The 7-Day Window Almost Nobody Is Hitting
The first product I ever built in healthcare was a mental health app for bipolar patients. Beautiful thing. Mood tracking, sleep logs, gentle nudges, a colour palette I fought about for two weeks.
Engagement was 15%.
Fifteen. Out of a hundred people who genuinely needed it, eighty five never opened it twice. I sat with that number for a long time, and what finally broke through wasn't a product insight. It was a timing one. The app was waiting for people to come to it. The people who needed it most were the ones least able to come.
So we stopped waiting. We called them instead, with AI, at the moment that mattered. Engagement went to 85%. That was the beginning of HANA.
I think about that a lot when I read the readmissions literature, because healthcare keeps rediscovering the same lesson and then filing it somewhere it can't find again.
Why do post-discharge follow-up calls matter so much?
Because there's a window, and it's brutally short. A Kaiser Permanente study of nearly 12,000 heart failure patients found that follow-up contact within seven days of discharge was associated with 19% lower odds of readmission within 30 days. Contact after day seven showed no significant benefit. None. The clock isn't a soft guideline, it's the whole intervention.
And the cost of missing it is not abstract. AHRQ puts US 30-day adult readmissions at roughly 3.8 million a year at an average of $15,200 each, and CMS can cut a hospital's Medicare payments by up to 3% under the Hospital Readmissions Reduction Program. That's covered well in HealthTalk AI's breakdown of the readmissions playbook, which is worth your time.
So if everyone knows this, why isn't it happening?
Because the maths doesn't work. That's it. That's the entire failure.
A clinic or unit discharging fifty patients a day cannot reliably reach all of them inside 48 hours by hand. Someone has to build the list. Someone has to dial. Patients don't answer unknown numbers. Friday discharges sit until Monday, which eats three days of a seven day window before a single attempt gets made. By Wednesday the coordinator is behind and triaging by gut, and the patients who slip are, reliably, the sickest ones.
Nobody in that chain is doing anything wrong. The work is simply larger than the people assigned to it. Always has been.
Can AI voice calls actually close that gap?
Yes, and not because the AI is clever. Because it's available.
The thing an automated follow-up call does that a human call list can't is start on time, every time, for everyone. Discharge hits the record, outreach triggers, the patient hears a voice within 24 hours whether it's Friday night or the middle of August. No backlog. No triage-by-exhaustion. Across HANA's deployments we see roughly 85% weekly engagement against an industry baseline of 15 to 20%, across 1M+ patient interactions with zero critical adverse events.
The interesting part is that patients aren't tolerating it. They prefer it. An AI call at 6pm on a Sunday is less intrusive than a human one at 11am on a Tuesday when you're at work.
What does a good automated follow-up call actually do?
Four things, in order. It reaches the patient inside 48 hours. It asks about symptoms and medications in plain language, in the patient's own language. It books the follow-up appointment inside the same conversation rather than saying "please call to schedule," which just recreates the phone problem you were trying to solve. And it escalates anything worrying to an actual human, fast.
That last one is the whole design philosophy. The AI isn't replacing clinical judgment. It's routing to it. Look at the clinical use cases and the pattern is consistent: automation handles reach, humans handle decisions.
What usually goes wrong with these programmes?
Two things, and I've watched both.
The first is treating it as a dashboard project. Data goes in, nobody's job changes, nothing improves. I learned this expensively in my DTC years, watching a Stripe dashboard hit a million dollars in a single day while the product underneath it was, honestly, bad. Scale hides problems. Metrics hide problems. A number going up is not the same as a thing working.
The second is complexity theatre. When I crossed Australia with a circus, the acts that drew the biggest crowds weren't the technically hardest ones. Fire chains beat aerial silk, every time, because people could follow what was happening. Same in health tech. The follow-up programme that works is the boring one that reaches everybody, not the elegant one that risk-stratifies twelve ways and reaches half. Integration should be boring too.
Key Takeaways
The evidence on post-discharge follow-up is settled and has been for years: contact within seven days, ideally within 48 hours, meaningfully lowers readmission odds, and contact after that window does close to nothing. The failure isn't knowledge, it's throughput. Manual outreach cannot keep pace with discharge volume, so the patients who get missed are the ones least likely to answer, least likely to have transport, most likely to come back through the ED.
Voice AI is useful here for an unglamorous reason. It starts on time. It doesn't get behind. It works weekends. When it's wired properly it books the appointment in the conversation and hands the worrying cases to a person while there's still time to do something. That's not a moonshot. It's a scheduling problem solved at scale, and the ROI math tends to work out around 31:1 for clinics that run it seriously.
Build the boring version. Reach everybody. Then get clever.
FAQ
How soon after discharge should a patient be contacted? Within seven days, and ideally within 48 hours. Research on heart failure patients found follow-up inside seven days associated with 19% lower odds of 30-day readmission, with no significant benefit from later contact. The sooner the better, and weekends count.
Do patients actually engage with AI follow-up calls? More than most people expect. In our deployments weekly engagement sits around 85% compared to a 15 to 20% industry baseline for patient-initiated tools. People answer a call about their own recovery, especially when it arrives at a time that suits them and speaks their language. Our outcomes research goes deeper on this.
Does automated outreach replace nurses and care coordinators? No, and programmes that frame it that way tend to fail. Automation handles the reach layer: dialling everyone, asking screening questions, booking appointments. Clinical staff handle escalations and judgment calls. The gain is that nurses spend their hours on patients who need them instead of on voicemail.
If you're running post-discharge follow-up by hand and losing the window, book a discovery call and we'll look at your numbers together.