AI Tools and
Behavioral Health

John Lynn: Welcome to the Healthcare IT Today CIO Podcast. I’m John Lynn, the founder and chief editor at Healthcare IT Today, and I’m excited to bring you the most practical healthcare CIO insights and perspectives. We know your job is challenging, and we want to help you be more successful. And our guest today is Patrick Williamson. He’s Chief Information Officer at Array Behavioral Care. Welcome, Patrick.

Patrick Williamson: Hey, John. How are you today?

John Lynn: Doing great. I’m really excited to come from more of the behavioral health side, and we got some interesting security, and especially with AI these days. You have lots of topics to get to, but before we go there, tell us a little bit about yourself and Array Behavioral Care.

Patrick Williamson: Yeah, sure.

I’ve been in the technology industry my whole career. I actually started out really young, just loving computers and tech, and got into infrastructure and computer networking early on. I was part of Geek Squad in the early days. I worked at a radiology company for 14 years, teleradiology, vRad. And now, for the last almost three years, I’ve been doing tele-behavioral health for Array.

A little bit about Array. We’re a leading tele-behavioral health practice. We span multiple care settings. We work in emergency settings. We work in outpatient facilities. And we also do health in patients’ homes, but it’s all virtual.

We’re a practice model, so different than a lot of behavioral health organizations where it’s a marketplace or a network of clinicians.

One of the things that really makes us unique or different is our technology platform. We use Epic. That was a strategic decision we made because a lot of our partners are on Epic, and so there’s just some strategic benefits there with data sharing, workflow integration, and getting connected more closely to the patient’s medical record.

It’s been a great journey over these last three years, and we’ve done a ton of work here. And now AI, this is the next pioneer here. So, excited to talk a bit more about that and what we’re doing today.

John Lynn: Yeah. No, that’s awesome. We know a couple health companies that are on Epic as well for some of those same reasons, the access to the data and the flow of the data that helps inform the decision-making. Yeah, we’ll probably dive into some of that in a minute, but I’d love to start with that topic of AI governance and security, and why it’s so critical and maybe even different in healthcare, and specifically in behavioral health when it comes to governing AI and making sure it’s secure.

Patrick Williamson: Yeah. Well, as we know in healthcare, the privacy of our patients is of the utmost importance. And in behavioral health, the behavioral health note, that’s one of the most sensitive pieces of data to protect just because those are personal. They’re designed to be between the clinician and the patient only. And so, when we share data between organizations, it’s for the context that’s really necessary. If it’s RCM, it’s just billing-related and things to help support the claim.

So, securing the note and ensuring that we have the proper controls and we know where the data is is very important to us and to our patients. And so, when we start getting into the realm of AI products and AI tools and using AI in the actual chart or the healthcare setting or anything related to the note itself, we just have to be very sensitive and very sure that we have the right safeguards and the right technology protecting it.

So we’ve put a lot of effort into ensuring that we’ve got our arms around our environment and our data and the use cases that we’re using for AI purposes today. So it’s something that we can’t compromise. We have to have it.

John Lynn: Tell me more about that. How are you approaching your effort to vet these solutions for your organization? And in that same token, obviously on the CIO Podcast on Healthcare IT Today, I think it’s hard for them to say no, right? You’re like, “Wait, that is a cool tool, and that could benefit the patient.” So it could be hard to say no. What are you doing to make sure that they’re appropriately vetted?

Patrick Williamson: Yeah. It’s funny. I was just telling someone recently that some weeks I feel like the chief wet blanket, where I just keep having to say no.

Right. It’s a tough part of the job because, as an engineer from my background, I’ve been on the innovative side of technology my whole career, and so I love new tools, new tech, and new use cases. So for me personally, it’s painful to say no.

But there’s a lot of responsibility with ensuring that the organization is safe, the patient is safe, and that we’re protecting the data. So when we started on our AI journey, we really got a lot of structure around it last year. We took this approach where we bucketed things into four different categories.

We’ve included automation as part of this as well since that’s oftentimes a solution that’s more appropriate than AI. Looking at a tiered foundation, we’re looking at automation as tier zero. It’s the plumbing. It’s connecting the systems, moving the data, automating the workflows.

Once we move up from there, tier one is the category of embedded AI into the current products that we’re using. That’s the AI that’s closest to the workflow or closest to the data itself. Think of Microsoft Copilot. That would be the embedded AI for the Microsoft stack.

After that, there’s vendor solutions. That’s kind of the tier two, going out and buying a third-party AI product that’s not currently integrated into your ecosystem. And then the third tier after that is custom development.

So we’ve used this framework that we created to help categorize and see where, when someone has a bright idea and says, “Hey, I’m looking at this company, and this isn’t something that they’re advertising or some marketing slick that they have,” we’ll categorize it and say, “Well, this is kind of in that tier two bucket. Let’s see what we have in tier one that could compensate. And how far can we get with maybe an existing product or platform we have before having to look outside.”

It helps ground the conversations into something that’s a bit more standard and structured, and it helps tell the story of whether it’s a good fit or not without it having to be me just saying no every time.

So we’ve tried to create a couple of these different structures and processes to vet the ideas.

I’d also say that we’ve learned this along the way, especially with automation. Sometimes there’s an opportunity for AI or automation, but we’re just not ready for it from a business workflow perspective. Maybe the workflow is a bit ambiguous. Or the data is not ready.

And so being able to assess that as well and say, “Well, actually it might be better if we just looked at the data integration first, cleaned up the process, and then really see what the opportunity is after that,” versus just throwing AI at it and hoping it fixes everything.

Because that tends to be the buzzword out there, that AI can help solve this. But you’ve got to have some prerequisites ready before that happens.

John Lynn: Sure. Well, it’s an interesting way to approach the avoidance of shadow IT in your organization by enabling them to know, “Hey, we’re going to see if we can use our existing tools,” which is an interesting Epic-first approach, as we often hear about on this podcast.

But also helping them know, “Don’t just go do it yourself. We’re going to work on this, and here’s why we aren’t working on it,” or, “Here’s how we can satisfy that need.”

How about things like PHI on ungoverned AI tools? I guess that would be, is that a tier three custom solution or maybe even a tier two solution? Are there any tools that you look at and you’re like, “Yeah, these are clearly HIPAA, they’re fine”? And what does it take for a tool to pass your vetting process?

Patrick Williamson: Right. Well, this is probably one of the more scary areas of what most companies are dealing with today.

It’s interesting. I was at the Gartner Data & Analytics Conference earlier this year, and I was part of the CIO roundtable. So I was able to talk to a ton of different AI leaders and CIOs about what they’re currently doing and what their struggles are.

One of the primary issues that people had today was not knowing what AI tools were in use in their organization and where people were putting their data.

And I think the approach that we’ve taken, I feel pretty fortunate with. We use an endpoint product called Zscaler. Since we’re a virtual company, all of our employees work from home. All of our clinicians. It’s a completely virtual workforce.

And so we needed a distributed security tool that would bring security to the endpoint. By deploying Zscaler, it gives us web filtering and packet inspection, malware protection, all of these things, which includes being able to govern generative AI websites.

It was probably early in 2024 where we blocked all generative AI platforms from being accessed from one of our endpoints.

We also have coupled all access to all of our systems through our Microsoft SSO. And the majority, especially if it’s any system that has PHI, you have to be on an Array device that accesses it.

So we almost have like three-factor authentication to get into any system that has PHI.

With these controls in place, unless someone was doing something very malicious or manually transcribing something, you’re not able to take data from an Array computer and put it into a third-party generative AI platform. You can’t access ChatGPT. You can’t access Claude.

So it’s given us this great control over the environment, and it’s allowed us to enable the right tools for the people across the company to use. I mentioned Microsoft Copilot being one of those.

Copilot has been a bit of an ebb and flow over time. They’ve been behind the eight ball a little bit at times, and then they play catch-up. But really this year, they’ve stayed lockstep with Anthropic and OpenAI in deploying the latest models and the latest features, and now they’re really starting to come up with their own.

So the capabilities of Copilot for general administrative purposes have become really great, actually.

So getting that foundational security has been the place where we focused initially because we needed to get guardrails and understand where people were accessing.

We got a jump start on that. But I hear people all the time say, “Well, yeah, we don’t have web filtering on the endpoint. So we have to roll out training or reminders that people are not supposed to put content in ChatGPT.”

And I think people at this point in time need technical controls. If you don’t have one, there’s specific AI endpoint tools that you can deploy if you don’t have something like a Zscaler.

So I think that’s probably the first thing about using ungoverned AI systems is you’ve got to get the controls in place.

John Lynn: Yeah. No, I love that approach because we love what Zscaler’s doing. It’s fascinating to hear about the evolution of Copilot as well.

On that note, what are some of the initial applications, especially in your behavioral health niche, where AI tools would make sense?

It’s great you’re using Copilot for administrative purposes. What other things are you using?

Patrick Williamson: Yeah. Well, Copilot is being used for back-office administrative use cases. We’re starting to think through some of the use cases where we could use it in administrative, technical, or clinical support roles.

Chart audits is one of those things that has been on our list of priorities for the year. We’re looking at a third-party, kind of that tier-two AI vendor, versus doing something more in the tier one that we already have access to.

Tier one would require a bit more build for us, so we’re balancing what’s the right use case for doing chart audits.

When it comes to inside the clinical setting, ambient is one of the things that’s talked about most often, and we’ve done some ambient pilots. And I think, just from my personal experience, I’ve been in doctor’s offices with my children.

And the ambient tools in that setting are amazing. I’ve noticed this huge shift. Instead of the clinician sitting at a keyboard typing and swiveling, they’re able to sit face-to-face and have a conversation. It totally changes the dynamic in the room.

Well, in a virtual setting like we have, the experience is different.

The notes that our clinicians are taking over a 52-minute therapy session, we’re not just typing away the whole time. There’s definitely notes being taken, but it’s definitely more of a conversation.

So we’ve actually seen less value in ambient in a therapy setting.

And then when we look at some of our other clinical services, we use a lot of structured templates and a lot of clinical checklists, and ambient’s not there to really assist with that.

So we’re still leveraging dictation, and dictation works really well for some clinicians, and others still prefer to type.

But we hear there’s some exciting stuff on the Epic roadmap around dictation into forms and some of the ambient tools that Epic is coming out with.

And so this is one of those areas where it’s like, well, the business value isn’t clear, and it’s not really calling our name, and we know it’s coming on Epic’s roadmap.

So this is a good decision point to say we’re just going to wait. Unless something changes, we’re just going to wait until the feature presents itself in that tier-one category.

So it’s in BH, I think it depends on the type of BH services you’re offering. And for us, there’s other Epic tools, chart summaries and other things within Epic, that are helpful.

But something that is as exciting sounding as ambient, that’s just an example of one of the use cases that we’ve said, “Yeah, it’s not really the right time for us yet.”

John Lynn: Yeah, that’s really interesting, and you’re right. It’s funny, when ambient first came out, I said telehealth is the perfect place for it. You’re already recording it, right? It makes sense.

But obviously with a behavioral health 52-minute session, a lot of the AI hasn’t been tuned for that because there’s a lot of extraneous information and other information.

So yeah, and we are recording this before Epic UGM, and I think it publishes after Epic UGM. So we’ll see if they announce something at the upcoming UGM that will help. So I’m sure you’ll be watching with bated ears.

Patrick Williamson: Yeah, it’s actually the number of features that get announced regarding AI at UGM over the last two years has been hard to keep up with.

We went to UGM last year, and I think there was 100-some features in development, and there was already 100-some live. I mean, I’m not going to get the numbers right, but it was staggering to think about the amount of stuff that they’re releasing.

So from a platform perspective, it’s certainly one of these instances where it’s like, well, there’s a lot of value in waiting unless you have a critical gap that you need to solve today.

John Lynn: Well, and it seems like they’re undercutting a lot of vendors on price, too. So that’s interesting. We’ll see how that plays out.

You mentioned that you’re on Epic, and like I said, we’ve seen this with other telehealth organizations where you can get access to their record in Chart Review and be in Epic, and being able to share that data and capture it from maybe the clinical setting.

Why is that valuable for behavioral health to have visibility into the full medical record as opposed to just your local records, if you will?

Patrick Williamson: Yeah. Well, I’ll tell a little story here. It’s one of my favorite stories that I’ve been part of since joining Array.

We were about to go live on Epic. It was November of 2024. We were all out in New Jersey at an office that we used to have there, and we had just finished doing the data ingestion and getting all of the patient charts created. We had shut down on a Friday, and we were opening back up on a Monday. So we had all weekend to do this data preparation, importing the schedules and all of that.

And so it was the first time that we were actually able to see the Care Everywhere network in action.

And Marlene, our VP of Therapy, she runs our therapy group. She pulled up one of her patients just to see how Care Everywhere worked.

And no joke, tears were brought to her eyes, and she was like, “Oh my gosh, I can see all of this rich data that I could never see before that has a direct impact on a person’s behavioral health.”

Things like sleep studies, blood pressure, the other medical visits that this patient had had. A lot of times, behavioral health symptoms that people are dealing with can manifest in other physical illnesses. And so if you’re treating the physical illness, you’re not actually treating the mental illness. You may not be able to get rid of the physical illness.

And so being able to marry these two, the mental health and the physical health, together, it gives the clinician, the behavioral health clinician, such a better, well-rounded picture of what a patient is really dealing with.

And so maybe it’s understanding what their blood pressure issues are and working through whatever stress is causing their blood pressure to be elevated instead of taking another medication.

And so Sarah, our Chief Medical Officer, came over, and Marlene was literally crying.

And it was just this moment of silence across the conference room as Marlene was telling the story, and it was just so impactful.

Many of us have all had some sort of behavioral health experience, whether it’s personally or maybe a family member. And so to be able to see Marlene be so moved with having this visibility.

I think this patient specifically had a sleep study that Marlene was trying to get access to. And just having the patient coordinate physically calling the sleep study place and saying,

“Hey, can you fax this over to my therapist?”

A lot of times patients don’t want to do that, and it’s just extra steps. You’re already dealing with someone who might have a mental health struggle going on.

So the fact that she was able to see this information, it was really impactful just to see the transformation that it can start to have on the care that’s being delivered.

So that was a major selling point to me when I was like, “Wow, we made the right call. I’m super thrilled that we chose to go with Epic because of the value that it’s creating for us.”

John Lynn: Yeah. Well, Judy’s husband who inspired Care Everywhere will be a really grateful to hear that story, I think.

Patrick Williamson: It’s an amazing story. I’ve told that story a dozen times in the last two years. It’s one of my favorite ones.

John Lynn: Yeah, what probably many people that are listening don’t know is when I started and implemented an electronic medical record, I did it in a health and counseling center at a university.

And that was our vision, the full seven areas of wellness, right? Physical, mental, emotional, spiritual, etc., etc. We have to treat the full student, and we wanted to be on one record so that the mental health and the physical health could all be together.

So I love that you’re doing that as well. That’s awesome.

We talked about this a little bit, but I would love to see how you’re approaching it. It sounds like you have the four tiers, but do we need more AI? Or do you think, for most organizations, it’s better to leverage the tools they already have?

And when do you differentiate, “Hey, let’s get another tool,” or, “Hey, let’s use Epic or Microsoft or whatever you already have?”

Patrick Williamson: Yeah. I mean, this is the dilemma of the IT leader, right? It’s what is the guidance that you can provide the organization to use the right tool for the job.

We have an internal use case where this one’s not super exciting, but it’s super practical.

We had an IT ticketing system that was just not modernized. It worked. It met our compliance needs, but it was difficult for the IT group to use.

In 2024, we signed up with an enterprise contract with a platform called Asana. It’s a work management system.

And Asana’s done a lot of good work embedding AI into the product and creating AI-powered workflows. And so we looked at an upcoming renewal cycle in our ticketing system and we said, “You know what, we could build this and make it better leveraging AI inside the system.”

It has the compliance that we need. We have the healthcare contracts, the BAA with Asana. So let’s just rebuild our ticketing system in this product that everyone is already using to manage other work and workflows.

And then we were able to leverage some of the built-in AI functionality, some of that tier one that I keep referring to.

And so this was just a practical example of us looking at the products we have, looking at a product catalog, the contract dates, and being proactive about, well, do we want to renew this product? Do we need to renew this product? And do we have something in-house that could compensate and take that functionality over?

And so this happens every day, where someone brings forward a request for an AI use case. It just happened to me two days ago.

And it was like, actually, we can do this with our automation platform, Workato. We can actually connect this thing and this thing together. It’s not an AI use case. It’s just data movement.

So when I started at Array, one of the things that I thought was really important was to turn the IT group into business consultants instead of IT ticket takers.

Not to say that there’s anything wrong with that, but we’re a smaller IT group. We’re 18 people, and we need to offer more value to the company and not just be here when they need us. We need to insert ourselves into areas to help guide and shepherd technology use across the organization.

And so moving into this business consultant-type mindset has allowed the individuals in the IT group to be involved in more business meetings, more business discussions, and then be able to help say, “You know what? We’ve got a product over here that can do this. We have a product over there.” Or, “Yeah, we haven’t looked into this yet.”

So it helps steer the conversations as the business teams are having them versus our business groups having to go find their own solutions and then just come to IT to implement.

So it’s kind of a mix. Do we need more AI or just better use of tools?

I always fall back to let’s look at the tech that we have. Let’s look for the simplest solution. Let’s look for something cost-effective. And really only use AI when it truly makes sense.

John Lynn: Yeah. I call it the keep-the-lights-on CIO versus the strategic business CIO.

And I think the former is starting to die, right? It’s dwindling. They’re retiring. They’re losing their jobs and being replaced.

So I think we’re seeing that shift across the industry, which makes sense.

Well, we always like to wrap up these episodes with a little career discussion, career advice, something like that.

So I’d love to hear from you. What advice would you give an aspiring CIO that may be considering a behavioral health organization? What should they know? What should they expect?

Patrick Williamson: There’s advice I like to give my team members. And it has a lot to do with what I was just talking about regarding moving into more of a business consultant-type role or mindset when interacting with others around the company.

And it’s really about being proactive and getting involved. Listening for opportunities to engage with folks, whether they’re talking about a workflow issue or maybe a technology problem.

Ask questions and see if there’s something that you can do to help with that, or maybe there’s a technology we have that can help solve the issue.

There’s so many features and functionalities that exist in our platforms today that it’s impossible for everyone to keep up on, but one of our jobs as technologists is to help shepherd people along the way and understand the best way is to use the platforms that we have and all of the features within them.

So the other thing I really like to tell my team members is try to anticipate what people are going to ask.

So if you send off an email and you answer some questions, anticipate what the follow-up questions are going to be and try to proactively answer those because it shows that you’re thinking ahead. It shows that you’re putting yourself in the shoes of the person that you’re interacting with.

And it goes along with that business consultant mindset. Just be more proactive and think ahead.

So those are things that are very practical. There’s nothing brand new about those, but I think it really starts to shape how people see you within the business when they see how you’re leaning into the problems and opportunities that are affecting how we operate or maybe some pain points that might exist.

So those are some real practical things that I like to tell my team members.

John Lynn: Yeah. That’s really good advice. I like to tell people, try to make it easy for them. Don’t make them ask three more questions. I think it’s the same lines.

And man, I remember going and sitting in the nurse’s station, sitting in the counselor’s office. And people might look at it and be like, “Wait, you’re not working. You’re just hanging out here chilling.”

And I’m like, “No, you don’t understand. You don’t know how many times the nurse comes back and says, ‘Oh John, I’m glad you’re here. I wanted to ask you this,’ or the counselor says, ‘Hey, I was going to send you an email, but I didn’t have a chance.’ Could you tell me”

Patrick Williamson: It’s like, well, one of my favorite email quotes, and I probably repeat this at least once a month, is, “I didn’t have time to write you a short email, so I wrote you a long one.”

And it’s so true. It takes extra effort to refine your thoughts and make them more clear and more succinct and easier to understand, especially when we’re describing something technical in nature.

And so I probably tell someone on my team that once a month. I’ll see a long email or a long Teams message go out, and I’ll respond with, “Didn’t look like you had time to write a short email, so you wrote a long one.”

And so It’s a fun situation and provides some coaching.

John Lynn: Good advice. Well, Patrick, this was a lot of fun. I appreciate you taking time to talk with us and share your insights and perspectives in this fast-moving, wild world of technology and AI in healthcare. And lots of opportunity, you know, and also some risks we have to definitely consider.

So, thanks for helping us understand that. And thanks everyone for watching and listening.

If you want to find more great healthcare IT content like this, be sure to check it out at healthcareittoday.com or search for the CIO podcast by Healthcare IT Today on your favorite podcasting applications. Thanks, Patrick.

Patrick Williamson: Thank you, John. Take care.