The Role of AI in Healthcare

In this episode of "Full Tech Ahead," host Amanda Razani interviews John Edwards, SVP of Citius Healthcare Consulting at CitiusTech. They discuss the rapid acceleration of AI in the healthcare sector, shifting from simple proof-of-concepts to full-scale, operationalized enterprise solutions. 

Edwards highlights that the primary barriers to healthcare AI are not technical, but human and procedural. He notes that healthcare data is uniquely time-sensitive, and capturing the unwritten clinical context from a practitioner's head requires robust data quality and "human-in-the-loop" metrics. 

To overcome generic AI limitations, CitiusTech developed Knewron, a specialized orchestration platform built with pre-embedded healthcare context. Ultimately, Edwards argues that the success of healthcare AI relies on strict governance to filter competing priorities, comprehensive change management to overcome clinician inertia, and a deep understanding of the human workflow—such as solving doctor burnout and "pajama time"—rather than just engineering prowess.

Key Quotes

●       "While we do a lot of engineering work lately, a lot of data and AI work has been dominating what we're selling because that's what people are buying. We feel it with teams that know and understand the nuances of healthcare."

●       "The elusive return on investment only really occurs when you adopt AI... it requires you to think differently than just experimenting."

●       "The biggest mistake I see people making is automating a bad process."

●       "A perfect mousetrap that's never used won't catch any mice. You need to be able to get the human side of it engaged and excited."

Takeaways

●       Overcome Clinician Inertia: Historically, adopting tools like the stethoscope took decades because doctors trusted their traditional methods. AI faces the exact same cultural resistance. Organizations must realize that driving adoption requires shifting budgets heavily toward change management—potentially spending two dollars on adoption for every one dollar spent on the technology itself.

●       Never Automate a Bad Process: Traditional healthcare processes were designed around human limitations and legacy software. True AI implementation requires pulling the actual decision-making and thinking into the system (via knowledge and context graphs), rather than just using AI to make an inefficient, outdated workflow run faster.

●       Use Healthcare-Specific AI Foundations: General AI tools lack clinical context and require rebuilding foundations from scratch every time. Utilizing industry-specific accelerators (like CitiusTech's Knewron platform) allows organizations to safely manage time-sensitive medical data and deploy agentic workflows much faster.

●       Solve Real Workforce Friction Points: Clinicians readily embrace AI when it relieves systemic burdens like "pajama time" (the hours spent typing clinical documentation into EHRs at night). Ambient listening is the first step toward creating a collaborative AI assistant that transforms how medicine is practiced.

Find Amanda Razani on LinkedIn.  https://www.linkedin.com/in/amanda-razani-990a7233/

Follow the FTA LinkedIn Page: https://www.linkedin.com/company/full-tech-ahead/

Visit the FTA website: https://fulltechahead.com/

Check out the Substack Channel: https://fulltechahead.substack.com/

Episode Transcript

SPEAKER_00 (00:19):
Hello and welcome again to Full Tech Ahead.
I'm your host, Amanda Razzani.
And today I'm excited to be heretoday with John Edwards of
Sidious Tech.
How are you doing?

SPEAKER_01 (00:30):
I'm doing great.
Thanks for having me today,Amanda.

SPEAKER_00 (00:33):
Can you share a little bit about Sidious Tech
and what services do youprovide?

SPEAKER_01 (00:39):
You know, Sidious Tech has been around for over 20
years, and all we've everperformed is healthcare
consulting.
So we do engineering support aswell as management consulting
for the healthcare market, payerprovider, health tech, and life
sciences.
We have decided to focus becausewe think by focusing on
healthcare, we can help ourclients with some of their
biggest problems and solve itwith people that really

(01:02):
understand healthcare ratherthan just people that understand
technology.
And so while we do a lot ofengineering work lately, a lot
of data and AI work has beendominating what we're selling
because that's what people arebuying.
We field it with teams that knowand understand the nuances of
healthcare.

SPEAKER_00 (01:19):
Okay.
Well, that goes into our topictoday, which is AI in the
healthcare business and theimpact.
So, first off, healthcare AIadoption, it's accelerating
rapidly.
What are some of the biggestchanges you've seen in the
market?

SPEAKER_01 (01:35):
You know, we have seen in the market a desire to
move away from just doing proofof concepts to show that you
understand AI and how to buildsomething to really wanting
full-scale solutions.
And so that's really excitingbecause the elusive return on
investment only really occurswhen you adopt AI.
AI has made it easier to buildthings or to find partial

(01:58):
results, but actually rethinkingyour workflow, your process, and
using AI requires you to thinkdifferently than just
experimenting.
And so it's great to see uhclients now ready to take that
leap.
When it first came out, I thinkthere was so much concern about
can I trust it?
Will it hallucinate?
Will I make decisions that willbe damaging to me?

(02:20):
And so there was a lot ofexperimentation that occurred.
Some companies are still stuckin the experimentation, but
we're seeing more and morepeople really wanting to get to
that scalable solution and uhfind that return on investment
from the work they're doing inAI.

SPEAKER_00 (02:35):
Wonderful.
Well, AI has enormous promise inhealthcare, but there is a lot
of complexity to it.
So, what are some of the biggestbarriers preventing healthcare
organizations from fullyoperationalizing it?

SPEAKER_01 (02:49):
I think that one of the challenges that makes sense
when you think about it, butpeople sometimes overlook is
that healthcare informationcomes in through time.
The tests that I had for mydoctor a year ago may not still
be relevant for the physicalthat I'm about to get and the
tests that I need now.
And so there's a timesensitivity to some of the

(03:10):
information that is part of thecontext that a doctor would use
in making a decision.
AI has to be taught how to treatthat information and which
information is valuable over along period of time, and which
information has uh less valueand has to be considered um
differently.
And so all that context of how adecision is made by a physician

(03:34):
or a nurse or a clinician has tobe trained in AI.
And the challenging thing isthose decisions are often being
made in people's heads in theireducation, and there may not be,
you know, documentation abouthow they used all this
information.
And so training something forscalability requires sufficient

(03:57):
information, a lot ofinformation, so you can better
get AI to make decisions thatare reliable, like a human would
make.
And so it is complex, it doesrequire thinking about quality
and how you are going to buildyour solution.
And it also requires you tothink about where you're going
to keep a person in the loop.

(04:19):
Because it's healthcare, evenwith all the quality that you
can build, there has to be a wayfor supervision of the model.
And so coming up with themetrics that matter is an
important part of healthcare AI,because there will always be
some voice inside of yourorganization or criticizing you
from the outside around you'reusing AI.

(04:41):
Can you trust it?
And so there can't be blackboxes.
There has to be clarity abouthow the decision is made, and
there has to be metrics aboutthe performance of the model
that gets built, not just themodel itself in its use.
So I'd say measures that matteris the first, you know, obstacle
that has to be overcome.

(05:02):
The second obstacle that has tobe overcome is inertia.
People used to listen to a heartby putting their ear to a chest.
The stethoscope came out and itgives you a much better sound of
a heart.
And you can't imagine going to aclinician today without them
listening to your heart withthat device.
But it took decades of thatfacility, that capability being

(05:22):
available for clinicians tochange the way they'd always
done it.
They had a practice, they had apattern, and even though there
was technology available, therewas inertia because the way that
I do things, I know, I trust, Itrust what I hear.
I don't know if I'm going totrust this other device or be
willing to adapt my practice.
AI is like that because AI isgoing to change the way some

(05:47):
people do their work if it'sbuilt correctly.
And so overcoming that inertiaof unwillingness to change,
doing it the way I've alwaysdone it, is a significant
obstacle.
And it's all about adoption.
You know, in a typical ITproject, adoption might be 15 to
20 percent of your budget aboutgetting a technology adopted.

(06:09):
There are some that say youshould be spending$2 for every
one on adoption if you want AIto work.
That's significantly differentspending and attention because
AI won't bring value unlesspeople trust it and adopt it.
So you have to have the metricsthat matter, and then you have

(06:30):
to overcome inertia of wantingto change to be able to get to
the value of healthcare AI.

SPEAKER_00 (06:37):
Absolutely.
That makes a lot of sense.
Well, you recently created anew, like a 2.0 version of a
tool.
Can you share that tool and whatproblems were you trying to
solve with it?

SPEAKER_01 (06:50):
Yeah, we we built a tool called, we called it
Neuron, K-N-E-W-R-O-N.
And what it is, is a tool thatallows you to more easily build
agentic AI.
It has the context of healthcarebuilt into it, it has the data
quality measures andcapabilities to track and
measure the model, but it alsocan integrate into existing

(07:11):
workflows or use algorithms fromother areas and be an
orchestrator of the agentic AIcapability that's desired.
We felt like healthcare is socomplex that if you used a
generic tool in AI, you'd haveto build all this context every
time to make sure that it wasinterpreting healthcare

(07:33):
information correctly.
Some of our clients have usbuild that way within the case
tools that they've chosen.
We build a tool, though, toallow us to go faster and to be
able to expedite the use of AIand healthcare through Neuron.
And so we have uh clients thatare using it today and
experiencing a more rapid umdeployment of their solution

(07:57):
because of what's alreadypre-built and the capabilities.
And so that was really the goalwas to offer healthcare an
option around building within ahealthcare-specific tool set.
Again, we we build outside ofthat tool as well because we're
consultants.
We we follow the architectureprinciples that have been
established at a given company.
But um, Neuron is an option thatthey can consider.

(08:19):
And uh some of our clients aredeciding to go down that path
for their AI builds.

SPEAKER_00 (08:24):
Well, when we look at all the AI tools out there,
it does seem a lot of companiesare still struggling with uh
implementing these AI tools uhacross the company.
What are some of thoseroadblocks?
Uh, why do you think they exist?
And what advice do you have forintegrating these tools more
easily?

SPEAKER_01 (08:43):
You know, Amanda, I think that having specificity in
your goals is important becausethere are going to be different
people with perspectives of whatAI can do, should do, could do,
whether the CFO, CEO, COO, or ondown in the organization.
But if you aren't clear aboutwhy you're doing AI, then it'll

(09:03):
be difficult to be able to getthe inertia overcome within the
organization to be able to makethe change.
I think the organizations thatare doing it best are those that
have set up some type ofgovernance.
Sounds burdensome and overhead,but that governance process
usually drives forward abusiness case and clarity about

(09:24):
why we're doing something.
So it moves it off of the datascientists' experiment land into
a managed implementation withclear goals and objectives and
buy-in from the areas that aregoing to be impacted by the
change.
And so that's a big shift.
And I think organizations thathave that type of governance
process and get that buy-in havea better chance of seeing the

(09:48):
implementation occur.
You know, we're working with uhmultiple clients that have open
lists of AI priorities.
And the challenge is everyonethat submitted one of those
ideas really, really wants it.
And often, a client I was atrecently had, I think, 49
submissions by their seniorleadership team, 49 different

(10:10):
projects they wanted.
They thought they had budget todo two.
That means 47 ideas aren't goingto be fulfilled that somebody
was passionate about.
And so you have to have aprocess to be able to choose
from a lot of great ideas thatcome in from creative people to
get specific things that you cando.
And then there will be peopleupset because their idea wasn't

(10:33):
processed, they're not gettingthe good stuff, you know, that
AI capability and AI help thatthey desired.
But with discipline and with aclear process by which you're
chasing those business outcomes,you know, AI can pay for itself
within a year easily and get areturn on investment because of
the possibilities it brings.
And so often, if you can putthat business case story

(10:56):
together, you can show a pathtowards self-funding and find
the money within youroperations.
But that money may come fromhaving different ways of doing
work.
And if you're not willing tochange what you're working on,
you may not be able to justifythe investment.
The biggest mistake I see peoplemaking is automating a bad

(11:19):
process.
The processes we all use werecreated when AI wasn't around,
when people had to do things,when when you relied on the
human brain and the person to beable to make every little
decision, and then they weresupported by an app that you
logged some of your results.
But most of the thinking anddecisioning was in your head and
in your work with others.
With AI, you you want to pullsome of that thinking into the

(11:43):
AI machine if you're gonnacreate agentic AI capabilities.
And so that causes the personthat used to have all those
ideas in their head to feelthreatened.
Because what I do and the valueI'm creating, they want to get
it from me.
And so overcoming thegovernance, you know, of this is
the right decision and we decidewe're gonna do it is one part of

(12:05):
the inertia.
But then at the ground level andthe process that's being used,
the knowledge workers need tostill feel appreciated, still
need to understand how their jobis going to be more interesting
and better and different.
Because if you don't have theminvolved in the process, it's
gonna be very difficult to getthem to adopt it later or to

(12:25):
make the changes that areneeded.
And so I think that um as wemove into this scalable
enterprise AI solutions, it'sgonna be more about the people
than about the technology.
More about changing how we dothings than generating something
with an AI chat window becausethe real value comes from

(12:48):
automating some of theseprocesses.
You know, I use AI every day nowin my daily work.
It makes me more productive,doesn't it?
You as well, Amanda.
I mean, I I can't imagine nothaving these tools.
My children use AI, and youknow, my 12-year-old is always
telling me things.
I said, Where'd you find that?
She goes, He goes, Oh, ChatGPTtold me.

(13:08):
He doesn't go to Google anymoreto ask questions.
He goes to Gemini or ChatGPT,and he's 12.
You know, the next generation ofworkers that are coming up will
have used AI in college and inhigh school and in their daily
life, and adoption is going tobe easier.
But all of us that came frombefore, that came from a
different world, have to find away to embrace the change, or

(13:33):
you could be an obstacle towardsthe success of the company.
You know, AI doesn't replacepeople, AI unlocks the knowledge
that people bring and allows youto do things you were always
able to do, just do them faster.
And it gives you the possibilityof doing other things too, which
is exciting.
But it's scary, it's scary tochange.
And I think that if people don'tthink about the human side, not

(13:56):
just the engineering side, thatthey're gonna miss the
opportunity to be an earlyadopter in AI because they won't
overcome that inertia, theywon't get the buy-in from the
people that need to change, andthey might just automate a bad
process.

SPEAKER_00 (14:10):
Right.
So it sounds like yeah, itsounds like the most successful
people are gonna be those whoare willing to be adaptable and
flexible.

SPEAKER_01 (14:18):
That that's it.
Yeah, and you know, it is partof it's a big change, and it can
create momentum inside of thecompany, but there's a lot of
blocks to it, and they're nottechnical.
The the technology is easiernow.
The AI tools that are out thereare amazing.
There's so many options of howto build things and how to
create the scalable solution,but the solution has to have

(14:42):
knowledge and context, knowledgeof healthcare practices and
context of how decisions aremade.
And creating knowledge graphsand context graphs as part of
the training of healthcare hasto be governed by that overall
quality process.
Then you can have reliablesolutions, but you still have to
convince people to use them.
You have to help peopleunderstand how this works

(15:03):
better.
I remember when we trained, uh,this is going to show my age,
people to use mouses rather thangreen screens with shortcut
keys.
And I remember nurses that wouldsay, but that's going to slow me
down.
I know the shortcut keys.
I don't want to point and click.
Well, it probably did slow themdown a little bit, but you
didn't have to memorize a bunchof things, right?

(15:24):
And get lost in green screens.
Now AI has the possibility ofchanging that again.
We can speak to the applicationmore readily rather than click
and type.
We can ask questions and reviewcontent and provide workflow
without hands on keyboards if wedesign the systems that way.
It creates the possibility offaster results.

(15:48):
And finding the places in yourbusiness process where those
faster results will be valuableand valued is part of that art
of prioritization and creatingcapability that alters the
company and its ability tofulfill its mission.
And so healthcare is tricky, butAI is very possible within
healthcare if built correctly.

(16:09):
And, you know, getting people toadopt it and understand it can
solve real friction points inthe workforce as well.
So many doctors complain aboutthe pajama time, the work
they're doing at home afterthey've left the office,
starting to finish up theirdocumentation.
And I remember uh in a doctor'svisit I had, I had a resident
that was there being trained,and he said, he goes, You're in

(16:31):
a in uh AI, right?
In consulting.
I said, Yeah.
He goes, promise me thatsometime in my career I will be
able to quit typing in the EHR.
He was using Epic, but he didn'tcare which one.
He didn't want to type anymore.
He goes, It's such a waste of mytime.
I said, It's just around thecorner.
Interesting, a few months later,when I was back at that same
doctor's office, the doctorasked me, Do you mind if my
digital assistant attends thismeeting too?

(16:53):
They had started ambientlistening.
Ambient listening will turn intonot just listening, eventually
they'll turn into suggestionsand workflow and creating a
partnership with the doctorabout changing how they practice
medicine.
But it starts with being willingto be recorded.
I remember when we first talkedabout recording, like doctors
won't want to be recorded.

(17:13):
They'll think it's just a placeto get sued.
You know, they didn't they'llresist.
They won't want, no, if you cansolve the friction of pajama
time of extra work that Ishouldn't be doing, I'm more
than happy to be recorded.

SPEAKER_00 (17:25):
Absolutely.
It seems like there arelimitless possibilities for the
future.
Well, if there was one keytakeaway that you could leave
our audience today with, whatwould that be?

SPEAKER_01 (17:35):
If you're pursuing a healthcare AI project, make sure
you have people on your teamthat understand healthcare, not
just the AI side.
The AI side is interesting andcomplicated and gotten easier.
But if you don't have knowledgeand context in your solution
about how healthcare reallyworks, you will be unlikely to

(17:56):
have the same impact that youshould have.
So as you're thinking about yourteam, don't leave out the BAs,
don't leave out the changemanagement people that help with
thinking about adoption.
Because a perfect mousetrapthat's never used won't catch
any mice.
You need to be able to get thehuman side of it engaged and
excited and helping deform thefuture of the business process.

(18:23):
Technology needs to support allthat, and it needs to be open to
that business input in a waythat's stronger than it's ever
been before because it'sbecoming personal.
It's about my job, my career, mydecision that my name's with.
And the business has to feelcomfortable or they are not
going to use your solution.
So involve the business, involveBAs, involve change management

(18:47):
is my strong advice if you'repursuing AI and healthcare,
probably more important than anyother industry around having
that business voice, because thedecisions we make affect us all.
Our health, our wellness, ourhappiness.
So thank you for your time,Amit.
And thanks for having me today.
This was great.

SPEAKER_00 (19:04):
Yes.
Thank you so much for sharingall your information with our
audience today.
It was a great discussion.
And thank you to our audience.
If you have any questions orcomments, share those and I'll
make sure to get back to you assoon as possible.
Have a wonderful day.