[00:00:00]: Shift Happens podcast.

[00:00:02]: Welcome to the Shift Happens podcast, where we explore the latest trends in technologies transforming the future of work. Today’s guest is Ghazenfer Mansoor, CEO and founder of Technology Rivers. Ghazenfer brings a rare blend of technical depth and a healthcare viewpoint, and in this episode, he shares what it really takes to make AI work in healthcare.

[00:00:20]: From designing for compliance from day one to building AI trust through small acts, Ghazenfer shares a practical, grounded view of how transformation actually happens in high-stakes environments.

[00:00:27]: Let’s get into it.

[00:00:40]: Hey everyone. Welcome back to another episode of Shift Happens, and today I am so excited to spend the next 30 to 45 minutes with Ghazenfer.

[00:00:52]: Hey, Ghazenfer, welcome.

[00:01:00]: Hey, D, thanks for having me on the show.

[00:01:05]: Before we dive right in and really unpack how AI is benefiting healthcare, the first question I ask all the guests is: what song do you think of that best describes change and the shift we’re experiencing right now?

[00:01:08]: To me, the best song is Don’t Stop Believin’ by Journey.

[00:01:20]: And it’s not just classic, it’s about holding through the uncertainty, which is exactly what healthcare leaders face when embracing AI. You don’t see the full result until some stage, or may never, because you keep adapting.

[00:01:24]: So persistence is key in any of the healthcare projects, whether it’s remote monitoring or rethinking home care scheduling. You hit roadblocks.

[00:01:35]: And the real difference between success and failure is believing in the vision long enough to see it work.

[00:01:40]: And that’s the gist of entrepreneurship as well.

[00:01:52]: Absolutely. This is one of my favorite songs. As you’re describing what it means to you, I think about your background, right?

[00:02:00]: One of the first verses made me think about your background. It says, “Just a city boy, born and raised in Detroit. He took the midnight train going anywhere.” Now, I don’t know if I’m divulging your secret here, but you don’t really have a healthcare background, do you?

[00:02:02]: I kind of got into healthcare by accident.

[00:02:13]: Many years ago, as I was working, I got a project with Veterans Health. That was my first entry into healthcare. I supported multiple projects for the Veterans Health Administration.

[00:02:20]: And then as I started this business, our second customer was a health tech company and, from one to another, we started building different apps.

[00:02:21]: Projects for healthcare companies, including for Johns Hopkins Innovation Center. And that’s really where, with one of those startups, we learned HIPAA and then we started building HIPAA-compliant products.

[00:02:40]: So now in 10 years, we have done close to 50 different healthcare applications, and about half of them are HIPAA compliant.

[00:02:52]: And as such, naturally now you’re a subject matter expert in the industry, right, apart from the technology itself.

[00:03:00]: Yeah. But obviously, healthcare is huge. We still believe that we know little about healthcare. We need SMEs for each area because for any of the projects that we are working on, many times there are so many complications that you do need people like oncology clinicians and others. Without these people we’re not good. Obviously building tech is a key part, and then the tech people and the healthcare people are two different worlds, and bridging that gap—knowing healthcare,

[00:03:10]: Knowing compliance, knowing the challenges, and working with these people for a long time puts us into

[00:03:35]: A good place when interfacing with them.

[00:03:40]: Absolutely. I remember—I’m dating myself—maybe 15 years ago I got into a project in healthcare to help an organization build an FDA Part 11–compliant document management system. And there’s,

[00:03:52]: Rightfully so, a lot of regulations on how information is handled, which relates to HIPAA as well.

[00:04:00]: But you’re right, it’s so massive. There’s a lot to learn and we’re talking about, at the end of the day, people’s lives being at stake. As you do your work and gain experience, what parallels do you see between transformation in healthcare and other industries, and what can

[00:04:01]: Healthcare leaders learn from other industries and vice versa, because now you’ve seen the healthcare world and how technology evolves in other industries as well?

[00:04:20]: In reality, the technology, digital, or transformation part is similar for both, with the exception of a few things. Obviously compliance is a bigger part, which is similar in financial and some other sectors as well.

[00:04:27]: But at the end, you’re still dealing with humans, whether it’s a clinician or a bank or any of those. So you want to build applications that are trusted. So I think when it comes to parallels,

[00:04:40]: It’s really the workflow. For example, if you are implementing anything, improving the existing workflows is key.

[00:04:55]: You have to build trust with your users before you go live. You have to measure results that matter to the end users. There are a lot of similarities when it comes to transformation in healthcare and in other industries, but more importantly,

[00:05:00]: When it comes to healthcare, you must make sure you know the compliance.

[00:05:15]: I think the bigger challenge I’ve seen is that many times just calling it “HIPAA” doesn’t mean much. You have to know deeply what that means.

[00:05:20]: Healthcare users are very sensitive, whether it’s doctors, clinicians, any of those. They are not going to touch any application if it’s not HIPAA compliant, because at the end it’s the entity that’s responsible for HIPAA.

[00:05:24]: They are the ones who are touching that. So making sure that those applications that are built for that industry are

[00:05:40]: Really compliant.

[00:05:42]: Got it. So speaking about these applications and the need to make them more compliant, can you share some specific examples with our listeners on the types of solutions you’ve developed, especially

[00:05:55]: AI-powered ones, and what impact and benefits these applications or solutions provided to customers?

[00:06:00]: Yeah, it’s interesting because there are a lot of things changing as well. As we are working on different applications, the projects that we have been working on for years—the AI at that time versus AI now—are very different.

[00:06:09]: For example, in one of the cases it’s a doctor’s consultation.

[00:06:20]: Let’s say any prognosis or anything that you are talking to your patients about. You identify something and now you want a second opinion, or you are sharing the experience. AI can give you a lot more insight based on all the available data that people have.

[00:06:30]: In the past, doctors were relying just on their knowledge and limited data.

[00:06:40]: Another common scenario I’ve seen even in real life: if you go to a doctor, many times they just don’t know your history more than maybe a couple of years. They don’t care about something that happened 20 years ago.

[00:06:42]: Now with AI, all that data is available. So the interesting aspect, for example, is having your healthcare records and deep research that gives you a lot more insights that are mind-blowing.

[00:07:00]: Sometimes you wonder yourself, “I never knew this was part of it.” Yes, you knew it at the back of your head, but now with AI you’re getting much deeper results. In our case, we work a lot with home care, assisted living, and the autism care industry as well.

[00:07:07]: So for example, in home care there are complex staffing and scheduling problems.

[00:07:20]: We created a tool that can automatically match caregivers to patients based on skill, location, availability, and workload. So as you’re assigning people,

[00:07:28]: Typically you would assign based on certain criteria, but then you realize that person is 90% occupied.

[00:07:40]: Their chest is a little bit overworked on one side. That hurts productivity. So providing a heat map and everything gives you good visibility of who has more availability and what the workload is. All these improvements,

[00:07:52]: With AI, we’re able to do much faster and quicker. Similarly,

[00:08:00]: Flagging inconsistencies in their reporting and logs, because you don’t want to get rejected by Medicaid after the people are gone. That means you are also losing money. So that’s helping you improve your payroll processes. These are real

[00:08:03]: Examples that we work with. On remote patient monitoring, when AI analyzes patient data through wearables and you’re getting data from different sources,

[00:08:20]: Do you wait for the patient to call you, or can it detect anomalies and then notify the patient and doctor?

[00:08:22]: I think that’s a very common and interesting use case because in many cases we get symptoms and we wait to go see the doctor till the end. Where AI can make a real big difference is how

[00:08:40]: This predictability allows you to take actions before that happens.

[00:08:51]: And being more proactive, for sure. Absolutely. We’ve seen the impact even in the news, where Satya Nadella at Microsoft was talking about certain research that humans can do and

[00:09:00]: The potential for finding cures for some illnesses, which is awesome,

[00:09:10]: Humans are important, obviously. But you’re not only dependent on human-centered decisions, because AI is augmenting by providing all the data so that you can make a better decision.

[00:09:17]: And that’s key, right? Sometimes there’s a

[00:09:20]: Misunderstanding that it’s on autopilot, where it’s not the case. To your point, it’s augmenting.

[00:09:22]: It’s your copilot. That’s why I love that Microsoft calls it Copilot. Because especially in healthcare, trust is important—trust and, more importantly, the validity of the data. So have you

[00:09:40]: Found, at least initially in the healthcare industry, that there was some pushback? “I don’t know about that, it’s too good to be true.” But it seems like there’s a lot of adoption and it’s well embraced in a lot of different applications coming out.

[00:09:55]: Right, right.

[00:10:00]: I think the pushback is always there. It’s human nature. Change is tough. People don’t want it. So bringing those people in and building the confidence of clinicians early makes a huge difference because,

[00:10:03]: Even outside healthcare, if you build something and just ask people to use it, they’ll have fear: “My job is going to be replaced,” or “Why is this happening? How is it going to impact my work?” People are not ready or they’re not comfortable that they can do it.

[00:10:20]: So there are a lot of issues, and that’s the main pushback we see

[00:10:22]: With AI as well.

[00:10:31]: So speaking about change management, I’m curious: have you seen any difference with

[00:10:40]: Adoption of AI-specific technology versus previous technology in the healthcare industry? And what are your tactics in driving adoption and usage?

[00:10:47]: There is always pushback.

[00:11:00]: There are challenges. I think the fear of bias in results shows up a lot. On one side, people say, “These are mind-blowing results,” but many times they say, “There are a lot of mistakes. Is it giving the right data? I don’t want to give my data because it goes into the AI and then it’s going to the world.” So it’s training and coaching them on what is the right way and how we are doing it. For example, our process includes cleaning up the data and de-identifying the data.

[00:11:20]: You can’t put any PHI-specific data into AI. Once people see how that whole process works, they feel more confident.

[00:11:24]: But yes, pushback is still there.

[00:11:40]: That’s a great point around adoption, and trust is important. Now, speaking of trust, we know there are a lot of compliance requirements like HIPAA,

[00:11:47]: Encryption, role-based access controls.

[00:12:00]: What are some of the key compliance areas people have to be aware of in the healthcare industry, and how do you navigate them?

[00:12:07]: Obviously HIPAA is key, and even within HIPAA, what is included is a big thing. We have a big checklist of

[00:12:18]: Items that need to be done. In a traditional office environment, you’ve seen documents with paper locks. When everything moved to the web, things changed. Now you have a phone. You carry the phone everywhere. You have a phone in the restaurant and you lose the phone or somebody gets access to it.

[00:12:20]: Suddenly that’s a violation of HIPAA. So how do you make sure that you are in compliance?

[00:12:40]: There is a long list of things you need to be aware of when it comes to mobile, desktop, or any device. And also

[00:12:55]: From an internal process perspective. For example,

[00:13:00]: As I mentioned earlier, de-identifying data. Data cleanup is important. What data you are putting into AI is key.

[00:13:02]: Making sure of encryption in transit as well as at rest. Even if your applications are built, at the infrastructure level it’s very

[00:13:20]: Important to make sure HIPAA compliance is there. And then audit logging: who is looking at the data, at what time, how long. It gets more interesting for web and mobile, because maybe you saw part of the data in a list versus when you go deeper,

[00:13:24]: Looking at the prescription, notes, or MRI.

[00:13:40]: So very specific

[00:13:41]: Control is needed. Access control is key. As you’re building these, or evaluating any applications you’re bringing into your organization, it’s important to have that compliance and make sure your applications follow it. And that’s a common challenge we’ve seen.

[00:14:00]: Many times we get asked, for example, “This application—can you make it HIPAA compliant?”

[00:14:10]: It’s like, “You have a house now. Can you add a basement to it?” Yes, you can, but it’s going to be a huge lift. Adding something from the beginning is different, because now your whole process changes: how you authenticate, how you handle data.

[00:14:20]: You may have, for example, caching in your architecture.

[00:14:24]: How do you remotely remove that? Things like that. All these come into play when it comes to compliance.

[00:14:40]: Another thing we’ve seen recently is how you handle certain compliance issues with certain regulations.

[00:15:00]: For example, the GDPR requirement says you should be able to delete the data. Apple and Google now put that as a requirement for any app: you have to give an option to delete. Since you have to delete all records, HIPAA says you can’t delete the data, because if it’s my data, someone looked at it at a certain time.

[00:15:08]: If I’m deleting my data, where is that? Where do you draw the line? Those are interesting use cases

[00:15:24]: And conversations.

[00:15:40]: That’s a great example. So how do you handle that? You’re exactly right: one says delete it—right to be forgotten under GDPR—because it’s my data,

[00:15:42]: But with HIPAA, you can’t. So where’s the compromise?

[00:15:55]: There are

[00:16:00]: Solutions. I wouldn’t call them compromises, but more about

[00:16:03]: How you implement certain rules within the compliance requirements of your organization.

[00:16:20]: There are certain business decisions you have to make. For example, do you

[00:16:24]: Anonymize all of the data but still keep everything, so that at some stage, if there’s a need, you can go back and see what happened? You have all of that patient data there, anonymized so that nobody knows whose data it is.

[00:16:35]: Is one

[00:16:40]: Strategy. Related to

[00:16:47]: That, the question of ethics comes up, especially with new AI capabilities, because as you described, it can do deep research and pull information from ages, as long as data is there. How should leaders think about ethical considerations, especially for

[00:17:00]: Applications that directly affect patients?

[00:17:02]: I think the bigger part is training. I’d say AI is not biased, but AI is biased based on the data you provide. So you should know: if my data is limited, then what? I’ll give you one example in one of our home care

[00:17:20]: Applications. AI recommends caregiver assignments, but the coordinator

[00:17:24]: Approves or doesn’t approve them because it may have missed, for example, the language barrier or a cultural match. You know how in certain cases, “I need a female only,” things like that. Once you have all the data, AI gives great results, but there are personal factors:

[00:17:28]: Even though I am in this boundary and my location is all good, which is followed for everybody, a person may not

[00:17:40]: Prefer to work with a certain patient. Things like those come in. So how does the AI know? You have to train AI. The more data you have, the better it is. When we talk about ethics or bias, it’s very important that

[00:18:00]: You have the right data, and you are also transparent. The patient, therapist, clinician—everyone knows why AI is providing a certain recommendation. It’s not that AI is biased; it’s based on the data. And human involvement is key. We talked earlier as well.

[00:18:02]: The final decision still goes to the human.

[00:18:20]: Yes, AI does everything. It auto-matches, but there are times when humans have to intervene.

[00:18:22]: That’s more of a business decision: in what cases do you want the whole end-to-end flow handled by AI versus cases where humans

[00:18:24]: Intercept? It goes case by case.

[00:18:40]: And this is where the human in the loop is critical, right?

[00:18:41]: You don’t just let it run autonomously and leave it alone.

[00:18:47]: Right. Bias starts with data. We diversify sources and test for those

[00:19:00]: Biases. We keep humans in the loop. That’s really the gist of it.

[00:19:03]: Shift Happens podcast.

[00:19:10]: Well, Ghazenfer, this has been an awesome conversation and I want to make sure I’m respectful of your time before we wrap up.

[00:19:22]: Advice would you leave for every healthcare leader before jumping in and rolling out a massive AI initiative or even an AI application?

[00:19:35]: I think data is the key. AI is not going to be magic if your data is not right. Cleaning up your data—AI is not

[00:19:40]: Going to come and help you improve your workflow if your data is not right.

[00:19:47]: It’s very important that you look at your specific use cases and clean up the data. Those are the two most important pieces because you can go to ChatGPT and ask it to

[00:20:00]: Generate any content. It can create it, but it doesn’t know how it’ll fit into your workflow.

[00:20:03]: You don’t want to just adapt what AI is giving you. You want to implement and create seamless workflows that you already have. AI is going to help you improve your workflow,

[00:20:20]: And really the good AI is where

[00:20:22]: People—your users—don’t even feel it’s AI, because it’s just a better workflow in their day-to-day.

[00:20:30]: So yes, data cleanup is important, and defining the workflows. It’s like any project you do or any job you have: if the requirements are not clear, that’s a very big reason most projects fail. AI adoption can also fail if you don’t have a very clear

[00:20:40]: Definition of what you want. Once you define these things, you can implement.

[00:21:00]: A couple of other things I would highlight: there’s a big misunderstanding that because everything is AI now, people feel, “I just bring in AI and it’s going to solve my problem.” Yes, for generative AI maybe, but for real AI it’s about the data.

[00:21:03]: That means the cost can be huge. The more data you have, the more you have to train. That process itself is not simple or small. You have to have those processes to make sure you are ready.

[00:21:30]: As you’re planning for AI, you want to make sure you understand those caveats.

[00:21:40]: One last thing I would add is: start simple. You don’t have to start a big project. You can start with a very simple

[00:21:45]: Clinical or administrative workflow—helping your staff start doing certain things with AI gradually, because that starts building trust.

[00:21:49]: Once they start building trust,

[00:22:00]: The work just keeps going. One of our clients came to us after

[00:22:03]: Three bad tries with different vendors. When they came to us, we created the first version in a month and a half. Once they saw it working,

[00:22:11]: Surprisingly, after that, every week we would do a one-hour call with their whole team and ideas kept coming.

[00:22:20]: Their team started bringing in, “Can we do this? Can we do that? How about this?” In two years, they grew more than double, and that whole process brought a lot more efficiency into their processes.

[00:22:30]: So I think it’s

[00:22:35]: Important to start small, because in any big project you don’t know the problem

[00:22:40]: Until you see it. You want to identify those very early. There’s less risk if you want to change direction. You want to change now rather than after

[00:22:47]: A year, when it’s too late.

[00:23:00]: That’s right. Again, I love the advice: make sure you have good data quality and it’s clean. Make sure you have the right processes and standards, because technology is technology. In the end, if you don’t have a sound process or

[00:23:11]: Standards, then it’s useless. And third, go for low-hanging fruit.

[00:23:20]: Start simple, get quick wins, and from there adoption will grow and new ideas will form. It’s also a way for you to learn in an evolutionary way versus jumping into big projects.

[00:23:30]: Well, Gazenfer, thank you so much. Grateful for your time, and for our listeners, thank you for tuning in. We look forward to hearing from you or seeing you in our next episode.

[00:23:57]: Shift Happens podcast.

[00:24:00]: And that’s a wrap for this episode. Ghazenfer really hammered home some key points.

[00:24:03]: Today we learned that success with AI isn’t just about the tech itself, but about practice. The fear of bias and job replacement is a real hurdle, and that’s why building trust is so important.

[00:24:20]: And it starts with transparently showing users how AI makes its recommendations. This episode was a great

[00:24:30]: Reminder that data is the key to everything. As Ghazenfer said, AI is not magic if your data is incorrect. And we can’t forget his advice to start small. It’s about building small wins that get your team excited and invested in the process. Until the next time.