AI is changing how clinicians detect risks, manage records, and make decisions. Experts in digital health explain the breakthroughs helping care teams act faster and deliver more personalized support. See how these tools are improving workflows while keeping doctors in control.
- Biopsy High-Risk Lesions Without Delay
- Prioritize Follow-Ups Before Risks Escalate
- Free Pediatric Clinicians for Family Care
- Reveal Mental Health Trends Between Visits
- Deliver Trusted Answers Within Existing Workflows
- Let Clinics Query Governed Data Directly
- Guide Fitness Choices at Critical Moments
- Keep Doctors in Control of AI
- Govern Autonomous Agents at Runtime
- Embed Predictive Insights Into Clinical Workflows
- Validate Blood Pressure Models Before Use
- Pair Ambient Notes With Dynamic Risk Signals
- Clear Insurance Barriers Before Appointments
- Turn Paper Records Into Clinical Intelligence
- Train Models Where Patient Data Resides
- Track Biomarker Trajectories Over Time
- Refine Orthopaedic Plans With Image Analysis
- Filter Preventive Data Through Physician Judgment
- Personalize Behavioral Care With Emotional Insight
- Predict Lows Before They Strike
- Eliminate Repetitive Request-Review Bottlenecks
- Bring Hands-Free AI Into Surgery
- Remove Compliance Friction From Telehealth
- Avert Medical Delivery Disruptions Early
Biopsy High-Risk Lesions Without Delay
I run AI-assisted dermoscopy on every suspicious lesion now, a system that scores a mole against a database of millions of prior biopsy-confirmed images and flags asymmetry, border irregularity, and color variation patterns my eye might rate as borderline. It caught a melanoma in situ on a patient last year that I would have watched rather than biopsied immediately, the lesion looked clinically unremarkable but the algorithm flagged a color pattern that pushed me to sample it that same visit. That case changed how I use the tool. I stopped treating the AI score as a second opinion I could overrule on gut feeling alone, and started treating a high-risk flag as a mandatory biopsy trigger regardless of how benign the lesion looks to my eye. The technology has not replaced clinical judgment in my practice, it catches the specific blind spot every dermatologist has, pattern recognition fatigue after the two hundredth mole of the day, when a subtle outlier starts to look routine simply from repetition.
Dr. Cameron Rokhsar MD FAAD FAACS, Founder & Medical Director, New York Cosmetic Skin & Laser Surgery Center
Prioritize Follow-Ups Before Risks Escalate
The most meaningful use of AI in digital health I have encountered is AI-assisted follow-up prioritization. In gastroenterology and liver care, patients often have repeated lab reports, imaging findings, symptoms, medication changes, and follow-up timelines. When this information is scattered, important changes can be missed or reviewed late.
An AI-supported system can help organize this information and flag patients who may need earlier attention, such as someone with rising bilirubin, persistently abnormal liver enzymes, falling platelet counts, worsening abdominal symptoms, or missed follow-up after an advised investigation. It does not replace clinical judgment, but it can help the care team notice risk sooner.
This has changed how I think about healthcare delivery by making follow-up more proactive rather than purely visit-based. Instead of waiting for a patient to return only after symptoms worsen, digital tools can support earlier reminders, clearer follow-up planning, and faster review of concerning patterns. It also improves communication because patients can better understand what needs urgent attention, what can be monitored, and when the next step is needed.
The biggest lesson is that AI is most valuable when it strengthens human attention. In digital health, its role should be to reduce missed signals, organize follow-up, and help doctors deliver safer, more timely care.
Dr. Jitendra Mohan Jha DrNB, Gastroenterologist & Hepatologist, Dr. Jitendra Mohan Jha Gastro & Liver Clinic
Free Pediatric Clinicians for Family Care
One of the most promising uses of AI in healthcare is its ability to support physicians by reducing administrative tasks and helping organize information more efficiently.
In pediatric gastroenterology, much of quality care depends on understanding a child’s complete health story, symptoms, growth patterns, nutrition, family concerns, and changes over time. AI-powered tools have the potential to help clinicians organize information, summarize records, and streamline documentation, allowing more time to focus on conversations with patients and families.
The key change AI can bring is not replacing the physician’s role, but enhancing it. Healthcare decisions still require clinical judgment, experience, and an understanding of each child as an individual. AI can assist by improving efficiency, but the human connection remains central to pediatric care.
I also see significant potential in AI-supported patient education. Families often have many questions about symptoms, nutrition, and healthy lifestyle choices. Responsible use of technology could help provide clearer, more accessible health information while supporting personalized communication between healthcare teams and families.
AI applications in healthcare are most valuable when they work alongside clinicians, improving workflows, supporting communication, and helping physicians deliver thoughtful, patient-centered care.
Mona Dave, Provider, Manisha “Mona” Dave, MD – Pediatric Gastroenterology
Reveal Mental Health Trends Between Visits
The most promising use of AI I’ve seen in digital health is also one of the least flashy. It’s the ability to pull together what a patient tells us between visits, like mood ratings, sleep, symptom questionnaires and notes, and show the pattern over time instead of a single snapshot. In psychiatry, a lot of what matters happens in the weeks between appointments, and people often struggle to remember how they felt even last Tuesday.
It has changed how I approach care. Instead of starting an appointment by asking “so how have you been?” and relying on memory, I can start with, “I noticed your sleep has been slipping for two weeks, tell me about that.” That makes the conversation more specific and more collaborative, and it helps me catch early warning signs sooner. Patients tend to feel more seen when I can point to something real in their own words.
I’m also careful about the limits. The tool can highlight a trend, but it can’t tell me what it means for a person’s life. That still takes listening, clinical judgment and trust. I see AI as a way to notice more and to free up my attention for the human part of the work, not as a replacement for it.
Shebna N Osanmoh, Psychiatric Nurse Practitioner, Savantcare
Deliver Trusted Answers Within Existing Workflows
The most innovative AI application I have implemented in digital health is an OpenAI-driven search-and-instant-answer product at my organization that transformed how teams access healthcare knowledge and make operational decisions.
I observed that healthcare professionals and operational teams were spending significant time searching across disconnected systems, documents, and knowledge bases. Traditional search tools returned lists of documents instead of actionable answers, creating inefficiencies and pulling valuable time away from higher-value work.
As a Product Manager, I helped define the product vision and roadmap for a generative AI solution that moved beyond traditional search to deliver faster, evidence-grounded answers. I partnered with 20+ cross-functional stakeholders across engineering, AI, cloud, business, and operations, using feedback from 30+ user interviews to identify pain points, prioritize use cases, and set measurable success metrics.
This implementation was innovative not simply because we used generative AI, but because we integrated it into existing workflows while maintaining accuracy, responsible data practices, and healthcare compliance. We continuously evaluated performance, incorporated user feedback, and focused on making AI a tool that augments human judgment rather than replaces it.
The results were significant: the product reached more than 20,000 users, processed over 1.74 million questions, achieved approximately 95% answer accuracy, reduced search effort by 70%, and increased user adoption by 40%.
This experience fundamentally changed how I approach healthcare delivery. I began viewing AI not as a standalone technology or another reporting layer, but as an enabler of proactive, workflow-integrated decision support. The real opportunity is to reduce administrative burden, make trusted information accessible at the point of need, and allow healthcare professionals to focus more of their time on meaningful decisions and patient care.
It also reinforced an important principle for me: digital health innovation must go beyond technical performance. It must combine clinical and operational relevance, trustworthy outputs, regulatory compliance, user adoption, and measurable business value. That is how I believe we can scale AI responsibly and create a lasting impact on healthcare delivery.
Sonam Kumari, Product Manager, Pharmacy Operations Strategy Technology Innovations
Let Clinics Query Governed Data Directly
I am Mark Sternig, CTO of Focus, where we provide IT, security, data, and AI services to healthcare organizations. One honest caveat on the framing: we do not deliver care ourselves, we build the data foundation the medical practices and health systems delivering it stand on, so I will answer from that seat.
The most innovative thing we have implemented is conversational analytics, letting a healthcare organization talk to its own data in plain language instead of waiting weeks for us to build a report for them. For years the process was painful. A practice sort of knew what it wanted, we sat through whiteboard sessions, built exactly what they described over two weeks, and it still came back not quite right. Now, on top of a cleansed and governed data model, someone at that practice, an administrator, a physician, an operations lead, can ask a question in plain language, see the dashboard, and dig deeper by simply asking the next question. No ticket, no meeting.
What changed for me is where our value sits. It moved from building reports to guaranteeing the foundation underneath them, the ingestion, the cleanliness, the governance, and the context layer that tells the AI what the data means, so a clinic can explore its own numbers without ever exposing PHI. In healthcare that last part is the whole ballgame. The innovation is not the chatbot. It is the clean, governed data beneath it that makes the chatbot trustworthy in a setting where a wrong or over-exposed answer is a patient-safety and compliance problem, not just a bad report.
Mark Sternig, Chief Technology Officer, Focus
Guide Fitness Choices at Critical Moments
The most useful AI implementation we’ve delivered in digital health was a recommendation layer inside a cross-platform fitness and health app in 2025. During the eight-month build, we connected an OpenAI model to each user’s recent activity and goals. The app drew on recent activity and goal data from HealthKit or Google Fit, then suggested one timely action, such as shortening a workout or scheduling rest.
What made it valuable was the timing. A conventional dashboard asks the user to interpret a chart after the fact. We brought guidance into the moment when someone was deciding whether and how to exercise. We also kept the scope narrow. Its recommendations stayed within the product’s fitness remit: workout duration, rest, challenges, and progress toward a goal. That boundary made the feature easier to understand, test, and trust.
The project showed why personalization works best as a sequence of small decisions inside the user’s routine. The team has to design the data permissions, recommendation logic, and interface together, because a technically sound suggestion can still fail if it arrives at the wrong time or asks too much of the user. After launch, early feedback highlighted the app’s clarity and practical tracking, while engagement moved upward as the client prepared the next monetization phase.
Define when the model should stay silent, show the user why a suggestion appeared, and measure whether people act on it.
Roman Surikov, Founder & CEO, Ronas IT | Software Development Company
Keep Doctors in Control of AI
Hi, I am a medical doctor by education and CEO of AskDocDoc (telehealth). AskDocDoc is an operator, not a clinic (we connect doctors and patients, and medical decisions belong to the doctors), so maybe my practical side is useful for you.
In our company AI went first where the work is operational – building the product, analytics and technical support. The case I like most is a sister project, where we built an AI assistant for doctors. The model helps the doctor consult patients; it does not replace the doctor’s judgment.
What didn’t work for us is vague tasks. The result looks fine but it’s wrong (and you don’t see it at first). That’s why a person checks the result.
Evgenii Arsentev, MD, PhD
Chief Executive Officer, AskDocDoc
Evgenii Arsentev PhD, Chief Executive Officer, AskDocDoc
Govern Autonomous Agents at Runtime
The shift that matters is not a single tool. It is AI that acts instead of AI that answers.
The most consequential thing I see in digital health is agents working inside EHRs, inboxes, and revenue cycle queues. They are not producing a recommendation for a clinician to read. They are taking steps. That is useful, and it changes oversight completely.
It changed my approach. I stopped treating approval as the finish line. Health systems approve AI at the front door, then lose sight of it once agents start acting. So the work moved to runtime: an inventory of every agent with an accountable owner, a permission check on each action, an audit trail the hospital keeps, and a stop control that does not depend on the vendor.
Boards no longer ask whether to use AI. They ask whether we could stop it if we had to.
Will Haver, Founding CEO, Skovos
Embed Predictive Insights Into Clinical Workflows
The most significant innovation in digital health is the evolution from static systems of record to predictive clinical intelligence layers that synthesize fragmented data in real-time. For decades, healthcare IT focused on the basic digitization of paper records, effectively turning physical filing cabinets into digital folders. True transformation requires moving beyond manual administrative tracking to deploy AI models that act on longitudinal data to predict patient outcomes before they escalate.
This approach directly addresses the persistent hurdle of interoperability. Instead of waiting for disparate hospital systems to sync perfectly, we now implement intelligence layers that extract and normalize data from pharmacy records, lab results, and wearables to provide a unified risk profile for every patient. In practice, this means identifying subtle physiological markers of sepsis or rising readmission risks long before they are caught by traditional manual observation.
Implementing these systems fundamentally changed my approach to healthcare delivery. I no longer view technology as a passive infrastructure for compliance, but as a system of utility where data works actively for the clinician. We have seen that even the most sophisticated models fail if they add cognitive load or force a provider to toggle between screens. Consequently, the priority has shifted to invisible integration-ensuring predictive insights are delivered directly within the existing clinical workflow.
Pragmatic execution in this field means moving past the hype of AI curing disease to focus on the immediate necessity of reducing physician burnout and improving acute care responsiveness. The goal is to restore the clinician’s focus to the patient by automating the heavy lifting of data analysis, ensuring that technology serves as a reliable partner rather than an administrative burden.
Kuldeep Kundal, Founder & CEO, CISIN
Validate Blood Pressure Models Before Use
The most innovative AI I’ve worked on in digital health was a blood pressure risk model that wasn’t allowed to affect any patient’s care at launch. We built a remote monitoring platform for hypertension, with adherence scoring and risk prediction drawing on BP cuffs, wearables and Epic records. We ran the model in shadow mode first. It scored real patients alongside the care team, but clinicians didn’t act on its outputs until they had checked its reasoning against their own calls. Every score came with an explanation they could review. After validation, risk-scoring accuracy was 30% better than the baseline.
The bigger shift for me was realizing the model was the easy part. Most of the work went into the data layer underneath it: getting device readings and Epic FHIR data into one consistent format. That layer cut device onboarding time by 70%. Now I start every healthcare AI project by asking whether the data can be trusted and whether clinicians can check the output. I only ask which algorithm to use after that.
Vlad Mironov, Head of Customer Success, Jelvix
Pair Ambient Notes With Dynamic Risk Signals
The most innovative implementation I’ve led was pairing ambient clinical documentation with longitudinal risk analytics in a dual-engine architecture. While our ambient AI layer transcribes patient encounters directly into structured EHR notes, the predictive engine streams data from labs and connected devices to calculate risk velocity, giving our teams early, dynamic visibility into how fast a patient’s risk profile is shifting rather than waiting for fixed thresholds.
This fundamental shift transformed our approach to care delivery: we stopped viewing AI as a tool for automating isolated administrative tasks and started building it as a continuous intelligence layer. By embedding validated predictive signals directly into native EHR workflows instead of isolated dashboards, we eliminated context-switching fatigue, streamlined clinical action, and gave our care teams real-time risk stratification without compromising clinical judgment.
Noah Gula, AVP at OSP, OSP
Clear Insurance Barriers Before Appointments
The most useful thing we’ve built isn’t the most futuristic: moving insurance work to before the patient arrives.
At QuickIntell we deploy AI agents across the revenue cycle and clinical documentation. One orthopedic practice, with six surgeons and three PAs across two locations, had missed pre-authorizations delaying payment, staff swamped by reminder calls, and under-documented notes. Over eight weeks it added pre-authorization checks before imaging and surgery, voice agents for reminders and rescheduling, and an AI scribe in consults. Our published case study reports that 90% of imaging and surgical cases were pre-cleared before the visit, no-shows fell from 17% to 9%, and documented comorbidities rose 25%. These are company-reported results from a combined deployment. The practice administrator put it this way: “Our surgeons now walk into consults with everything prepared.”
What it changed for me is where I look for value. It comes from timing and ownership more than from the model. When the authorization is settled before the appointment and every open requirement has an owner, the patient doesn’t learn about a coverage problem at the front desk. Clinicians still review and sign every note; the AI removes the chasing.
Rahul Agrawal, Founder & CEO, QuickIntell
Turn Paper Records Into Clinical Intelligence
The most innovative use of AI I’ve implemented isn’t the most technically impressive one. It’s the one that quietly removed hours of work nobody thought could be removed.
At Medicai, we built AI to handle chronic patient data intake. Picture an oncology patient arriving at a public hospital with 100 pages of printed records from previous institutions. Historically, the oncologist or their resident read every page by hand and reconstructed the disease timeline before care could begin. It took hours per patient, and in busy departments, it didn’t always happen in full.
The file is now scanned at reception. By the time the doctor opens the record, it’s chronologically ordered, labeled by document type, with diagnosis, staging, and treatments already extracted. A physician recently told us it saved five hours of work on a single patient. We launched it four months ago, and 10 hospitals now use it to process thousands of files every month.
What makes it innovative isn’t the extraction itself. It’s what happens after. Once patient history is structured, rather than sitting in paper and free text, it becomes usable data. That opens the door to patient registries filtered by cancer type or specific mutations, and to treatment recommendations aligned with clinical standards such as ESMO.
It fundamentally changed our approach to healthcare delivery. We stopped thinking of AI as a tool that makes a single task faster and started treating it as the layer that turns scattered records into a foundation for better decisions. The clinician still makes every call. What changed is that they start from a complete picture instead of building one from scratch.
The most valuable AI in healthcare often targets the unglamorous work surrounding care, not care itself. Return clinicians their time, structure the data along the way, and new capabilities follow.
Andrei Blaj, Co-founder, Medicai
Train Models Where Patient Data Resides
The most innovative use of AI in digital health I’ve come across is federated learning, which trains a model across many hospitals without any patient data leaving the building. Each hospital trains the model on its own records and shares only what the model learned. The best-known proof came during COVID, when about 20 hospitals worldwide trained a shared model to predict which patients would need oxygen, and it outperformed what any one site could build alone.
It matters because the biggest barrier in health AI isn’t the algorithm. It’s that the data needed to build good models sits in hundreds of systems that legally can’t pool it. Federated learning turns that from a legal standoff into an engineering problem.
It changed how I approach AI for sensitive environments. I used to treat privacy as a compliance review at the end. Now it’s the first architecture decision: where the data lives, what leaves, and what an attacker could learn from the model itself. My research on anonymization showed that privacy techniques carry real accuracy trade-offs, so that choice can’t be bolted on later.
The breakthrough in health AI wasn’t a smarter model. It was letting the model travel to the data instead of the other way around.
Mahesh Shivnatri, Technology Leader, Deloitte
Track Biomarker Trajectories Over Time
The most valuable use of AI in our practice is not diagnosis. It is longitudinal data interpretation.
A single patient can generate hundreds of data points each year across advanced lipid testing, metabolic markers, hormone panels, genetics, body composition, continuous glucose monitoring, sleep data, and fitness testing. The challenge is no longer access to data. It is turning that volume of information into something clinically useful.
Traditionally, physicians review the latest result and compare it with a reference range or a prior value. That approach can miss the more important signal: trajectory.
We use AI to organize a patient’s data across time and identify patterns that are difficult to see when results are reviewed in isolation. Before a visit, I can quickly see which markers are improving, which have plateaued, and which are gradually moving in the wrong direction.
That changes the clinical conversation. A fasting insulin level may still be technically “normal,” but if it has increased over three consecutive visits, that trend may be more meaningful than the individual number. Detecting those shifts earlier gives us an opportunity to intervene before a patient crosses a traditional diagnostic threshold.
AI also changes how physicians use their time. Instead of spending the visit searching through historical results, more time can be spent interpreting the data, discussing tradeoffs, and deciding what action makes sense for that patient.
Patients benefit from the same visibility. When they can see exactly when a biomarker changed after starting a medication, adjusting nutrition, or changing their training, the relationship between behavior and outcome becomes much clearer.
The guardrails are critical. AI organizes, compares, and flags. It does not make the clinical decision. The physician remains responsible for interpreting the findings in the context of the patient’s history, goals, and overall health.
The real opportunity for AI in medicine is not replacing clinical judgment. It is giving physicians a better view of the patient over time.
Dr Amir Saeed MD, Co-Founder, Prosper Medicine
Refine Orthopaedic Plans With Image Analysis
One of the most useful applications of AI I’ve come across in orthopaedic care is AI-assisted analysis of medical images and surgical planning. For conditions such as knee arthritis or patients being considered for joint replacement, AI can help analyze X-rays and other scans and give us more detailed information about the joint. I see this as a useful second layer of support because it can help with planning, measurements, and comparing changes over time. The final decision, however, still depends on the patient’s symptoms, examination, scans, and clinical judgment.
What has changed for me is the way I think about technology in patient care. AI is not there to replace the doctor; it can help us spend more time understanding the patient and making a better-informed plan. For example, if an AI tool highlights a change on an image, I can discuss what it means with the patient and explain the treatment options in simple terms. I believe the real value of AI in healthcare will come when it makes care more precise and useful without taking away the human side of medicine.
Dr. Jaswinder Singh, Owner, Dr. Jaswinder Singh – Orthopedic Surgeon, Joint Replacement & Sports Injury Specialist
Filter Preventive Data Through Physician Judgment
I’m Niko Hems, Head of Strategy & Medical Communications at YEARS, the Berlin-based preventive medicine clinic.
Here’s my take: the most interesting thing happening with AI in health isn’t what most people expect. As AI makes more health data available, from wearables, imaging, genomic, the real bottleneck isn’t collecting it anymore but it’s interpreting it. And that’s where physician judgment becomes more valuable.
At YEARS, we work with a 3-Tesla whole-body MRI, advanced biomarker panels, and cardiovascular diagnostics. The data is rich. But what matters for a client is what a physician decides to act on, what to monitor, and what to set aside. More data without that interpretation is noise. AI can help sort the noise, but someone still has to make the clinical call.
Where I think AI is genuinely useful in preventive medicine:
– Imaging and pattern recognition — AI-assisted analysis of MRI scans or full-body skin screening can flag things faster and sometimes more consistently than the human eye alone
– Longitudinal data tracking — when you’re following someone over years, AI can identify subtle shifts in biomarkers that signal emerging risk before a single number goes outside the normal range
– Risk modeling — combining genetics, blood markers, imaging, and fitness data into a composite picture of future health risk is something AI is genuinely good at
Where I think the hype is outpacing the evidence:
– AI-driven longevity claims — epigenetic age clocks, biological age measurements, and AI-generated health recommendations often sound sophisticated but lack the clinical validation to drive treatment decisions
– Wearables as health solutions — your smartwatch gives you data, not decisions. Most people don’t have the framework to know what to do with it
Niko Hems, Longevity Expert. Head of Strategy & Medical Communications, YEARS
Personalize Behavioral Care With Emotional Insight
The most innovative application of artificial intelligence in digital health is the development of emotionally intelligent, adaptive systems designed to map and respond to individual psychological and relational needs. Rather than acting as a simple automated triage tool, advanced natural language processing and pattern-recognition software can now evaluate complex emotional states, linguistic shifts, and behavioral trends across various health care touchpoints. This technology fundamentally shifts healthcare delivery by moving systems away from rigid, one-size-fits-all models toward hyper-personalized care. In behavioral health and complex systemic care environments, AI assists practitioners by tracking subtle longitudinal changes in a patient’s emotional presentation—such as emerging distress or shifting relational boundaries—between clinical hours.
Streamlining healthcare practices with these tools changes the operational landscape in several key ways:
Reducing Administrative Burden: Automating routine documentation, intake sorting, and tracking frees up practitioners to focus entirely on human-centric, empathetic care delivery.
Enhancing Care Continuity: Bridging the gap between clinical sessions by providing secure, context-aware support ensures patients maintain momentum in their treatment plans.
Targeting Interventions: Helping multidisciplinary teams quickly spot individuals at risk of falling through systemic cracks, allowing providers to allocate resources precisely where intense human intervention is required. Ultimately, this integration does not replace the irreplaceable depth of human connection. Instead, it acts as an intelligent scaffolding that relieves operational friction, allowing health systems to meet people precisely where they are emotionally.
Amanda Phillips LCSW, CEO, MY AP Therapy
Predict Lows Before They Strike
The most useful AI I’ve seen in healthcare isn’t a chatbot. It’s the predictive alert on my dad’s CGM that reads his trend line and predicts a low about 20 minutes before it happens. Before this, my dad’s lows would be discovered at 3 a.m. when he was in a hypo state, all shaky and confused. But now my phone buzzes and my mom rushes him some juice. The night remains boring, which is the point of all this. The first month it was switched on, it raised so many false alerts that we nearly turned it off. But after his doctor had tweaked the thresholds, it was so much more useful. Now I’m a big fan of health tech that warns you early and almost never false alerts.
Ace Zhuo, CEO | Sales and Marketing, Tech & Finance Expert, TradingFXVPS
Eliminate Repetitive Request-Review Bottlenecks
The particular application of artificial intelligence that has stuck with me the most involves an instance when I witnessed AI perform all sorts of repetitive informational tasks from within a digital health workflow process. From the technology side of the implementation, I was particularly not impressed by the AI model itself but by its effects on the users of the system.
Prior to implementing the AI into the workflow process, people were manually reviewing and categorizing some 1,000 request cases a month, each taking about 8 minutes to do. Using AI to classify the information from each request case and send it along to the correct person cut the average time spent per request case to about 3 minutes after 12 weeks or so.
That seems easy, but it was completely rethinking how to approach developing healthcare products. We needed to ask much tougher questions. What happens if the AI is incorrect? What data is it supposed to have? What decision-making can it help with, and at what point does a person need to approve the decision?
After having more than 15 years experience as a software engineer and technology executive, I realized that the success of an AI-based product is not about putting AI everywhere. It is about identifying a real bottleneck and eliminating work there without introducing new risks elsewhere.
The lesson has also taught me to be more mindful of automation. Just because something can be automated does not mean it should be done automatically, especially in healthcare.
That is what I have learned from this experience. I would begin with the workflow, identify repetitive tasks, measure the outcome, and always include a human in the process where there will be an impact. The use of AI is relevant where it adds value to the healthcare process and not because it is AI.
Vitaliy Kononov, Co-Founder & CTO, Atty
Bring Hands-Free AI Into Surgery
We’re helping turn the operating-room display into an AI assistant that surgical teams can interact with hands-free.
I’m Tolga Tarhan, founder and CEO of Atomic Gravity. Our team partnered with SoftAcuity to develop AI capabilities for Elsa, its Brilliant Assistant platform, developed in collaboration with Barco. Elsa combines voice interaction, sensing, multimodal AI, and edge computing within a medical-grade surgical display. It is designed to help clinicians access imaging, communication tools, and applications through natural voice commands, bringing assistance directly into their workflow.
The vision comes from SoftAcuity founder and CEO Bruce Kennedy, who previously founded surgical-imaging company PCM, acquired by Arthrex in 2011. Drawing on that experience, Bruce saw an opportunity to make the surgical display a point of access to information, communication, and intelligent assistance in the operating room.
Working on Elsa changed how I approach building AI for healthcare delivery. We had to think beyond the model to the physical environment: how clinicians interact with the system, what information matters in the moment, and how assistance fits into an existing workflow. That means starting with a specific source of friction for the care team and designing around it.
My biggest takeaway: useful healthcare AI needs to fit the way care is delivered. The goal is to make information and tools easier to access while keeping clinicians in control.
Tolga Tarhan, Founder & CEO, Atomic Gravity
Remove Compliance Friction From Telehealth
One of the most innovative uses of AI I’ve worked on is DalloAI, which is trained on the real-world knowledge of consultants who have spent years working inside regulated industries. Instead of using AI just to answer questions, we’re using it to help healthcare businesses stay compliant, understand what they need to do next and act on changes in their market.
That has changed how I think about digital health delivery. The opportunity isn’t just using AI in the consultation itself. It’s using AI to remove the operational and regulatory friction around healthcare, so new telehealth models can launch faster, stay compliant and reach more people.
For me, that’s where AI becomes really powerful: when it makes healthcare innovation easier to actually deliver.
Kyle Hunt, CEO, Dallo
Avert Medical Delivery Disruptions Early
The most innovative use of AI in digital health I’ve encountered is predictive technology that identifies potential disruptions before they affect time-sensitive healthcare deliveries. In nationwide logistics, I’ve seen how AI can evaluate traffic, weather, route history, and shipment data to flag risks to medical supplies or other critical deliveries much earlier than a traditional tracking system. In one situation, an anticipated transportation delay allowed the delivery plan to be adjusted proactively rather than waiting for a missed milestone to trigger action. That changed my approach by reinforcing that AI is most valuable when it gives people enough time to make a better decision, not when it simply automates one. For healthcare organizations, I recommend using AI around specific operational bottlenecks—such as inventory shortages, delivery delays, or demand forecasting—while keeping human oversight for decisions that affect patient care. The technology works best as an early-warning system that helps teams become proactive instead of reactive.
Talya Turgeman, Owner, Worldline Express
