A patient calls to move a follow-up appointment. The clinical record knows the appointment, the referring physician and the reason for the visit. Somewhere else in the same organisation, often at a different vendor entirely, the outreach platform knows that person as an email address with an open rate and a campaign history.
Neither system is wrong. Neither knows what the other knows.
That split runs through large provider organisations, and it is not an accident of procurement. Clinical systems of record were built to be authoritative, auditable and slow to change. Marketing platforms were built to be fast and to segment. The two rarely agree on something as basic as which record refers to which human being.
Closing that gap has turned into a live budget question rather than a theoretical one. Health systems are buying customer relationship tools designed for retail and installing them inside an environment governed by federal privacy law. The construct they are importing, the 360-degree customer view, was invented for companies whose main constraint on knowing everything about a customer was cost. In a provider setting the constraint is a statute, and the customer is a patient.
Sai Saketh Sunkara built one of these integrations end to end for a large healthcare system, holding ownership across architecture, integration and analytics on a Salesforce Health Cloud and Marketing Cloud Engagement platform assembled from nothing. By the end, six or more enterprise systems were feeding one model of the patient, among them an Epic electronic health record, a Teradata warehouse, MuleSoft integration services, Data Cloud and Datorama.
The hardest part, by his account, was not the number of systems. It was the age gap between them. Bridging a legacy electronic health record with modern cloud CRM platforms required a custom pipeline built for resilience, because the two sides did not merely store different fields. Their data models were incompatible at the level of what an entity is and how it changes over time, and a pipeline written on the assumption that both ends would eventually agree tends to break on contact with the first real record.
Underneath that sat a problem that sounds administrative and is not. Patient identifiers were inconsistent across the systems. Until those identifiers could be resolved to the same person with enough confidence to act on, a genuine unified view was not available at any price. Identity resolution is unglamorous work, and it is where projects of this kind usually fail quietly, because a false match in a healthcare setting is not a mildly awkward email.
The obvious objection arrives at about this point. If clinical systems and marketing systems are now joined, is clinical information being used to market to patients?
Sunkara's answer is architectural rather than reassuring, which is the more useful kind. Balancing HIPAA-compliant governance against marketing activation, he says, required a carefully segmented architecture that separates clinical and marketing data flows. The shared identity spine lets the organisation recognise a person consistently across the estate. What is permitted to cross between the clinical side and the activation side is deliberately narrow and defined by design rather than by policy memo. The point of the segmentation is that a unified view of who someone is does not imply a unified view of everything known about them.
The approach did not originate on this project either. Sunkara had already published research on building a 360-degree customer view using a cloud data platform, working through the identity resolution and data harmonisation principles in the abstract before he met them in a regulated production environment. Practitioners far more often ship first and write it up later, if they write it up at all. Running the sequence in the other direction meant the architecture arrived carrying a stated method, rather than being reverse-engineered afterwards from whatever survived the build.
On results, he puts the operational figure at a 25 percent reduction in costs, achieved by consolidating patient data management and outreach onto the single platform. He also reports a 15 percent improvement in patient adherence, which he attributes to personalised communication journeys running on the unified data model. That second number is his own operational measure rather than an independently reviewed one, and it is worth separating from the first. Adherence is a clinical outcome, and a claim that outreach moved it is the sort of thing that would ordinarily rest on a controlled study with a defined measurement window rather than on a platform count.
Where he expects the field to go next is predictive. Segmentation driven by artificial intelligence, paired with predictive analytics, will in his view push health systems away from reacting to patients who have already dropped out of contact and toward what he calls "catching care gaps before they widen." That is a considerably more sensitive capability than a reminder email, and it lands on precisely the boundary his segmented architecture was built to hold. Anticipating who is about to disengage from their own care requires exactly the sort of clinical signal that governance is designed to keep on one side of the wall.
The broader shift he describes is organisations moving toward unified platforms that treat "patient engagement as part of care delivery, not a separate function." Whether that reads as reassuring or as alarming depends entirely on where the walls sit and who checks them. Engagement counted as care is a stronger mandate for outreach teams and a stronger obligation on the people designing the data flows beneath them.
That leaves the field with a hiring problem as much as an engineering one. Sunkara argues that people entering this work need fluency across compliance, clinical data structures and modern cloud architecture, not CRM skills alone, because translating a research-grade framework into a regulated system at scale is where those disciplines have to meet in practice. Provider organisations have no shortage of people who can build a communication journey, and no shortage of people who can read a privacy rule. The scarce ones are those who can do both at the same time without pretending the other does not apply.