Last month, 7,000 patients across Fortis, HN Reliance Foundation, and Sahyadri left hospitals without the usual long wait for final bills-a quiet but significant shift in a healthcare system where discharges typically stretch for hours. A new artificial intelligence system lets patients settle their accounts quickly and avoid the frustrating wait.
Medi Assist, one of India’s largest third-party administrators for health insurance, used AI to predict each patient’s out-of-pocket expenses with a margin of plus-minus 500 rupees, allowing for faster settlements and fewer delays.
“It’s the difference between leaving a hospital frustrated or feeling relieved,” says Satish Gidugu, CEO of Medi Assist. “For many, the uncertainty around bills at discharge can be as stressful as the treatment itself. We wanted to change that.”
This AI-powered process is part of Medi Assist’s broader push to streamline claims processing and ease patient experiences. Handling over 8 million insurance claims annually across more than 15,000 hospitals, the company found it necessary to adopt AI-not just for efficiency but also to manage the sheer complexity of the data involved.
The impact goes beyond quicker discharges. Medi Assist is also tackling a longstanding problem in health insurance: fraud. Since February, its machine learning models, which analyse over 160 parameters for each claim, have doubled insurers’ savings by identifying fraudulent claims before making payments.
“Insurance fraud is a persistent issue,” explains Dhruv Rastogi, SVP and Medi Assist Head of Data Science. “Our models continuously learn, spotting patterns often invisible to manual processes.”
Medi Assist’s use of AI addresses the everyday challenges of administration and the larger, systemic issues that have plagued health insurance for years.
The Data Challenge: Building the Foundation for AI
Becoming AI-ready was no small feat for Medi Assist. The company faced a daunting challenge in the form of fragmented, inconsistent data, ranging from handwritten hospital notes to scanned bills and unstructured insurance documents.
“When I joined Medi Assist 11 years ago, hospitals were still faxing us charges,” says Satish. “We weren’t even thinking about AI back then. The challenge was simply stopping the paper flow.”
Digitizing data became the first critical step. In India, many hospitals still rely on handwritten documents, and even simple medications like Crocin 500 can be listed in varying ways depending on the hospital’s system.
“There were no standard formats,” explains Dhruv. “We had to manually digitize millions of documents-hospital bills, insurance policies-with no immediate return on that investment.”
Over four years, Medi Assist painstakingly processed 70% of hospital bills, down to the individual line items, categorizing everything from surgeries to non-medical supplies. This created the structured data necessary for AI models. Likewise, millions of insurance policy documents were standardized to allow for automated interpretation.
“It wasn’t a matter of flipping a switch,” Dhruv adds. “Every detail, every line item, had to be organized for AI to learn from it. It took years to build that foundation.”
Once this infrastructure was in place, Medi Assist began training AI models on millions of records-claims, member interactions, and hospital data. The goal wasn’t just to automate tasks but to make sense of the vast amount of chaotic data.
“Our focus was on making the AI smarter and more accurate over time,” Dhruv says. “The data needed to be structured so that AI could continuously learn and improve.”
AI in Action: Solving Real-World Problems
Medi Assist uses artificial intelligence to address two major challenges in health insurance: speeding up patient discharges and detecting fraudulent claims.
One of the most immediate impacts has been the ability to predict out-of-pocket expenses for patients, significantly reducing waiting times for final bills. Hospitals must compile charges from multiple departments-billing, pharmacy, labs-before generating a final invoice. Medi Assist’s AI steps in to speed up this process.
“We can predict the final bill before the hospital even finishes compiling it,” says Dhruv. The AI estimates out-of-pocket costs with high accuracy by analysing millions of previous transactions, factoring in treatment types, and reviewing the patient’s insurance policy.
As a result, patients can leave hospitals faster. Last month alone, 7,000 patients left sooner than usual because AI-generated estimates were accurate enough for hospitals to proceed with discharges.
“It saves patients from hours of frustration,” Satish explains. “It brings peace of mind during an already stressful time.”
But AI’s impact doesn’t stop with billing efficiency. It’s also helping combat insurance fraud, a widespread issue that inflates insurers’ and policyholders’ costs. Medi Assist’s AI models analyse over 160 data points per claim, detecting irregular patterns and flagging potential fraud.
“The AI catches things often missed by manual review,” Dhruv says. “It’s not just about flagging suspicious amounts-it’s about recognizing patterns we’ve seen in fraudulent claims.”
Claims are assigned a fraud propensity score, and those deemed suspicious are sent to human investigators for further review. This combination of AI and human oversight has led to significant savings for insurers. Since the system went live in February, fraud savings have doubled.
“The more data the system processes, the smarter it becomes,” Dhruv adds. “We’re able to catch fraud faster and with greater precision every day.”
Ethical AI: Ensuring Fairness and Accountability
Ethical considerations are at the forefront as Medi Assist scales its AI efforts. The stakes are high-AI now influences decisions affecting millions, from hospital discharges to fraud detection.
“Bias is a real concern in any AI model,” Dhruv acknowledges. “We have to ensure algorithms don’t make skewed decisions, especially in an area as complex as health insurance.”
Medi Assist trains its AI models on a diverse dataset to mitigate bias, considering over 160 factors for every claim. This breadth helps ensure that no single factor disproportionately influences the outcome.
“A model that only looks at a few variables might focus on irrelevant details, like where a patient is from,” Dhruv explains. “Using a wide range of data points ensures the AI sees the full picture.”
Medi Assist also uses an in-house team of doctors to review data before it’s fed into AI models. This extra layer of oversight ensures that the information guiding the AI is accurate and relevant.
“Validating the data first helps prevent the AI from learning from flawed or incomplete information,” Dhruv says. “It’s about getting the foundations right.”
Once the AI models are deployed, they’re continuously monitored through “drift engines,” which detect any signs of bias or deviations from expected norms. Regular audits are conducted to ensure fairness across demographics such as gender and location.
“We actively manage the AI systems,” Dhruv notes. “It’s not enough to just set them up and let them run. We have to make sure they remain fair and unbiased.”
The Future of AI in Healthcare
For Medi Assist, AI’s potential extends far beyond faster hospital discharges and fraud detection. The company believes AI will fundamentally change how healthcare is experienced at every level, from the hospital bed to insurance claims.
“We’re just scratching the surface of what AI can do,” Satish says. “But as we move forward, keeping the human element central will always be our priority.”
(Pankaj Mishra is the AI Editor at NDTV. He shapes stories and conversations that help explain AI in India and how it influences people’s lives and work)
