In healthcare, AI is increasingly being asked to do more than summarize information or answer questions. It’s moving beyond those routine tasks to schedule appointments, route patient requests, provide follow-up instructions, and flag issues such as incomplete referrals or pending authorizations for the appropriate team.
As AI’s role in healthcare continues to expand, the context surrounding each action becomes just as important as the response that the technology generates.
That context is rarely contained in one place. For example, a scheduling system may show that an appointment is confirmed, even though an authorization workflow still shows an unresolved issue. Or a patient’s current language preference may be documented in one system, while an outreach platform references an older record and uses the wrong language. These systems may be accurate within their own boundaries, yet an AI tool acting on only one of them can still produce an incomplete next step.
This challenge becomes more important as healthcare organizations look to move AI into everyday operations. Deloitte’s 2026 Global Health Care Outlook found that about 30% of health systems were operating generative AI in select areas, but only 2% had deployed AI across the entire enterprise.
As AI takes on a larger role in these workflows, CIOs will need to evaluate not only whether an AI system works but also whether it has sufficient context to act appropriately across the patient journey.
Records are only part of the patient’s current story
Healthcare’s digital foundation has come a long way. According to the Office of the National Coordinator for Health Information Technology, more than 99% of non-federal acute care hospitals have adopted certified electronic health records. ONC also reported that by 2025, 76% of hospitals were participating in all four interoperability domains it measures: sending, receiving, finding, and integrating electronic health information.
In practical terms, that means many hospitals have built the ability to move patient information across systems and make it available where it’s needed.
The challenge now is making sure AI can understand what’s happening with the patient in that moment, so someone isn’t given outdated instructions, an incorrect status update, or guidance that no longer reflects where they are in their care journey.
I think of this challenge as a context gap. The data may exist somewhere in the organization or within other platforms, yet the AI may still lack the combination of identity, history, workflow status, preferences, permissions, and recent activity needed to understand what’s happening now.
When that gap goes unnoticed, automation can carry an incomplete understanding further into the patient journey.
The faster AI acts, the more important context becomes
Fragmentation has always created operational friction. For example, consider a patient trying to schedule a routine screening. They may submit a request online, call when they don’t receive a clear next step, have to repeat the same information to a staff member, and then follow up again if the referral or authorization status is unclear.
On the staff side, each of those touchpoints requires someone to check multiple systems, reconcile conflicting information, and manually close the loop.
AI changes the speed and scale at which those fragmented processes occur—and worsen.
An automated system can send appointment reminders, confirm or reschedule visits, provide preparation instructions, route referral questions, or flag an unresolved authorization issue for follow-up within seconds. If the context behind that decision is out of date or incomplete, the same speed that creates efficiency can also rapidly spread confusion across channels and teams.
I believe that’s why AI evaluation must look beyond whether a model produces an accurate response. Healthcare leaders also need to understand how the technology fits into real workflows, what information it relies on, and how its decisions are monitored over time.
Three context tests before scaling
AI can only act appropriately when the surrounding data, workflows, and oversight are strong enough to support it. Below are the three questions CIOs should ask to determine whether an AI system has enough context to take the right next step across patient-facing workflows and where gaps are most likely to emerge.
-
Can the AI identify the right patient information across the journey?
Identity is the starting point for context. CIOs should understand how patient information is matched, updated, and governed across every system involved in the workflow, including what happens when records conflict or a caregiver is authorized to act on a patient’s behalf.
Difficult cases are especially useful to test. If two systems contain different phone numbers, which source takes precedence? If a demographic field changes, how quickly does that correction reach the AI? These exceptions reveal whether the technology can maintain a reliable view of the patient and their preferences.
-
How current is the AI’s understanding of the patient?
A patient’s situation can change quickly. An authorization may clear, a referral may move forward, an appointment may be rescheduled, or a staff member may resolve an issue manually. Even if the AI has access to the right systems, it can still give outdated information if those changes haven’t reached it yet.
This is where timing becomes critical. CIOs should understand how quickly important updates become available to the AI and which systems it relies on to determine what’s happening now. How soon after an authorization clears does the AI see that change? Can the AI recognize when information may be out of date and route it for review before responding to the patient?
-
Can the AI close the loop when the situation changes?
Some AI tools are designed primarily to answer questions or provide information, such as explaining preparation instructions or sharing the status of an appointment. Others may be authorized to take the next step, such as rescheduling a visit, routing a referral issue to the appropriate team, or sending a required form to the patient. Once AI begins taking those actions, CIOs need to understand what the system is allowed to do on its own, when it should stop, and when a person needs to step in.
If an AI assistant discovers that a referral is incomplete, can it route the issue to the appropriate team with the relevant history attached? If a patient reschedules, do other systems tracking appointments receive the update?
After deployment, measure whether AI is closing the loop
Once an AI tool is deployed, organizations often track metrics such as adoption, response accuracy, utilization, or time saved to understand whether it’s working as expected. Those metrics are useful, but they only capture the interaction itself. For patient-facing AI, CIOs also need to look at what happened afterward and whether the interaction helped move the patient’s care forward.
Did a scheduling interaction lead to a completed appointment? Did the referral reach the appropriate next step? Did the AI resolve the request, or did the patient contact the organization again through another channel? When the AI escalated an issue, did the receiving team have enough information to act without asking the patient to start over?
These considerations track the state of the journey beyond the AI interaction itself, giving CIOs a clearer view of whether automation is helping preserve continuity throughout the patient journey.
AI readiness depends on context readiness
Healthcare organizations have spent years making it easier to capture and share patient information across systems. As AI takes on a larger role in patient-facing workflows, the next challenge is making sure it can use that information in the right context, at the right time, to help move the patient forward and receive the care they need.
For CIOs, the practical test is whether an AI system can understand who the patient is, what’s changed in their journey, what remains unresolved, and who needs to act next. That context will shape whether automation carries the journey forward with continuity or leaves the patient to connect the pieces.
