If your clinicians won’t touch the AI tool you just rolled out, the model probably isn’t the problem. Ask why healthcare AI projects fail and the answer usually comes down to trust. AI fails for reasons regular software doesn’t: clinicians won’t act on recommendations they can’t explain, patient data ends up in models it shouldn’t, and biased data quietly produces biased results. In healthcare, trust has a price of entry: compliance. It’s a problem you have to solve before any algorithm can help.
Ghazenfer Mansoor, founder and CEO of Technology Rivers, dug into exactly that problem as a guest on AvePoint’s #shifthappens podcast with host Dux Raymond Sy. His team has built close to 50 healthcare applications, about half of them HIPAA compliant, and on the show he shared what it takes to earn clinicians’ trust, from HIPAA controls to clean data to human oversight. Here are the key lessons from the episode.
Listen to the full episode: The Foundation of AI Adoption Success in Healthcare
Why Healthcare AI Projects Fail Without Trust and HIPAA Compliance
Digital transformation in healthcare, Ghazenfer says, isn’t that different from any other industry. You improve the workflows people already have, you build trust with users before go-live, and you measure the results that matter to them. As Ghazenfer puts it, “at the end, you’re still dealing with humans.”
What’s different is compliance, because that’s where clinician trust in AI starts. Your clinicians know their organization is on the hook if patient data is mishandled, so a tool’s output doesn’t matter if it can’t prove it protects that data. In Ghazenfer’s words, healthcare users “are not gonna touch any application if it’s not HIPAA compliant.”
So before you ask what AI can do, make sure your team trusts how it handles data. Once they do, the payoff can be significant.
How AI Is Transforming Clinical Decisions With Deeper Insights
The most direct payoff is better clinical decisions. When a physician spots something and wants a second opinion, they used to rely on their own knowledge and whatever limited data was in front of them. AI can now pull insight from all the data available.
That includes history your doctors probably aren’t looking at. Physicians often see only the last couple of years of a patient’s history, and something from 20 years ago rarely comes up. With AI working across the full record, those older details come back into view, and the connections can surprise even the patient.
The physician still makes the call, which keeps a human in the loop. AI is “augmenting by providing all the data so that you can make a better decision,” as Ghazenfer describes it. And clinical care isn’t the only place this pays off. Some of the fastest wins are operational.
How AI Solves Staffing and Scheduling Challenges in Home Care
If you run a home care, assisted living, or autism care organization, you know scheduling is a hard matching problem, and it’s one of the clearest uses of AI in home care. The right caregiver needs the right skills, the right location, and open availability.
Workload is the factor manual scheduling tends to miss. You assign someone who fits every criterion on paper, then find out they’re already 90% booked, and productivity drops. Technology Rivers built a tool that matches caregivers to patients across skill, location, availability, and workload, with a heat map that shows who actually has capacity. Your coordinators get balanced assignments far faster than they could make by hand.
How AI Enables Proactive Care Through Remote Patient Monitoring
Remote patient monitoring is where AI moves you from reactive care to proactive care. If you’re collecting data from wearables and other sources, you don’t have to wait for a patient to call. AI can detect anomalies and alert both the patient and the doctor.
That matters because most people notice symptoms and put off seeing a doctor until it’s urgent. The value, Ghazenfer explains, is that “this predictability would allow you to take actions before that happens.”
Every one of these use cases runs on patient data, though. That puts compliance back at the center, and ignoring it is part of why healthcare AI projects fail.
What HIPAA Compliance Really Means for AI in Healthcare
Calling a product “HIPAA compliant” doesn’t mean much on its own. You need to know what it covers. Building HIPAA-compliant AI starts with where patient data goes. Never put PHI into a general-purpose AI tool. Ghazenfer’s team cleans up and de-identifies data as part of its AI process, and showing clinicians that process is often what wins their confidence. Any AI service that does handle PHI needs HIPAA safeguards and a signed business associate agreement (BAA).
Beyond the data itself, check for these controls:
- Device security. A phone left in a restaurant can become a HIPAA violation, so mobile and desktop access both need protection.
- Encryption of data in transit and at rest, at the infrastructure level as well as inside the application.
- Audit logging that records who looked at data, when, and for how long.
- Granular access control. Seeing a patient’s name in a list is different from opening their prescription, clinical notes, or MRI, and your system has to tell those apart.
Apply the same checklist to any vendor selling you HIPAA-compliant AI, not just to software you build.
It’s also far cheaper to get this right from the start. Bolting compliance on late is part of why healthcare AI projects fail. Ghazenfer compares it to adding a basement to a house that’s already built: possible, but a huge lift, because authentication, data handling, and even caching all have to change. If you’re planning a new product, HIPAA-compliant healthcare software development with these controls built in from day one saves you that rework.
AI Bias Starts With Data: Why Human Oversight Still Matters
Compliance protects your data. The next question is whether that data is good enough to trust.
“AI is not biased, but AI is biased based on what data you provide.” (Ghazenfer Mansoor)
Go back to the caregiver matching tool, because it shows the limit of AI in home care: it only knows what’s in the data. The AI recommends an assignment, and the coordinator sometimes rejects it because the model missed something it had no data on: a language barrier, a cultural match, or a patient who only wants a female caregiver. Better data gets you far better recommendations, but where the data falls short, you need a person to catch what the model can’t see.
So keep a human in the loop for final decisions. And make your AI’s recommendations transparent enough that coordinators and clinicians can see why it suggested what it did.
Start Small With Clear Workflows to Scale AI in Healthcare
Before you launch anything, define the workflow. Unclear requirements are one of the planning mistakes that sink any healthcare software project, and they’re a big part of why healthcare AI projects fail too. If you can’t say clearly what you want, AI won’t figure it out for you.
Then start small. Pick one simple clinical or administrative workflow for your first healthcare AI implementation, and help your staff use AI there. Small wins build the trust that bigger projects need.
One Technology Rivers client came in after three failed attempts with other vendors. The team shipped a first version in about six weeks. Once the client’s staff saw it working, weekly one-hour calls turned into idea sessions, and over two years the client more than doubled.
Starting small also makes mistakes cheaper. “You don’t know the problem until you see it,” Ghazenfer says, so it’s better to change direction now than a year in. For more on building software people actually adopt, see Ghazenfer’s book, Beyond the Download.
The Common Thread: Trust Built on Data You Can Defend
Every lesson from the episode points to the same answer: trust. Clinician trust in AI comes when a tool is compliant, when patient data is protected and the data it learns from is clean, when its recommendations are transparent, and when a person keeps the final say. That trust grows fastest through small, visible wins, not one big launch. Ghazenfer covers why most AI projects fail before they start outside healthcare in another interview.
If you’re planning a healthcare AI implementation, start with one workflow. See more of what Technology Rivers has built in its portfolio.
