Mayank Aggarwal, Business Analytics Director at Evernorth Health Services
Artificial intelligence in healthcare has reached a new level of maturity. Predictive models can now detect risks much earlier and help flag high-risk patients before they even come to clinic. Dashboards reveal patterns and anomalies in data that previously only analysts could unearth using weeks of work, and AI algorithms are able to scan diagnostic images consistently with a very high degree of reproducibility that enable faster decision-making for clinical review. Advanced AI tools are currently aiding in the interpretation of different types of diagnostic imaging techniques such as PET scans, MRIs, and X-rays, thereby, doctors can go through diagnoses much quicker and be more confident of their findings. A change in the direction that would be of benefit to such a sector where time is often a factor influencing the result.
Inability to bridge the insight-action divide
Even with these advanced tools available, it is not automatically true that the right decisions and actions will be made. A model might point out a high-risk condition in a patient, a dashboard might show a cost anomaly, a generative AI app might summarize a mountain of physician notes in a blink. However, the harder bit is ‘doing something about it. Insight generation may be the fastest-growing capability in healthcare AI but knowing how to act on it, at the point of care, with the right person, in the right moment, still lags badly behind. Closing that gap is one of the most consequential opportunities left in the field.
Why context and workflow matter as much as accuracy
The reason this happens is not because these models aren’t smart enough, it’s because they are detached from the context of the clinicians using them. A risk score delivered out-of-band and unrelated to the clinician’s current task is going to get ignored, regardless of the accuracy of the information.
The most effective AI will always be the one that meets the user where they are in their workflow. If a nurse has to open up an entirely different application to see an AI-driven risk score, that’s one more step they might skip. However, if that same risk score shows up in the claims or care management software they use every day, there’s a far greater chance they’ll actually review it. At the point of adjudication, a claims professional is going to notice alerts and scoring relevant to the case they’re already working on. They’re more likely to consider the information and adjust their ruling based on it. Put in another part of the application, or in a completely different application, and the information might as well not be there.
The same goes for human judgment. While it may seem counter intuitive, humans need to have a role in the decision-making process. A care manager reviewing a claim might have a far greater understanding of context than the machine learning model which generated a risk score or flag. People can account for nuance that an algorithm can’t, and human judgement is often necessary to achieve the desired outcome. The companies that are doing things right aren’t trying to eliminate the human touch, they’re trying to put it closer to the point of action.
Decision intelligence as the missing framework
This is where decision intelligence comes in and offers a practical way forward. By framing models, dashboards, and algorithms as inputs to a particular human decision, decision intelligence focuses on outcome: What action should be taken based on the insight? Who should take it? And how can the system best present information to these decision makers? This viewpoint turns AI into a decision-making production system – a single framework encompassing analytics, operations, and human factors.
Building decision intelligence into everyday operations
Building decision intelligence into everyday operations doesn’t require abandoning existing systems of wholesale. It’s often a matter of starting small: looking at where insights already come into the process and realizing where a decision is actually being made. Then, it’s a matter of redesigning those systems so that the outputs of decisions are folded into the existing workflow, rather than existing alongside it. Likewise, capturing the lessons within the broader organization through knowledge management or publishing research on decision intelligence within healthcare will allow disparate experiments to generate something greater than the sum of their parts. As these capabilities develop within global centers of excellence and within healthcare-specific domains, this virtuous cycle will kick in, and day-to-day operations will see more of the cutting-edge work being done within AI healthcare systems.
The road ahead
The next chapter of healthcare AI will not be defined by the most sophisticated model, but rather by the most productive path from any insight to action. This is a challenge that favors thoughtful analysis over sensationalism, and it favors the team that has spent most of their time studying their own decision-making. For an industry built on high-stakes decisions, that focus on decision intelligence may be exactly the discipline needed to turn today’s impressive AI capabilities into tomorrow’s better outcomes.
