Sir Ganga Ram Hospital, is a leading multi‑specialty institution based in New Delhi. As the hospital’s scale and complexity have grown, so have the operational and clinical pressures: rising call volumes, continuous tendering and procurement, and increasingly demanding diagnostic workloads in the laboratory. Against this backdrop, Arun Kumar Goyal, CIO, at Sir Ganga Ram Hospital, has set out to redefine three key pillars of the hospital, patients, clinicians and back office, using a deliberately data‑lite AI strategy that focuses on small, high‑impact deployments rather than massive data overhauls.
The Challenge: Volume, Complexity and Fragmented Decisions
The hospital faces three intertwined challenges. On the patient side, thousands of people contact the institution every day, many unsure which specialist to consult when they experience multiple symptoms such as weakness, breathlessness, knee pain or arthritis‑like conditions. The call center, with about 25 people, receives “somewhere around 3,000 to 4,000 calls per day,” and, as Goyal admits, “we are missing a huge number of calls” because of sheer volume, leading to dropped calls.
In the back office, the hospital runs a “365 days, 24X7 tendering process” covering approximately 70,000 SKUs and 5,600 tenders. Hundreds of vendors quote for the same items in different units and packaging formats.
“One will give you the price of a single tablet, one in a strip, another guy gives you the price of a box, another guy gives you the liquid,” he says. “We have to normalize the whole data.” Comparing these bids and making decisions about margins, vendor performance and delivery reliability is complex and heavily manual, consuming an entire team’s time throughout the year.
In the laboratory, the hospital needs better visibility into how the lab is performing, which tests are most in demand, what trends are emerging, how many tests are rejected versus received, and how much reagent is consumed per test compared to standard norms. These insights are vital to streamlining operations but are not easily available in real time.
Goyal decided to “choose small use cases so that we are aware of how it will impact our operations, whether these are really AI cases or not.” The idea is to start with data‑lite applications, build awareness and capability, and then move towards more complex clinical AI such as decision support systems.
Pillar 1: Patient: AI Appointment and Chatbot
In the patient pillar, the flagship project is an AI‑driven appointment and chatbot system. Here, patients can speak with AI agents to seek doctor’s appointments, reach out to the appropriate doctor and, crucially, be guided to the right consultant based on symptoms. Goyal describes it as “patient appointment and chatbot, where patients can speak to the agents and seek doctor’s appointments, and reach out to the appropriate doctor.”
The key problem is simple but common: a patient may not know whether to consult orthopedics, cardiology or another specialty when faced with multiple conditions. “Suppose a patient is having weakness, breathlessness, knee pain or some arthritis type of problems, you don’t know whom to consult.” With the new system, “you just have to feed your symptoms and it will guide you to the right consultants.” The same system allows patients to fix appointments directly, and there will be an app for both iOS and Android, along with access through the hospital website and the call center.
In the call center, AI agents sit as the front layer. “These agents will be filtering more and more patients,” Goyal notes. The AI agent provides basic information and triage calls; if it is not able to resolve a query or has never handled a similar case, “then it will pass on to the human in real time so that the patient is not getting frustrated.” The human agent receives the information already captured by the AI and continues the interaction. Because the hospital is currently missing a large number of calls out of the 3,000–4,000 handled daily, Goyal expects that “AI agents will take care of those dropping calls, right, and they will not drop the patient calls.”
The same agents will sit on the website and in the mobile application, ensuring consistency across channels. Vendors, including global technology providers and specialized AI developers, have already presented demos, and the hospital is in the process of evaluation.
“I am hoping in the next three, four months” for implementation, says Goyal. He is clear that this patient‑facing AI project is intentionally data‑lite: “The appointment you completed with the doctor’s database and then time slots, that’s it. It’s not huge data.”
Pillar 2: Back Office: AI‑Enabled Tendering and Procurement
The second pillar, back office, focuses on the tendering process. Sir Ganga Ram Hospital’s tendering runs all year round, “24 by 7,” covering tens of thousands of SKUs and thousands of tenders. Hundreds of vendors participate, quoting for the same medicine or item in different units and packaging formats: single tablets, strips, boxes, liquids and more. As a result, the hospital has to normalize the whole data before any meaningful comparison or decision‑making.
Here, the AI agent’s first role is normalization: standardizing units of measurement, item descriptions and packaging types so that vendor quotes become comparable. Once this is done, the AI supports decision‑making. It helps answer questions such as “Where is the highest margin? Which is the best medicine? Which is the best performing vendor? Who offers services fast?”
“These decisions are today being taken manually, manually or semi‑automatic. With AI, the time and effort involved in the technical tender process will be reduced. There is a specific team which is always engaged all through the year. AI agents will not only provide the regular application, right, but the decision‑making as well,” Goyal says.
Again, he stresses that this is a learning phase: by implementing this application, the hospital will understand whether AI is truly the right solution, and “we will come to know what is the infrastructure required for this—how many GPUs consumption, how many token consumption, what will be the cost of these running through.”
Pillar 3: Laboratory – AI for Lab Performance
The third pillar is laboratory performance. “We are exploring the performance of the laboratory in terms of what is the return of the patient, which are the most performing tests, what is the trend, how many tests are rejected, how many received, what could be done, what is the standard agent consumption per test, how much are you consuming,” says Goyal.
The AI agent will provide answers to these operational queries and help streamline the laboratory: understanding rejection patterns, test performance, and reagent usage, and pointing to areas where processes can be improved.
Data‑Lite AI and the Road to Clinical Decision Support
A distinctive element of Goyal’s approach is the focus on data‑lite AI applications in the first phase. “Before you implement AI, I think the most important part is to have your data,” he observes, which is why he has selected use cases that depend on relatively simple datasets.
The doctor’s appointments doesn’t need too much data. Similarly, in laboratory performance, the hospital will expose test performance metrics and reagent consumption rather than deep clinical histories.
More data‑heavy applications, such as clinical decision support systems (drug interaction checks and patient history‑based decisions), are reserved for a second phase.
“We are planning to implement a clinical decision support system where the AI agent will help clinicians with the help of previous or current data of the patient and aid in decision‑making. This requires richer data including a drug interaction database and patient history,” says Goyal.
Hybrid Architecture
Architecturally, Goyal is planning a hybrid approach. “AI will be trained on our assets,” he says, referring to the hospital’s own data. Initially, he is strategizing to keep patient data in a closed environment. “There may be situations where we need to go to the public,” he acknowledges. In that case, he envisions “creating a separate environment for the agents where they will be and then within the data.”
For instance, “patient data would be in a closed environment, but if I need to do global components comparison, or seek some suggestions from the AI to give me a global or a national perspective, I need to put my data on the public and then get it from them, or maybe get the perspective from the national or global perspective, bring that data into the closed environment and then compare with the environment. That would be a hybrid approach.” He adds, “I would say it’s a two different approach, one closed and one public, with careful separation.”
Ultimately, he sees these initial projects as defining the hospital’s future AI roadmap. “The outcome of this project would actually define the future AI strategy,” he notes. “It is a learning case, I would say, not only for us but for the whole management,” he adds.
