A patient goes home with a stapled packet, a new prescription and a warning to call if anything feels wrong. By dinner, the instructions are fuzzy. By morning, the incision is warm. The hospital has already moved on to the next bed. This is the small, hazardous gap CipherHealth has spent 17 years trying to close.
The New York software company began in 2009 with an unglamorous idea: contact every discharged patient, automatically, then send the concerning answers to a person who can help. Today that idea has widened into an enterprise platform for outreach and rounding. CipherOutreach handles appointment reminders, pre-care preparation, post-discharge check-ins and ongoing monitoring by voice, text and email. CipherRounds gives nurses and leaders structured programs for checking on patients, staff and hospital spaces. An integration layer moves the resulting answers into electronic health records and work queues.
That makes CipherHealth easy to misunderstand. It is not primarily a survey company, though it gathers feedback. It is not an EHR, though it writes into Epic and other systems. And it is not simply a chatbot, despite its new conversational AI. Its real product is the handoff: a patient's sentence becomes a tagged issue, the issue gets an owner, and the owner has a clock running.
The $80 nobody wanted to spend
The cleanest example comes from UCSF Health. In a published video, a population-health executive put the old cost of a post-discharge call at $82. Automation brought the figure down to $2. Another UCSF program using CipherHealth reached 91 percent of a 1,300-patient bundled-payment cohort, with a median care-coordinator follow-up time of three hours and reported savings of $959,000.
The arithmetic is seductive, but the operating change matters more. A nurse no longer spends equal time dialing patients who are recovering normally and those who cannot fill a prescription. Software attempts the broad contact. Patients answer a short set of clinically approved questions. Routine responses close automatically; pain, medication trouble, confusion or worsening symptoms land in the appropriate queue. The expensive human minutes are reserved for ambiguity and risk.
“I'm most proud of being able to reach families in their primary language.”Anu Asnani, Seattle Children's Hospital
Seattle Children's shows what that model looks like when the first design is not good enough. Its program began with English and Spanish calls. When the hospital's data showed higher risk among Vietnamese- and Somali-speaking families, those languages were added. Automated calls were followed by another attempt and a text; clinical concerns went to one of four rotating transitional-care nurses, while nonclinical coordinators handled routine problems.
Between January 2023 and December 2024, the hospital reported a 72 percent engagement rate on clinical questions, a 36 percent reduction in pediatric readmissions, more than 2,700 inpatient days avoided and estimated annual savings of $6.9 million. Those figures come from one program, not a randomized trial. Still, the copyable part is concrete: start broad, study who is not being reached, add language and channel choices, then separate clinical from nonclinical work.
The operating loop behind the software
Rounding is not the same as resolving
Inside the hospital, CipherRounds attacks a related leak. Rounding is standard practice: a leader visits a patient, a manager checks in with staff, a facilities team inspects a room. The old process often ended on paper, in a spreadsheet or in somebody's memory. CipherHealth standardizes the questions, flags a concern in real time and routes it to food service, nursing, facilities or another accountable team.
The company now groups patient, staff and location rounds in one product. Patient rounding focuses on experience, safety and service recovery. Staff rounding looks for burnout and retention risk. Location rounding tracks equipment, hygiene and regulatory tasks. New ambient-listening software can capture a bedside conversation and draft structured documentation while the nurse remains with the patient. AI summaries, sentiment analysis and a three-month interaction timeline are intended to make the history legible before the next conversation.
The distinction from competitors is less about any single feature than the breadth of the loop. Patient-engagement vendors such as Get Well, Artera, Press Ganey, Luma Health and Innovaccer overlap with parts of the product. EHR vendors supply native messaging. Hospitals can also assemble call-center tools, forms and spreadsheets. CipherHealth's counterargument is accumulated workflow: more than 15 years of scripts, integrations and escalation patterns, with rounding and outreach sharing context.
First the cloud failed. Then the server needed a hammer.
CipherHealth's infrastructure history is unusually candid. CTO Nate Perry-Thistle has said the company started in the cloud, but the early arrangement became cost-prohibitive and difficult for compliance. It moved into colocated facilities, acquiring the racks, physical machines and headaches that came with them. During one failure, a facility tried what he described as a ball-peen-hammer restart. He called the CEO and warned, “this might be the one.”
The company later moved toward modern cloud infrastructure, using Kubernetes and Google Cloud tooling while keeping an eye on customer trust and portability. The episode explains something about the current sales pitch. Healthcare buyers do not just buy a clever interface. They ask where data lives, how alerts are audited, whether messages are delivered and what happens when a dependency fails. CipherHealth now advertises HIPAA-compliant AI architecture, HITRUST-aligned controls, HL7 and FHIR connections, and bidirectional links with Epic, Oracle Health, MEDITECH and dozens more.
Its 2018 strategic investment from JMI Equity was reported by company databases at $37.5 million, though CipherHealth itself did not disclose the number. Databases put total financing around $55.5 million before later undisclosed debt and venture rounds. Pricing remains private and enterprise-specific. A Frost & Sullivan report described products licensed à la carte, which fits a modular sale: start with outreach or rounding, integrate it, then expand programs.
The founder returns to a harder AI market
Randy Cheung, Zachary Silverzweig, Alex Hejnosz and Darshan Bachhawat are listed as founders in company databases. CipherHealth's current leadership pages foreground Cheung, now executive chairman, and Silverzweig. Cheung's original insight came from seeing a communication gap between caregivers and patients. Silverzweig brought a more personal encounter: his own cancer treatment showed him how easily follow-up and coordination can fracture, even in respected systems.
Silverzweig returned as chief executive in May 2026 after having led product, engineering and operations earlier in the company's life. The timing says plenty. Hospitals are surrounded by AI demonstrations and have become selective about what reaches production. CipherHealth is not trying to win with a free-floating medical oracle. It is inserting AI into workflows customers already run: summarize recent interactions, let patients answer in natural language, prioritize risk, document rounds and preserve a governed escalation path to a clinician.
That is a disciplined place to use AI. A response such as “my incision feels warm” carries more signal than pressing 1 for yes. But the model should not improvise a diagnosis. It should capture the language, map it to an approved protocol, write the context into the record and bring in a trained person. The useful automation is not the one that sounds most human. It is the one that knows when to summon one.
When the playbook breaks
- No accountable owner: a faster alert is still neglect if nobody is staffed to receive it.
- Bad contact data: automation cannot reach a disconnected number or infer a patient's preferred language.
- Loose integration: forcing staff to toggle between systems recreates the burden the platform promises to remove.
- Unapproved scripts: clinical committees must define questions, thresholds and escalation paths before launch.
- No baseline: without reach, callback, readmission and experience measures, a hospital cannot tell whether the program changed anything.
What another operator can copy
The transferable lesson is not “buy more software.” It is to design communication as an operating system. Choose one expensive gap, such as post-discharge calls or unresolved inpatient complaints. Define a baseline cost and outcome. Automate the repetitive contact, not the clinical judgment. Give every exception a named team, urgency level and closure rule. Feed the result into the record people already use. Then inspect the misses: the languages, channels and populations that do not respond.
CipherHealth's customer evidence is strongest when the organization around the product is visible. Intermountain paired automatic calls to every discharged patient with a 24-hour nurse advice team. Seattle Children's divided routine coordination from clinical escalation. UCSF used automation within a population-health program rather than treating it as a marketing channel. The software created leverage because someone redesigned the work.
The company says its platform has now seen more than a billion encounters, 200 million interactions and 17 million resolved issues. Those numbers may give its AI more context, but scale alone is not the achievement. A hospital wins when the obscure sentence, the missed prescription or the warm incision reaches the right person before it becomes tomorrow's readmission. The follow-up call got cheaper. The handoff became the business.