New Delhi, Nov. 18 -- For enterprise AI to deliver meaningful outcomes, building the right technology may be only half the challenge. The other half is understanding how people work, where decisions are made and what information they need to make them better.

Indian healthcare major Apollo Hospitals' collaboration with IT giant Microsoft offers an example of this approach. Rather than develop an AI application around a predefined brief, Microsoft's forward-deployed engineers (FDEs) worked alongside Apollo's clinical governance and technology teams to understand how patient safety reviews were conducted and build a solution around those workflows.

The resulting Daily Tiered Huddle (DTH) agent is designed to help clinicians examine adverse events, including patient falls, medication errors and readmissions, by bringing relevant patient information together and assisting with case assessments.

The scale of the task is significant. Apollo reports around 600 adverse events every month across its 55 main hospitals. Earlier, clinicians often had to retrieve patient histories, medication records and test results manually from different hospital systems. Assembling this information could take hours or even days, delaying the review process.

The AI agent, developed using Microsoft Foundry, draws on hospital data stored in Microsoft Fabric. It consolidates relevant information, generates case summaries and prompts clinicians with questions to support their investigations.

For instance, when a patient returns to the intensive care unit within 48 hours, clinicians can access the treatment timeline, medication history and laboratory results through the agent, rather than depend on multiple teams to compile the information.

The CIO's outcome-led approach For Ashokkan Somuveerappan, chief information officer at Apollo Hospitals, the project represents a different way of approaching enterprise technology development.

The FDEs worked directly with doctors, observed daily clinical governance meetings and collaborated with Apollo's technology team to map existing electronic health records against the information required for safety reviews. This helped translate clinical expertise and established processes into the agent's capabilities.

Somuveerappan says the engineers were focused on the overall outcome rather than completing narrowly defined technical assignments. Traditionally, application development, database management and user interface design would involve separate engineering teams. The FDE model allowed engineers to work across these layers to address the problem.

"They were more glued to the outcome that they needed to achieve," he says, describing it as a different and innovative way of working.

Data handling was also part of the development process. Microsoft engineers used synthetic data for evaluation and testing, rather than Apollo's electronic health records. Patient information and the data generated through the review process are stored in Microsoft Fabric in accordance with the hospital's data-handling standards.

The deployment is beginning to show operational gains. Apollo's records indicate that more than 60% of cases reviewed through the huddle were closed within three days after the agent's rollout. This allows clinicians to concentrate on the remaining cases requiring deeper investigation, root-cause analysis or clinical outcomes review.

From automation to continuous learning The more significant aspect of the deployment is its ability to learn from clinical reviews. The agent captures the questions raised, discussions held and decisions made during daily huddles. This information feeds back into the system, helping it refine the questions and assessments it generates for subsequent cases.

Clinicians remain responsible for evaluating the recommendations and making the final decisions. According to Somuveerappan, the extent to which clinicians need to edit the agent's suggested questions and recommendations is one indicator of how well it is acquiring organisational knowledge and context.

The system is also being developed to identify patterns across incidents. In one instance, it flagged a cluster of central-line-associated bloodstream infections in a bone marrow transplant unit, prompting corrective action.

Apollo has reported a decline in patient falls and medication errors between July and September 2026 compared with the corresponding period in 2025. Its annual safety culture survey also recorded an almost 30% improvement in staff satisfaction with how errors were handled. These are hospital-reported outcomes and do not establish that the AI deployment alone caused the improvements.

Apollo is now exploring the application of similar capabilities to tumour board reviews and doctor handovers, with the longer-term ambition of identifying deviations from treatment pathways and potential safety risks earlier.

For CIOs, the deployment highlights a shift in enterprise AI adoption: from implementing tools that automate existing tasks to building systems that absorb organisational knowledge, support decisions and improve through continuous feedback. At Apollo, the technology challenge is no longer just about making information accessible, but using it to strengthen how the organisation learns and responds.

Published by HT Digital Content Services with permission from MINT CIO CIRCLE.