1Principal Cloud Architect - Optum, Eden Prairie MN, USA. Email: ashishkumarmanne@gmail.com
2Lead Software Engineer - Optum, Mckinney TX, USA. Email: mohinijindal28@gmail.com
3Sr Manager Quality Engineering - Optum, Ashburn VA, USA. Email: sandeepsaini1176@gmail.com
4Cloud Development Manager - Little Elm TX, USA. Email: Abinayasiva88@gmail.com
5AI/ML VALIDATION SPECIALIST - ISOFTECH INC, CHANTILLY VA, USA. Email: suvega85@gmail.com
6Independent Researcher - Aubrey TX, USA. Email: pcjakku@gmail.com
*Corresponding Author: Tirumala Ashish Kumar Manne, Principal Cloud Architect - Optum, Eden Prairie MN, USA. Email: ashishkumarmanne@gmail.com
Background: Healthcare enterprises are being more reliant on data-driven decision-making to handle the large-scale clinical and administrative operations of appointment adherence management, utilization review, care coordination, and claims processing. Conventional batch-driven analytics systems add latency, lower workflow responsiveness, and have low scalability with workloads of the scale of the enterprise. The opportunity to operationalize real-time decision intelligence in healthcare settings is provided by the emergence of cloud-native, event-driven architectures and deployable machine learning (ML) systems.
Objective: The objective of this research was to design, deploy and test a cloud-native, real-time AI-based decision intelligence system with the capability of enhancing operational activities, predictive precision, and infrastructure scaling of enterprise health care processes. To measure performance improvement, the system was compared to a legacy system that operated on a batch-processing basis.
Methods: An event-driven, distributed architecture was created that encompassed streaming data ingestion, real-time feature engineering, containerized ML inference services, workflow orchestration, and enterprise level observability controls. The data used to model workflows related to enterprise appointment management was an open-source healthcare operations dataset (medical appointment no-show dataset; 110,527 records). The computations of real-time features were based on scheduling and clinical variables and a supervised ML classifier was deployed as a scalable inference microservice to forecast no-show probability. The proposed real-time architecture and a batch-based rule engine were compared based on operational cycles of the same kind. The key performance indicators were the decision latency (milliseconds), the workflow throughput (decisions/hour), predictive discrimination (area under the receiver operating characteristic curve [AUC], precision, recall, F1 score), infrastructure utilization efficiency, and cost per 10,000 decisions processed. The stress test was conducted at 150 per cent of the estimated peak load to determine the stability of the systems.
Results: Compared to the baseline of a batch AI architecture, the real-time AI architecture achieved a 52% (1,800 ms vs 860 ms) reduction in median decision latency, and a reduction in 99th percentile latency by 61%. The workflow throughput was also 41 percent (120,000 vs. 169,000 decisions/hour). The predictive performance was also enhanced, as AUC (0.74) to (ML model) was 0.88, and the precision score (0.62) to (0.79), recall (0.58) to (0.76), and F1 score (0.60) to (0.77). Auto-scaling infrastructure minimized the cost per 10,000 decisions by 28 percent and removed service level agreement violations during peak-load tests. Under nominal conditions, feature computation latency was less than 300 ms and at the 99th percentile in stress testing, feature computation latency was less than 250 ms.
Conclusions: An event-driven cloud AI decision intelligence platform was much more responsive, predictive and cost effective than a traditional system of batch-processing. The proposed architecture is viable to healthcare operations by integrating real-time inference directly into enterprise operations and adding governance, auditability, and scaling on-the-fly capabilities. These results justify the shift to the continuous and real-time decision ecosystems in the enterprise healthcare setting and need additional confirmation in the multi-site and real-world clinical implementation.
Keywords: Artificial Intelligence, Real-Time Analytics, Healthcare Operations, Cloud-Native Architecture, Decision Intelligence, Workflow Automation, Enterprise Systems
How to cite this article: Manne TAK, Jindal M, Saini SS, Sivagnanam AM, Surianarayanan G, Jakku PC. Real-Time AI–Driven Decision Intelligence for Enterprise Healthcare Operations: Cloud-Native Architecture, Workflow Integration, and Measured Operational Outcomes. Int J Drug Deliv Technol. 2026;16(12s): 1039-1051. DOI: 10.25258/ijddt.16.12s.115
Source of support: Nil.
Conflict of interest: None