Your Post-Discharge Follow-Up Calls Aren't Happening. Let's Be Honest About Why.

I watched a Stripe dashboard cross $1M in a single day once. Different life, different company, direct-to-consumer, and I remember standing there feeling like a genius.

The product was bad.

Not catastrophically bad. Just quietly, structurally bad in ways that a big enough number at the top of a screen will hide from you for about eighteen months, until it doesn't, and then you're in a room with your team explaining why churn ate everything. Scale hides problems. It never fixes them.

I think about that whenever a clinic tells me their follow-up program is fine.

Why do post-discharge follow-up calls get skipped?

Because nobody has the hours, and the hours are the only input that matters. Your MA has a full panel, your nurse is triaging inbound, and the discharge callback list is the thing that gets done if the day goes well. Most days don't go well.

So the list gets triaged silently. Highest acuity gets called, everyone else gets a portal message they won't read. Plivo's write-up on post-discharge voice AI puts the mechanism plainly: limited nursing time, inconsistent call coverage, and delayed escalation, all landing exactly when patients are most vulnerable.

That's not a staffing failure. It's arithmetic.

What does a voice AI follow-up call actually do?

It calls every discharged patient on schedule, asks the questions your protocol already specifies, and escalates the ones who answer badly. That's it. It's not diagnosing anything.

In practice a call runs three to five minutes. Are you taking the medication as prescribed. Any fever, any drainage from the site, any pain worse than yesterday. Do you know when your follow-up appointment is, and do you have a ride. Red flags route to a human the same hour, and everything else lands in the chart as a structured note.

The specific clinical use cases matter more than the technology does. A post-op orthopaedic call and a CHF weight check are not the same conversation and shouldn't share a script.

Will patients actually pick up the phone for AI?

Yes, and this genuinely surprised me the first time. I'd built an app for bipolar patients before this and got 15% engagement. Then we started calling patients with AI instead of asking them to open something, and weekly engagement hit 85%.

Eighty five against an industry baseline of 15 to 20%.

The reason is boring: answering a phone requires no learning. No download, no password, no reading. My daughter is 10 and she can't be bothered with most software, but she'll answer a call. Your 78 year old post-op patient is the same, for entirely different reasons.

The recent multi-site virtual nursing study in npj Digital Medicine found something adjacent and striking: across nine hospitals, virtual-nurse-assisted discharges had 3.7% 30-day ED readmission against 13.3% for traditional in-person discharge, at matched baseline risk. Human presence at the right moment, delivered remotely. Voice is the cheapest way to buy more of those moments.

What should a clinic measure in the first 90 days?

Three things, and none of them are minutes saved. Contact rate, which is what percentage of discharged patients you actually reached. Escalation yield, which is how many red flags surfaced that your staff would otherwise have found in the ED. And no-show rate on the follow-up appointment, because that's where the revenue lives.

We've published what those curves looked like across deployments in our case studies, including the ones where month one was mediocre and month three was the point.

Across 1M+ patient interactions in 5 countries and 3 languages, zero critical adverse events. I say that number carefully, because in this business the absence of a bad headline is the actual product.

How do you start without ripping out your EHR?

You don't rip anything out. The calls read from your existing discharge list and write structured notes back, which means the integration surface is small and boring on purpose. Setup detail sits in our technical docs.

Start with one cohort. One procedure, one discharge pathway, thirty days. Measure the three numbers above against your own baseline, not against a vendor's slide.

And run the money honestly before you sign anything, because 31:1 ROI is a real number and it's also an average that hides variance. Clinics with high no-show costs and high readmission exposure clear it easily. Others take longer. The pricing math is public for exactly this reason. We're fully open-source and self-hosted with no OpenAI dependency, which matters less for your ROI and enormously for your compliance officer. That decision, and why we made it, is part of why HANA exists at all.

Key Takeaways

Your follow-up program isn't broken because your staff don't care. It's broken because the number of patients needing a call exceeds the number of calls a human panel can place, and every clinic solves that gap by quietly narrowing who gets contacted. Automation doesn't make your team faster, it removes the triage decision entirely.

The channel is the whole game. Portals and apps ask patients to learn something at the worst moment of their year, which is why engagement lands near 15%. A phone call asks nothing, which is why it lands near 85%. Start with one discharge pathway, measure contact rate and escalation yield and no-show rate for ninety days, and let those three numbers argue for the rest. Don't let a big number on a dashboard convince you the underlying thing is working. I've made that mistake with actual money.

FAQ

Do patients need to consent to an AI follow-up call?

Yes, and it should be explicit at discharge alongside the rest of your communication consent. The call also identifies itself as automated at the start. Patients who prefer a human get routed to one, and in practice very few ask.

What happens if a patient reports something urgent on the call?

The call escalates to your on-call clinician within the same hour, with the flagged responses attached. The system never gives clinical advice or changes a care plan, it collects, flags, and hands off.

How long before a clinic sees a measurable difference?

Contact rate moves in week one because coverage is immediate. No-show and readmission effects need a full ninety days to separate from noise. If you want to walk through what that would look like against your own volumes, book a slot on my calendar.