I built a patient engagement app nobody used. Then I called them instead.
The app had 15% engagement. Fifteen. I'd spent months on it, convinced that the right interface would pull bipolar patients back day after day. It didn't. People opened it once, maybe twice, then never again. (feels like another lifetime, honestly.)
So I stopped building screens. I called the patients instead. With AI.
Engagement went to 85%. That number is the entire reason HANA exists.
I'm telling you this because there's a piece going around right now, a good one, that says the vendors are lying to you about patient engagement. And it's mostly right. So let's be real about what actually works.
Do AI voice agents actually improve patient engagement?
Yes, but not the way the sales decks claim. The honest answer is that voice works when the alternative is silence. Most clinics aren't choosing between a fancy app and an AI call. They're choosing between an AI call and nothing, because the front desk is drowning and nobody has time to dial 200 discharged patients.
A recent Deepgram analysis made the point that peer-reviewed studies on voice agents are still thin, and the 30-50% booking lifts in vendor pitches don't have the receipts. Fair. But the absence of a trial isn't the absence of a result. We've run over a million patient interactions with zero critical adverse events, and the engagement gap is not subtle.
Why do apps and portals keep failing patients?
Because you're asking sick, tired, often older people to do work. Download this. Remember a password. Tap through five screens. The portal assumes the patient will come to you. Most won't.
My 15% app failed for the same reason every portal fails. It put the burden on the person least equipped to carry it. The phone flips that. The phone calls them. No download, no login, no app store. My grandmother would answer a phone before she'd ever find a patient portal, and she's exactly who post-discharge follow-up is for.
That's the whole insight. You go to the patient. You don't wait. And once you internalize that, every clunky portal and abandoned app starts to look like the same mistake wearing different clothes, a system designed for the convenience of the institution instead of the reality of the person who's sick.
What makes a voice agent worth deploying instead of just adding noise?
A voice agent earns its place when it has bounded scope, captures structured data, and escalates fast. That's it. The 2026 deployments that work all share that shape. The ones that flame out tried to make the AI a doctor.
HANA never gives therapeutic advice. It asks the structured questions, pain level, meds, whether the follow-up is booked, and the second something crosses a red line, a human gets pulled in. You can see how that maps to real clinical use cases on the site. The AI is the first line of engagement, not the last word on care. Get that boundary wrong and you've built a liability. Get it right and you've built reach you literally could not staff.
Isn't this just replacing nurses with robots?
No, and honestly this framing drives me a little crazy. Healthcare doesn't have too many nurses. It has way too few. Every survey says the same thing, staffing shortages are cutting patient access, and the labor isn't coming back.
When the routine 72-hour check-in is automated, the nurse isn't out of a job. The nurse is freed to handle the patient who actually said something scary on the call. That's not replacement. That's triage at a scale a human phone tree can't touch. I'd rather my best clinician spend her afternoon on the three patients in trouble than on the 197 who are fine.
What about cost? Is the ROI real or is that a pitch too?
The ROI is real, and it's not close. Post-discharge readmissions cost real money, often $15,000 to $20,000 per readmission, and every avoided one pays for a lot of automated calls. We see roughly 31:1 return for clinics running follow-up at scale, and you can pressure-test that math against your own numbers on our pricing page.
I watched a Stripe dashboard hit a million dollars in a single day once, at a different company, with a product that wasn't actually good. Scale hides problems. So I'm allergic to vanity numbers now. The readmission math isn't vanity. It's a line item your CFO already hates.
Key Takeaways
Where I land is this. The skeptics are right that the voice AI market is full of inflated claims, and you should make every vendor show you production evidence, not a sandbox demo. But the skepticism shouldn't tip into doing nothing, because the patients waiting on a follow-up call don't get to pause while you wait for the perfect randomized trial.
The thing that actually moves engagement isn't a smarter app. It's removing the work from the patient and going to them directly. My 15% app taught me that the hard way. The 85% taught me it the easy way. Pick bounded scope, structured capture, fast escalation, and a vendor who'll still be there at month 18. HANA is open-source and self-hosted on purpose, so you're not betting your patient data on someone else's roadmap.
FAQ
Are AI voice agents HIPAA compliant?
They can be, and any serious vendor will sign a BAA. HANA goes further by being self-hosted and open-source, so your patient data never has to leave your control or depend on a third-party model provider.
What patient follow-up tasks are safe to automate today?
Post-discharge check-ins, medication adherence outreach, appointment reminders, and care-gap recall. The shared rule is capture and escalate, never diagnose or change a dose. Anything clinical routes to a human.
How fast do clinics see results from automated follow-up?
Usually within the first cycle of calls. Engagement jumps immediately because the patient no longer has to initiate contact, and the readmission and no-show reductions show up in the first few months. If you want to see what that looks like for your practice, grab a call with me.