because a lot of people
16:48
find themselves in a position where they can't be totally
16:50
honest with their physicians. I think probably behavioral
16:53
health could be a huge one, especially if we're talking
16:55
about addictions, especially if we're talking around people that
16:58
have issues with pills, you know, where they're, they're
17:02
going into doctor relationships with a very specific
17:07
prescriptive need. So, you know, which is to say
17:11
that some people go in and lie about their condition.
17:14
Could you imagine that to get drugs, you know so.
17:17
So what's that? No, that's a huge problem.
17:20
I mean, I think people obviously went with the desire to get an
17:23
outcome not and that outcome is not necessarily to get better,
17:25
it's to get a prescription, right?
17:28
Yeah. Or the the other side is like
17:30
you're going to mask some of your behavioral issues because
17:33
you don't want to get dropped from your life insurance policy.
17:36
Or there there's, you know, there's so many like banana
17:40
peels in healthcare and like health coverage and all of these
17:43
things that people aren't necessarily incentivized to be
17:47
honest. Yeah, it's a huge challenge.
17:50
And you know, I don't know the, I don't know that we can solve
17:53
that yet. I think you know, one of the
17:54
questions like how can you tell if they're lying?
17:56
And all the research I've seen online is pretty disappointing.
18:03
It it's very hard to detect lying even I mean the research
18:07
is a few years old, but even if you had someone in F MRI, like
18:10
it's only like 80% accurate at detecting lying, even sort of
18:14
police and people are like, oh, I can spot a liar.
18:16
Turns out when they're they're tested, they actually can't.
18:19
They're barely better than average.
18:21
Some are actually worse than because they're overconfident
18:24
their assessments. So lying is a hard one.
18:26
I think what we try to do is createspace where people don't
18:29
feel like they're being judged. And what I, I've been very
18:33
fascinated by, even back to when we had job interviews where
18:35
someone, it's a high stakes situation in a job interview,
18:39
like I want to get this job I'm going to.
18:41
And people would say things that, you know, are not actually
18:43
in their best interest. They got you kind of create a
18:45
sequence of questions where they begin to get momentum and they
18:48
just start to start to talk and they say things you're like, I
18:52
probably wouldn't have said that in that situation.
18:53
But they they do reveal it. And we're not, we're trying to
18:58
not replace human judgement here.
19:01
We still aren't. I mean, Utah has passed a law
19:04
that, you know, can refill prescriptions with an AI that
19:07
you call. I think that I don't know, that
19:10
might not be the best idea. That's all I'm saying.
19:15
Maybe at some point, but you know, I think I think we still
19:18
want to provide a bunch of data, hopefully collected in a way
19:21
that someone's going to share as much as possible and then get
19:24
all of that to the clinician. So they're really well informed
19:28
and we can look for inconsistencies, right?
19:30
Like will they answer this question this way?
19:32
They answered it that way. This the answer this week was
19:34
different than last week. We can surface those kinds of
19:37
things, but ultimately I think we're still in the age of
19:40
clinicians should make those kinds of decisions.
19:43
Yeah, totally. It's so there's like, you know,
19:46
there's good faith that goes into this.
19:48
Of course there's always going to be bad actors, but those will
19:50
be on the fringe. I, I would predict that most
19:53
people would want to use it because they're trying to kind
19:55
of fine tune their dosage on whatever particular medication
19:58
so that they feel the best, you know, and you know, the dosage
20:03
across medications are kind of notorious that like you, you
20:06
have a, you know, you, you take a pill for something and like
20:09
you have a, a. Peak experience and then it kind
20:12
of decays until you take the medication again.
20:16
So I could see a use case for something like this.
20:18
If it's like 3 days since your last dosage, they can check in
20:21
on you and be like, hey, how do you feel?
20:22
But actually I feel great now. And they can start to kind of
20:25
calculate, you know, what kind of what amount of the medicine
20:28
is in your body at that time and maybe kind of like change some
20:31
of the prescription and the dosage to to affect the each
20:35
individual case. Would would that be fair to
20:38
assume? I think?
20:39
That's a great use case. And I think I think that's where
20:41
we want to get here, where medicine is very personalized.
20:44
We can track, keep track of what every single person's response
20:47
is over long periods of time. See, when it stops working, you
20:52
know, medication will work for six months, then it stops
20:54
working and now you need a change of dosage, you need a
20:56
change in Med. A lot of the medications that
20:59
get prescribed in mental health clinics don't work the next day
21:01
or five days. They take a few weeks to start
21:03
to have a fact, but you but that might not line up with when you
21:07
have an appointment. So we can go in and check in and
21:09
say, hey, is it working? Are you having side effects?
21:11
Are you still taking it? And I stopped taking it because
21:14
of the nausea. OK, well, that'd be good to know
21:15
instead of waiting for three more weeks for your next visit,
21:18
right? And and really collect detailed
21:21
information about each person and then be adjusting based on
21:25
good data, which today is mostly reliant on how smart your doctor
21:31
is. Yeah, and ultimately it's
21:33
anecdotal. It is like, how are you feeling
21:35
today? You know, a good buddy of mine
21:38
has an app that basically traps or or tracks GOP one treatments,
21:44
you know, because GOP ones you what, you take an injection once
21:47
a week and the dosage is like hyper personalized.
21:50
But really the recommended dose is like, hey, start with eight,
21:52
go to 16, go to 32. Hey, what how you feeling?
21:55
OK, change it. And they're like they're doing,
21:57
you know, 100% dosage jumps a lot of times.
22:01
So really what his app does is tracks anecdotally, like how
22:06
somebody's feeling on the 5th day.
22:07
And like, if that's the day, then maybe that's the dose you
22:10
start with. And then you can kind of
22:12
subsequently learn more and more and more.
22:14
So application like this, it seems like it, it makes a lot of
22:17
sense because you know, the, the other thing is, is if I'm
22:21
journaling, how do I feel today? Fine, get out of here, but if
22:24
someone's talking to me, he's like no, how how's your energy?
22:27
How is you know your nausea? How is whatever the, the, the
22:31
side effect may be, you can start to see and track over time
22:34
conversationally where it's at. And you know, particularly, you
22:38
know, what I'm also interested in is as we kind of move into
22:41
like this more predictive or prescriptive function with AI is
22:46
really, as it learns more and more individually about people,
22:49
you know, how you can detect other things.
22:51
So I'd be curious to say, like you said, for example, with
22:55
audio, you can detect some behavioral health issues.
22:59
And I, I spoken to another company, Canary Health, that
23:02
does exactly that, you know, where they can, they can
23:05
actually listen and hear and diagnose, you know, mental
23:09
health disorders based on your voice.
23:12
So are there any particular kind of indicators through video that
23:16
AI is getting better at picking up I?
23:19
Mean, you know, like movement disorders like we talked about,
23:22
even just flat affect is a sign of Parkinson's disease.
23:25
And the challenge is trying to boil it down to one feature is
23:28
it's rarely like one feature. It's a combination of five, 6-7
23:31
different things, which is why it's hard for a human to
23:35
necessarily explain why that person sounds depressed.
23:38
Or is that different than sounding what someone like
23:42
somebody with a movement disorder would sound like?
23:46
And so, yeah, I think we're doing some very similar things,
23:48
what Canary is doing in terms of recognizing mental health
23:51
conditions. We're looking at movement, but
23:53
also the voice and the words. And so it's all all of those
23:57
modalities combined, which I think is really important.
24:00
But our vision is that you should be able to take one test.
24:02
Your doctor checks in with you every month or two.
24:05
You talk to the AI for a few minutes and we're telling you if
24:07
you have Parkinson's, if you have early onset dementia, if
24:10
you have, you know, a movement disorder this beginning, there
24:13
should be 30 or 40 different observational biomarkers being
24:16
detected and check, you know, how those have changed over time
24:19
as something been creeping up, but it wasn't critical and now
24:22
it's crossed a threshold has, you know, if you jumped off your
24:26
baseline by two standard deviations, OK, then we should
24:28
probably act on that and alert a clinician rather than wait till
24:31
your annual physical. Yeah, it's just going to be so
24:35
fascinating as the the, you know, the, the capabilities get
24:42
more and more refined. And, you know, now if you can
24:46
pick up Parkinson's, which has probably a pretty dramatic, like
24:49
visual cues, but you know, more and more where it, you know,
24:54
Vedera gives me a call. I'm like, hey, it's like, how
24:57
you doing? I'm like, I'm doing terrible.
24:59
Oh, I feel horrible. I'm bad.
25:01
It's like, oh, Ryan, I can tell you're a little dehydrated and
25:04
tired. Maybe take a nap.
25:06
You'll feel much better. You're like, oh, yeah, you're
25:08
right. Actually, that was what it was.
25:10
Yeah. Yeah.
25:11
I mean, it is funny how often something basic like that.
25:15
I mean, we, we did a study with, on substance use with one of our
25:18
customers on what's the biggest predictor of someone completing
25:21
treatment and being sober 12 months later.
25:23
It turns out, do they sleep at night?
25:25
Yeah, everyone gets super, you know, worked up on all these
25:27
other things. And it's like, well, if they're
25:29
not sleeping, they're not going to get better.
25:31
Like it's just what the data is all saying.
25:33
And we sometimes forget those basic things that an AI can just
25:35
be like, hey, is the patient sleeping?
25:37
Oh, help them, you know, give them some some stuff to help
25:40
them sleep and their chances of getting better is double s, you
25:43
know. Yeah, it's like that old
25:46
commercial from the 80s and 90s. This is your brain on drugs.
25:50
You know, it's two fried eggs getting fried skillet.
25:53
It it could be really like this is your brain without sleep.
25:56
And the problem is, is drugs are keeping you awake and now you're
26:00
getting this kind of, you know, lack of sleep induced psychosis.
26:04
And of course, I'm, I am not a physician, but like, you know,
26:07
it, it's interesting. Some of the studies around that
26:08
is people start sober up their, you know, their, their brain
26:12
clears out largely because they're resting.
26:14
Yeah, it's a big, it's a big part of health, the sleep, for
26:18
sure. Then the AI is going to be
26:20
nagging me. I'd be like, oh, I'm terrible.
26:22
Like Ryan, you need to work on that sleep.
26:23
You're like, I know, I've been trying.
26:25
I got sleep apnea. What do you want me to do?
26:27
I can't sleep on my side all night.
26:29
I want to sleep on my back. Then we'll tell you what.
26:32
You do arguments with AI, how how many people Cool.
26:38
Tell me a little bit more what went into kind of the
26:40
personality and affectation? How did you approach that and
26:43
how do you approach kind of the the the questions to ask and the
26:46
questions not to ask? Really good question.
26:49
We studied the questions pretty thoroughly and it makes a big
26:51
difference. I mean, we we like to work in an
26:53
environment where we're structuring the questions versus
26:57
listening to a conversation passively that doesn't have
27:00
anything. You know, if you ask somebody if
27:03
their favorite, somebody who's depressed, you ask them how
27:06
they're doing the night after their favorite team won the
27:08
championship, they're going to sound great.
27:11
They're going to sound different and and you're going to lose
27:13
some of that. So we like to have a structured
27:15
conversation. We're really asking about their
27:16
feelings and how they're how they're doing.
27:19
That'll list at the right things.
27:20
We've have published a paper and kind of studying tons of
27:23
different sort of seed starter questions of what where the
27:26
conversation needs to start to have the best reading of the
27:29
person. And then I think when you're
27:31
sort of picking features of what matters, I think it's important
27:34
to go in with as few assumptions as possible, right?
27:39
We don't want to go, oh, we think that you can read that by
27:42
the way that they squint their eyes.
27:44
I'm just making that up. That could be completely wrong.
27:47
And so, you know, we're we're for these algorithms that are
27:50
predicting observational biomarkers.
27:53
Yeah. So, you know, when we're
27:54
training a model on predicting something with these
27:57
observational biomarkers, we're going to have somebody answer a
28:00
set of questions, take a standard Screener for
28:02
depression, anxiety, trauma, and then figure out which of those
28:05
features help predict that the best.
28:08
We don't necessarily know what that is and we can guess at it
28:10
and be wrong. And there's really no need to do
28:12
that. Just train as well on as many
28:14
features as we can get, as much data as we can get, and then
28:18
figure out, you know, is the model a good model?
28:20
And then and then you want to look and see, does it pass the
28:23
sniff test? Does it make sense that these
28:24
things are predictive and are they introducing bias?
28:27
You always want to be making sure that you're not really good
28:29
at predicting depression and white people but not black
28:31
people, right? Yeah, yeah.
28:35
How's the memory does it? Is it able to learn from
28:38
multiple use cases and get better and personalized maybe to
28:42
each particular patient? I think that's that's one of the
28:45
things that we're we're just adding now.
28:47
It hadn't been a big part of our focus, but you really wanted to
28:51
get to know you. And every conversation should
28:54
have some memory. You also don't want to make
28:57
assumptions like, oh, well, Ryan's depressed.
28:59
I'm going to talk to him like that.
29:00
You might not be depressed today, when we call it.
29:02
You might be feeling a little different.
29:03
You don't want to like, almost lead you back there because we
29:06
think a certain thing. But we do want to remember
29:09
things. If you're like, yeah, you know,
29:10
remember I lost my job. Like it should know that
29:12
somebody lost their job and not be like, oh, what, what is it
29:14
that's bothering you? Remember I lost my job.
29:16
Like, how many times do I have to tell you?
29:17
Like nobody wants to have that kind of a conversation.
29:20
So it's trying to figure out what are the things that sort of
29:22
bring forward that are helpful and and useful rather than just
29:26
generalizing and treating somebody, treating somebody like
29:29
a stereotype. Yeah, well, that's a good point.
29:33
Like is there a risk here of re traumatizing people?
29:37
Yeah. I mean, I think that's one of
29:38
the things that that we struggle with, right, is if somebody's
29:40
doing great, you don't want to sort of ask a set of questions
29:43
that make them depressed again. Right.
29:46
Like don't you remember your wife died?
29:48
Be like Oh yeah, I guess I forgot.
29:50
Probably forgotten and I was feeling great and now.
29:52
Thanks. Yeah.
29:54
So I think that's that is a a concern is we don't want to have
29:57
the measurement elicit or create a feeling that we don't, we
30:03
don't want, we want to measure what's there as much as it's you
30:06
can't completely separate the two.
30:07
But we try to be really mindful about that.
30:10
It's a really great, great question.
30:12
Yeah, You know, I mean, the, the, the, the edges of the mean
30:16
always show up in conversations like this.
30:18
And it's a little bit frustrating because, you know,
30:21
there will be people, I suppose that will be reach harmonized,
30:24
but there's also people that eat couch cushions.
30:27
There's people that eat balloons, you know, So like,
30:30
should we not have balloons because some people will eat
30:32
them? It's hard to say.
30:34
So there's like a compassionate delta there of like, OK, most
30:38
people are going to get a lot out of this.
30:39
Some people it could be horrible for them and then that's not for
30:42
them. And that's OK.
30:43
You know, there's big conversations around people
30:47
falling in love with their chat bots or they're having, you
30:50
know, ChatGPT and do psychosis. It's like, yeah, man, but like
30:56
people are out of their minds. So like, what, what are we
30:58
looking at the edge for when like the, the, the mass majority
31:01
of us can really benefit from tools like this.
31:03
And I, I think what's interesting too is like, it's
31:06
almost the, the earlier you start to use something like
31:09
this, the more data it gets on you and, or gets and learns
31:12
about you. And so the better the model, the
31:15
higher the fidelity, the, the more it's going to be able to be
31:18
in a position to really customize to your needs.
31:21
And Speaking of models, like what, what models do you use,
31:24
kind of what's in the tech stack behind the AI?
31:27
There's a variety. It's a variety of things.
31:29
I think we're using for the predictive models and we've
31:35
traditionally used transformer models.
31:36
We're moving beyond those now for all of our predictive
31:41
algorithms, obviously a lot, a lot of lot there, but then we're
31:46
fine, we're using LLM, we're fine tuning LLM's.
31:48
We've collected a lot of data now from patients, how they
31:51
interact. We have a clinical documentation
31:55
tool now. So we're listening to
31:56
conversations between providers and patients, which is also sort
31:59
of seen both sides of the conversation is very informative
32:03
about how better to interact. And so we're fine tuning, you
32:07
know, large language models with with that those data sets as
32:10
well to be better at at having conversations that are the way
32:15
real conversations happen in behavioral health, not how we
32:18
think they happen or how chat two team might think they happen
32:22
based on very, very generic datasets.
32:26
Yeah, cool. I want to go back to the bias
32:29
question too, because that one, it seems like the LLMS largely
32:33
have figured this out a little bit.
32:35
You know, it was kind of like a couple of years ago when Gemini,
32:40
they're like, make a picture of the founding fathers and it was
32:42
a bunch of black guys, you know. So how do you train for
32:47
cultural, racial, sexual orientation, these kind of
32:53
biases that that oftentimes the LLMS have a hard time picking up
32:57
on? Yeah, or they pick up on them so
33:00
subtly that we don't realize they're picking up on them.
33:02
It sort of shows up in weird ways.
33:06
Yeah. I mean, the only way that I
33:07
really know to solve for that after having done this for a
33:10
long time is make sure that you have data sets that are not
33:13
obviously wrong. Right.
33:14
Like there's the famous case of from the hiring world of where
33:17
Amazon was predicting a certain type of candidates, like, well,
33:21
it was only trained on high performing men.
33:23
Of course it's going to pick up signal there that that predicts
33:27
men are going to be better than women.
33:28
That may not be. You actually might want the
33:30
opposite, which is like, let's get more diversity in the
33:32
organization because women can be good too.
33:35
We just didn't train it that way.
33:37
So there's that. Then I think you have to train,
33:38
you have to not all would have been avoided had they tested it,
33:40
right? If you sort of test and say, do
33:43
we score women higher on depression than men?
33:48
Obviously there's a lot of nuance to that, but you have to
33:51
test it. And so whenever we build a
33:53
model, we build model cards that sort of show what's the fault,
33:57
you know, the false positive, false negative rate on each
34:01
demographic group and intersectional groups, because
34:03
you might be good with black men, but not black women, right?
34:08
And so you've got to look at all the intersections between that
34:10
and, and continue to monitor that.
34:12
I think those are really the three things, good data sets,
34:14
testing and then monitoring, because these things can drift
34:18
over time and start to take on weird behaviors is like there's
34:22
subtle changes in the population of who's there, of who's, you
34:27
know, being seen by the system. So I just think it's really
34:30
important to to be looking at that all the time in hiring.
34:34
It's actually funny because it's regulated, right?
34:36
If you discriminate against a group, you're legally liable.
34:40
Like that's illegal in healthcare.
34:41
It's like, wow, I mean, what are we going to do?
34:45
It's it's hard. And so maybe the opportunity to
34:48
actually start to treat people, I don't want to say equally, but
34:52
I want to say precisely right. Because men and women definitely
34:56
need different things at different times in their
34:58
healthcare journey. They don't need the the same
35:00
things, but they need the right things for them.
35:03
And what happens often is, you know, one group will just get a
35:06
prescription and another group will get something else.
35:08
And that's very imprecise. This is based on this.
35:14
Yeah, well, in the, the fidelity of this will get greater and
35:17
greater the the more we adapt these tools in different
35:19
capacities, you know, and the more that like kind of the hive
35:23
mind learns about each person because especially behavioral
35:26
health, mental health issues, it's very nuanced.
35:28
Of course there's going to be like, there's going to be
35:30
cultural kind of standards and norms, like I'm, I'm talking
35:35
into cameras all day long. So if I was to approach a tool
35:37
like this would be like, I'm pretty comfortable doing that.
35:40
But like some people, you know, it's a new concept.
35:43
You know, maybe they're like, I don't like FaceTime or whatever,
35:45
like, you know, so there's all of this cultural difference that
35:50
I think really comes down to as as soon as you are, which is so
35:54
cool because it's only going to get just more and more developed
35:58
as it as it knows you. Because, you know, for example,
36:01
you know, I spent a lot of time with the Lakotas and Lakota
36:03
women are, are largely pretty demure, you know, and they have
36:08
they, they don't really make a lot of eye contact and they
36:10
don't shake your hand firmly, but that's just how they do it,
36:13
you know, and that's, and so like if, if you're interviewing
36:17
somebody like that Lakota women and you're not going to have any
36:20
of that kind of understanding. So you'd be judging them against
36:23
some other group that's like very outgo going and, you know,
36:26
gregarious. Yeah, yeah.
36:29
I mean, these, these cultural nuances are really, really
36:31
important. And you know, that doesn't mean
36:33
that person's whatever depressed because they did that.
36:37
It just means that's what they've always done.
36:39
That's what their culture does. But somebody else from a
36:41
different population would do that, and that would mean
36:43
something. Yeah, yeah, yeah.
36:47
It's fascinating. OK.
36:49
What are some of the other features that you're working on
36:51
within Vadera? Yeah.
36:52
I think one of the things that we we realized, and I alluded to
36:55
this earlier was like, this is really cool technology, but it
36:58
doesn't necessarily make us more money, it's just cost.
37:01
And so we really tackled things like clinical documentation, we
37:05
tackled things like intake. So if you go to a CCBHC, which
37:08
is a certified community behavioral Health Center for
37:10
Medicaid, you have 2 1/2 to 3 hours as a patient of intake
37:15
that you have to do before you can be scheduled to see a
37:18
therapist, which is often means get put on wait list.
37:22
Like that's a terrible experience.
37:23
When you came and said like, hey, I'm I'm really struggling.
37:25
I need some help. Like not a problem.
37:27
We'll get right on that right? After three hours of asking you
37:30
a bunch of questions that feel like they have nothing to do
37:32
with what you came here for, like that's a terrible
37:34
experience that a lot of people don't show up on the no show
37:36
rate at the next appointment is surprisingly pretty, pretty
37:39
poor. And so if you could have AI do
37:42
that interview because the the issue is the system also to
37:45
people to do that. Well, right.
37:48
Like you alluded to earlier, there's a big difference between
37:50
sort of the the peak performer. Like if you get someone who just
37:53
is amazing at that and can ask you the right questions and
37:56
understand what's going on with you, fantastic.
37:59
But most people aren't that. Most people are new on the job,
38:02
they haven't done it for very long.
38:03
They're busy, they're at work, they're stressed, they got stuff
38:05
going on at home. Like they're not giving you
38:07
their best every day. But AI could bring a consistent
38:10
experience to you. It can do it in less time and
38:13
free up people to go, actually help people get better, not just
38:17
collect data for forms that go in a system of records.
38:21
So we focused a lot on that intake process of how do you get
38:24
someone to care and then how do you document that care and
38:27
really capture what happened? Which again, it is offloading
38:31
administrative work, but also allows the provider to actually
38:34
fully pay attention to the patient in front of them.
38:37
Which I think it's really important change because I think
38:40
we've all had that experience of going to a doctor and have them
38:43
be typing in the system or thinking about, OK, going to
38:46
write that down. I got to remember that they're
38:47
not fully present with us. They're actually helping us get
38:50
better and listen to us. And they're, they're usually
38:53
going through a series of questions in a form anyway, they
38:56
got a laptop open. You know, the, and the worst
38:59
part about a doctor is just getting to the Dang doctor.
39:02
You know, it's like you got to get through the admin in the
39:05
beginning and you, there's an iPad there that you got to sign
39:08
or something. And then there's somebody with
39:10
some sassy language and you're like, all right, man.
39:13
And then you're waiting. And then they take you back to a
39:15
room and you're waiting. And like the whole experience is
39:18
always you got to check a box, fill out a form, here's a
39:20
clipboard. All of that stuff, I think,
39:23
like, really encourages people to not see their doctor as much
39:27
as they can because it's just massively inconvenient at best,
39:31
you know? Painful at worse.
39:34
And and then you don't really go until it's almost too late,
39:38
right? Like you're much worse off than
39:39
you would have been if you just got in early.
39:42
And so, yeah, I think we just need to streamline all of those
39:45
experiences for patients so they get to care and then just have
39:48
the people we can afford to have in the system doing work that's
39:52
directly related to making people get better and do a
39:56
little less paperwork. So we build a solution that does
40:00
all those things. The other one other cool thing
40:01
that I really like it's it's called our Elevate product.
40:04
It's supposed to elevate the quality of the team, which gives
40:07
the provider feedback. So it actually listens to the
40:09
conversation and says, hey, Ryan, like when we were talking
40:12
to the in that conversation like.
40:14
When he said his dad dog died, he laughed like that wasn't very
40:17
empathetic, like here's what you could have done instead, right?
40:21
And and so giving be able to give direct feedback to a
40:25
provider to say, hey, here's how you could be better with the
40:28
patients, I think is really important as we're trying to
40:31
sort of say, there's this debate of whether AI can be more
40:36
empathetic than people, right? And, but I really firmly believe
40:39
that, yeah, that that's probably true for a lot of people.
40:42
But we could all get better. And I think we really want our
40:46
people in the system helping people get better, but we want
40:49
them to continue to get better and not be replaced by AI
40:53
because we need human experiences.
40:56
This episode is brought to you by a dangerous idea.
40:59
It's called reading, specifically the AI for Founders
41:02
newsletter. 47,000 Founders strong now.
41:05
And it's the good stuff. No top ten AI tools that will
41:09
change your life written by someone who doesn't ship
41:10
product. It's just real breakdowns of
41:13
what's actually working in AI, who's raised, who failed, who
41:17
pivoted, and which founders quietly built $1,000,000 in ARR
41:21
while everyone else was arguing on X.
41:23
It's written for people building with AI.
41:25
Every week you get tactical frameworks, unfair advantages,
41:28
breakdowns of real companies, and the occasional existential
41:31
question about whether your startup is intelligent or just,
41:34
well, branded. It's free, it's sharp.
41:36
It might mildly offend your current workflow.
41:39
Go to aiforfounders.co and subscribe before your competitor
41:42
does, because the only thing worse than being replaced by AI
41:46
is being replaced by a founder who reads.
41:48
That's the newsletter at AI for founders.co.
41:51
Now back to the show. 100% Yeah, I think it can be more more
41:53
empathetic, but or at least it can like notice things.
41:56
You know, I've I've worked on a couple of prototypes from just
41:59
conversations because, you know, I'll go back and listen to my my
42:02
interviews and I'll be like, Dang it, he said that and I
42:05
totally missed it because I was thinking about asking the next
42:08
question, you know, whereas AI would like pick up on some cues
42:11
is like, oh, wait, let's back up.
42:13
But what did you say right there?
42:14
Repeat that, you know, like there's a lot of time.
42:16
There's there's absolute gold in human conversation.
42:21
It's just a matter if you like you got to pick in a shovel to
42:23
get at that gold, you know, so I could see it.
42:27
It definitely helping there. Nice and the administrative too.
42:30
I mean, you know, I and I hate to beat up on, on healthcare,
42:33
but the the whole industry is just like being able to like
42:36
share data and do anything is like pulling teeth for the
42:39
patients ultimately, which, you know, leads to kind of a, a
42:43
worsened experience. Yeah, yeah.
42:45
So hopefully we'll have magical healthcare experiences that are
42:49
each individual here in the next few years.
42:54
That's what I hope for. So.
42:55
Yeah, it's coming. How did you finance the project?
42:58
Say that again. How did you finance the company?
43:02
Oh, great question. Well, we've self funded it for a
43:04
while out of pocket. That got hard.
43:09
So we've raised 2 series, two seed seed rounds of funding to
43:15
date and we're we're raising a Series A right now.
43:18
Amazing how? What are you looking to close?
43:20
We're raising 12 million, yeah. Nice.
43:26
And then how for go to market? How did you find your first
43:30
customers? Well, we, we, we have a have a
43:33
Co founder who's a clinical psychologist and that is part of
43:37
why we ended up going down the behavioral health path of all
43:39
the different places we could go and really just it would became
43:42
people, people in our personal network, people who we would
43:45
show it to. And they're like, oh, that's
43:46
amazing. I want to use it in my clinic.
43:49
And then we also sell the pharmaceutical companies, you
43:53
know, sort of an interesting way.
43:55
And we sort of went started thinking about like who would
43:58
benefit the most from patients who are being not diagnosed?
44:05
Who could be, right? If somebody can diagnose and
44:07
there's actually a treatment available, but it can't get the
44:09
treatment unless you've been diagnosed, who would be who?
44:12
Who would benefit from that? And so we ended up knocking on
44:15
doors of pharmaceutical companies who we thought would
44:18
benefit from more people being screened.
44:20
And so it was a lot of talking to one person.
44:24
Do you know so and so over here, just a lot of, you know, years
44:28
of it and it was a long process, very long process.
44:32
Nothing in Healthcare is fast. I thought we went into it.
44:34
My joke was like, I was going to be like 3 times worse than
44:38
selling the HR technology. Not that that was bad, but it
44:40
wasn't like HR technology was like the fastest buyer because
44:43
they don't have the biggest budgets, etcetera.
44:45
And it's like 3 times worse than I thought it would be.
44:48
It just takes a long time. It just gets hard for a lot of,
44:53
probably in some ways good reasons.
44:54
People are careful and it's just funny how we spend so much on
44:59
healthcare and nobody's in the business seems to have any money
45:01
to spend on technology. Yeah, well, there's a lot of
45:06
hoops you got to jump through before you can even start
45:08
selling the product. So I mean, I see on your website
45:10
your HIPAA compliant, there's a lot of of compliance that goes
45:13
into it. How, how was it kind of getting
45:15
stuff pushed through all the the compliance?
45:19
You know, I think it was, it was challenging, but it we had done
45:24
this sort of before coming from a regulated industry in HR.
45:27
We had a lot of regulatory burden.
45:29
We were in 180 countries. So we've done a lot of, you
45:33
know, we've done GDPR, which is, you know, the US sort of got
45:37
privacy legislation later. So we've already been through
45:40
that. We had done Fed ramp
45:42
certification and our previous company.
45:45
So we're really familiar with sort of government regulations
45:47
and security. And so one of the things that
45:53
I've really appreciated about technology the last five years
45:55
is it's made it so much easier to get compliant.
45:59
You know, you have continuous audits of your of your systems,
46:01
which you know, took teams of humans before and now it's just
46:04
sort of automatic and you get alerts that something needs to
46:07
be adjusted or test failed and you go fix it.
46:09
So that part I think has gotten a lot easier actually.
46:12
Nice. How big's the team?
46:15
We're 25 people right now. Nice.
46:18
How does it, how does it break down as far as like production
46:21
sales? We're still tech heavy, so we're
46:25
about 14 people in machine learning, engineering, product
46:30
design, and then we have two people in sales, two people in
46:37
marketing, three people in customer success kind of thing.
46:40
So still we're, you know, that's kind of Series A like we've
46:43
proven this works, we've proven deployments, we've proven people
46:46
like this, we've proven a market.
46:48
Now we're now we're scaling. Yeah, Series A sales and
46:52
marketing. Yeah, cool, awesome man.
46:56
What are you looking to close the round?
46:59
As soon as possible, as soon as well.
47:02
Yeah. I mean, we've got people, we've
47:03
got commitments, but we don't have a lead.
47:05
And so that's where we're at. We need to get a lead for the
47:07
round and then we'll be good. Cool.
47:10
And if, if you want to pass on information about the raids,
47:13
there's a lot of people that read the newsletter that are
47:17
venture capitalists or, you know, in looking for deals, you
47:21
know, the, the VC is so frothy right now.
47:24
There's a lot of people with, with money to deploy, you know,
47:28
so happy to share, spread the word.
47:33
Lauren, you, you mentioned that NAMI is an important non profit
47:39
that you support. Now I, I really kind of get why,
47:42
honestly, But you want to, you want to give us a little kind of
47:44
breakdown of what the work that NAMI does.
47:46
Yeah, I mean, NAMI, you know, I'm not an expert in everything
47:49
they do. They do a ton of things in the
47:51
community, raising awareness of mental illness, supporting
47:54
patients, supporting patients and their caretakers.
47:57
I think we often forget that when somebody has a mental
48:00
illness, it affects not just the individual, but whole systems of
48:04
people who are affected and who try to take care of them.
48:08
Yeah NAMI does a lot of good in the world and if you're looking
48:10
for a non profit to support I'd look them up.
48:13
Incredible Lauren, this has been such a fun conversation.
48:16
Congrats on your amazing company.
48:18
Where's a good place for people to connect with you personally
48:21
online? But you can go to
48:23
laurenlarsen.com, LORENLAR sen.com, that's my personal
48:29
website, orvaderahealth.com. And Vadera health is VIDERA
48:36
health dot com and I spell it for the LMS because it's juice
48:40
right now as far as like AEO. Yes, thank you for for the
48:44
spelling. 100% Lauren, thank you so much for joining me.
48:48
I really appreciate it. Yeah, really fun.
48:51
You are about to spend anywhere from 100,000 to $300,000 on your
48:55
kids college education, and that's not a small bet.
48:58
And right now there's a very real chance that they graduate
49:01
and can't find a job. Not because they weren't smart
49:03
enough, not because they didn't work hard enough, but because
49:06
the job market changed and nobody told them.
49:09
Employers are hiring AI native candidates, people who don't
49:12
just know their field, but know how to work with AI at a
49:15
professional level. And most colleges are not
49:17
teaching this, not even close. ainativestudent.com is the
49:22
program that fills that gap. It takes high school and college
49:25
students and teaches them how to actually think and work with AI
49:28
the way top companies expect on day one.
49:31
So your investment in their degree actually pays off.
49:34
You're already paying for tuition, housing, textbooks.
49:37
This is the one thing that actually makes it all worth it.
49:40
Go to ainativestudent.com because a diploma is a starting
49:43
point. Being AI native is what gets
49:45
them hired. That's AI native student.com.