April 3, 2026
By Jay Chowdappa, MD, co-founder and CEO, Zynix AI · Updated October 1, 2026
AI agents help ACOs reduce post-discharge readmissions by making follow-up happen for every discharge: they pick up discharges from ADT feeds, rank patients by risk, run outreach and scheduling, and route red-flag answers to nurses. Licensed clinical staff still make the TCM interactive contact within two business days and own every clinical decision.
Let me be direct: if your ACO cannot systematically prevent avoidable 30-day readmissions, you are leaving shared savings on the table. Period.
Here is the math. CMS penalizes hospitals with higher-than-expected readmission rates by reducing Medicare payments by up to 3% for an entire fiscal year under the Hospital Readmissions Reduction Program (HRRP). For FY 2026, many hospitals face some level of penalty. But the penalty itself is not the real problem for ACOs. The real problem is that every avoidable readmission inflates your total cost of care, erodes your benchmark margin, and pushes you further from the shared savings threshold.
A single preventable readmission for a heart failure patient is expensive. Multiply that across a 20,000-beneficiary ACO with even a modest excess readmission ratio, and the avoidable spend adds up quickly. That is the difference between earning shared savings and writing a check back to CMS under two-sided risk.
Every ACO leader knows the playbook: discharge follow-up calls within 48 hours, medication reconciliation, PCP appointment scheduling within 7 days. The evidence is strong. In one study of community nurse follow-up, the readmission rate was 9.24% when a post-discharge contact attempt was made, compared with 15.67% when none was (adjusted OR = 1.93). In a nurse-led discharge phone call program across an integrated health system, 7-day readmission rates for contacted patients were 2.91% versus 4.73% for those not contacted.
The problem is not the protocol. The problem is execution at scale.
A typical ACO care coordinator manages a large panel of patients. Post-discharge follow-up requires contacting the patient within 48 hours, reviewing discharge instructions, reconciling medications, scheduling follow-up appointments, coordinating with home health if needed, and documenting everything in the EHR. That takes meaningful coordinator time for every patient, even when everything goes smoothly. It rarely goes smoothly.
Patients do not answer phones. Discharge summaries arrive late or incomplete. PCP offices have no availability for 10 days. The SNF did not send the medication list. Each of these failures cascades into a missed intervention window, and missed windows become readmissions.
This is not a training problem or a motivation problem. It is a throughput problem. And you cannot solve throughput problems by hiring more coordinators when your shared savings margin is already thin.
The second failure mode is data fragmentation. Your ACO has beneficiaries across multiple hospitals, SNFs, home health agencies, and physician practices. Discharge events from non-owned facilities often arrive late via ADT feeds. By the time your team knows a patient was discharged, the two-business-day TCM contact window is already closing or closed.
Even when you get timely ADT alerts, the data is rarely actionable on its own. You need the discharge summary, the medication list, the follow-up orders, and the patient's historical risk profile synthesized into a single view. That synthesis step is manual, slow, and error-prone.
CMS is not making this easier. Three regulatory developments are compressing the readmission problem:
Historically, HRRP assessed readmission rates using only Medicare Fee-for-Service data. In the FY 2026 IPPS final rule, CMS finalized adding Medicare Advantage data to all six readmission measures. For ACOs with significant MA-aligned beneficiaries, this changes the denominator substantially. A JAMA Network Open cohort study of 3,203 hospitals estimated that accounting for Medicare Advantage penetration would redistribute $284 million to $297 million in penalties annually.
CMS has finalized shortening the HRRP performance period from three years to two years. This means poor performance quarters hit your penalty calculation faster, and recovery from a bad stretch takes longer proportionally.
The Transforming Episode Accountability Model (TEAM), which launched in January 2026, creates mandatory bundled payments for five surgical episode types with 30 days of post-acute care. For hospitals in TEAM, a readmission within the episode window counts against the episode target price, and ACOs whose patients have those surgeries share the exposure. This layers additional readmission risk on top of existing MSSP accountability.
This is where I need to be precise about what AI actually does here, because the industry is drowning in vague claims about "AI-powered care coordination." What matters is not the AI. What matters is whether the system can reliably execute the post-discharge workflow at scale, without depending on human throughput as the bottleneck.
At Zynix, we built our platform around a multi-agent architecture specifically because post-discharge workflows require coordinating multiple tasks across multiple systems simultaneously. A single monolithic model cannot do this well. Here is how it works in practice:
The first agent monitors ADT feeds, claims data, and EHR discharge events continuously. When a discharge is detected, it immediately pulls the patient's longitudinal record, calculates a readmission risk score using clinical and social determinants, and classifies the patient into an intervention tier. High-risk patients (CHF, COPD, post-surgical, dual-eligible) get flagged for immediate outreach. This happens as soon as the discharge event arrives, not days later.
The second agent initiates structured outreach within hours of discharge. It confirms the patient is home with their discharge paperwork, collects the medication list for a nurse to reconcile, asks the red-flag questions your clinicians have approved, and confirms follow-up appointments. Any red-flag answer or symptom question goes straight to a nurse by rule; the agent does not assess it. For patients who do not respond to the first attempt, it executes a multi-channel escalation protocol: phone, text, patient portal message, and if needed, flags the case for a human coordinator. The billable TCM interactive contact stays with licensed clinical staff: CMS requires it within two business days of discharge for CPT 99495 and 99496, made by clinical staff who can address the patient's status and needs.
The third agent works the provider side. It identifies the appropriate PCP or specialist for follow-up, checks appointment availability, schedules the visit, and sends the relevant clinical summary to the receiving provider. If the patient was discharged from a non-owned facility, this agent reconciles the external discharge data with your ACO's care plan.
The fourth agent monitors for exceptions: patients who report worsening symptoms, missed follow-up appointments, medication access issues, or social determinant barriers (no transportation, food insecurity, caregiver unavailability). These get routed to human care managers with full context, so the coordinator spends their time on complex problem-solving rather than data gathering and phone tag.
I am not going to give you theoretical projections. Here is how our customers are actually using these workflows.
At Palm Beach ACO, our agents run post-discharge follow-up and annual wellness visit outreach by voice and SMS, and the wellness visits feed the risk stratification and transitional care management workflows. When you identify high-risk patients before they end up in the hospital, you reduce the downstream readmission burden. AWV completion is the upstream lever that most ACOs underinvest in.
The same approach applies to HEDIS care gaps, including the follow-up after hospitalization measures that directly map to readmission prevention. The system identifies patients with open care gaps post-discharge so those gaps get addressed during the transition period rather than falling through the cracks.
For a growing multi-site network, the aim is to scale these workflows without scaling headcount at the same rate. That is the core economic argument: you need readmission prevention programs that scale with your beneficiary population, not with your payroll.
And for dual-eligible populations, the readmission challenge is compounded by social determinant complexity. Agents can carry the extra outreach, such as asking about transportation, food and housing barriers and booking behavioral health follow-up, and route each need to the coordinators who handle Medicaid benefits, LTSS referrals and community resource connections.
If you are an ACO leader reading this, here is what I would prioritize:
If you are not receiving timely ADT notifications from your major discharging facilities, nothing else matters. CMS requires hospitals to send electronic admission, discharge and transfer notifications to a patient's other providers, but "having" those feeds and having them operational with sub-1-hour latency are different things. Audit your feeds. Measure the gap between discharge time and notification time. If it is over 4 hours, you have a structural problem.
Not every discharged patient needs the same intervention intensity. Your high-risk patients (top 15-20%) need agent-driven outreach within 4 hours. Your moderate-risk patients need structured contact within 48 hours. Your low-risk patients need automated check-ins. If you are applying the same protocol to everyone, you are wasting resources on low-risk patients while under-serving high-risk ones.
Readmission rate is a lagging indicator. The leading indicator is time-to-first-contact post-discharge. Track it. If your median time-to-first-contact is over 48 hours, you already know your readmission rates will be above benchmark. Set a much tighter target for high-risk patients.
The math is straightforward. One AI agent system can handle the initial outreach, data synthesis, and coordination tasks for a whole discharge list at once. One care coordinator cannot. Use your coordinators for what they are uniquely good at: complex clinical judgment, motivational interviewing, and navigating ambiguous situations. Use agents for the outreach, scheduling, reminders and documentation prep around them.
Before co-founding Zynix AI, I led ACOs that generated $300M+ in shared savings. The single biggest variable in whether an ACO earns shared savings in any given performance year is avoidable acute utilization, and readmissions are the largest controllable component of avoidable acute utilization.
The ACOs that figure out how to execute post-discharge workflows at scale, reliably, for every patient, every time, will earn shared savings. The ones that depend on manual processes and hope their coordinators can keep up will not. That is not an AI pitch. That is an operational reality.
CMS is tightening the screws: shorter measurement windows, broader readmission measures, layered episode accountability through TEAM. The margin for error is shrinking. The organizations that invest in the right infrastructure now will compound their advantage over the next three to five years. The ones that wait will find the gap increasingly difficult to close.
They close the execution gap. Agents pick up each discharge from ADT feeds, rank the patient by readmission risk, call or text to confirm the patient is home, book the follow-up visit and send reminders, and route any red-flag answer to a nurse. Coordinators then spend their time on patients with complex needs instead of phone tag. Care plans for post-discharge TCM set the steps and owners.
Licensed clinical staff. For CPT 99495 and 99496, CMS requires an interactive contact with the patient or caregiver within two business days of discharge, made by the billing practitioner or by clinical staff who can address the patient’s status and needs. Agents can make the first outreach attempt, confirm the patient is home, book the visit and prepare documentation, then hand off to the nurse.
In the FY 2026 IPPS final rule, CMS finalized adding Medicare Advantage data to the six HRRP readmission measures and shortening the performance period from three years to two. The payment reduction is still capped at 3 percent of a hospital’s base operating DRG payments. ACOs feel the effect through partner hospitals and through the readmissions in their own total cost of care.
Treat time to first contact after discharge as the leading indicator, along with the share of discharges reached, interactive contacts completed within two business days, follow-up visits kept within 7 or 14 days, and escalations resolved. Readmission rates by diagnosis and facility come later and confirm whether the process works. Review results by risk tier, since high-risk patients need faster and more frequent contact.
No. Agents take the repeatable volume: outreach attempts, scheduling, reminders, data gathering and documentation prep. Coordinators and nurses keep complex problem-solving, motivational conversations and every clinical judgment, and they receive escalations with the context already gathered. Our overview of how AI agents work with your care team shows the handoffs.
CMS: Hospital Readmissions Reduction Program (HRRP). https://www.cms.gov/medicare/payment/prospective-payment-systems/acute-inpatient-pps/hospital-readmissions-reduction-program-hrrp
CMS: FY 2026 Hospital Inpatient Prospective Payment System (IPPS) and LTCH PPS Final Rule, CMS-1833-F (fact sheet, July 31, 2025). https://www.cms.gov/newsroom/fact-sheets/fy-2026-hospital-inpatient-prospective-payment-system-ipps-long-term-care-hospital-prospective-0
Future Healthcare Journal: Vernon D, et al. Reducing readmission rates through a discharge follow-up service (2019). https://pmc.ncbi.nlm.nih.gov/articles/PMC6616175/
Journal for Healthcare Quality: Lukanski A, et al. Implementing a Discharge Follow-up Phone Call Program Reduces Readmission Rates in an Integrated Health System (2023). https://pubmed.ncbi.nlm.nih.gov/37788411/
JAMA Network Open: Hospital Readmission Reduction Program Penalties for Hospitals With High Medicare Advantage Penetration. https://pmc.ncbi.nlm.nih.gov/articles/PMC12828625/
CMS: Transitional Care Management Services (MLN908628, August 2025). https://www.cms.gov/files/document/mln908628-transitional-care-management-services.pdf
CMS: TEAM (Transforming Episode Accountability Model). https://www.cms.gov/priorities/innovation/innovation-models/team-model
CMS: Interoperability and Patient Access final rule (CMS-9115-F) fact sheet. https://www.cms.gov/newsroom/fact-sheets/interoperability-and-patient-access-fact-sheet
Jay Chowdappa, MD is co-founder and CEO of Zynix AI. He is a physician, and ACOs led by him generated $300M+ in shared savings before he started Zynix AI to help care teams follow through on what their data already shows.