The 1% that keeps leaders awake: Accuracy anxiety takes centre stage at ET AI Conclave & Awards 2025

At the inaugural edition of the ET AI Conclave & Awards 2025, held on February 26, 2026, in Bengaluru, a central tension animated one of the day’s most closely watched panels: how accurate is accurate enough when AI moves from experimentation to real-world deployment?

Titled Precision AI: The High Stakes of Accuracy, the session brought together healthcare and enterprise leaders to examine what happens when the margin of error shrinks or disappears altogether. Moderated by Pratik Bhakta, the discussion featured Dilip Jose, MD and CEO of Manipal Hospitals; Abhijeet Vijayvergiya, Co-founder and CEO of Nektar.ai; and Laina Emmanuel, Co-founder and CEO of BrainSightAI.

The speakers agreed that while AI’s promise is immense, its precision, or lack thereof, can carry vastly different consequences depending on context.


AI in healthcare: Delivering for the last mile
Opening the discussion, Emmanuel framed AI not as a job-displacing force but as a long-awaited problem-solving engine, particularly in medicine.

“At least in the healthcare space, I think there’s so much that can be done. There are so many unsolved clinical problems… for the lack of computational infrastructure, which can help solve those problems, [they] have not been solved so far,” she said.

BrainSightAI’s advanced brain mapping solutions, she explained, were made possible only after cloud infrastructure matured enough to handle “really heavy computational neuroscience workflows” and make them accessible to doctors and hospitals across the country.

Her excitement was rooted in AI’s ability to augment human understanding. Recounting a recent interaction with researchers, she described how an AI model could interpret a complex mathematical equation with minimal context and help researchers apply it to “very complex problems.”

“And I feel like that is the future that I am very excited about,” she said.

A cautionary tale: The moral complexity of care
If Emmanuel embodied optimism, Jose introduced a deliberate counterweight.

“In an AI conference, I know it’s difficult for me to tell everybody to be a little cautious of AI, particularly in the context of healthcare,” he began.

Healthcare, he stressed, is “emotionally sensitive, morally complex, context heavy, and oftentimes value or principle laden.” These are not environments where probabilistic systems can operate unchecked.

“We can’t be 99% accurate in healthcare. We can’t have any hallucination in healthcare,” he said bluntly. “For most other sectors, maybe the 1% miss is a lost sale… In healthcare, it’s a lost life.”

With 50 hospitals across its network and nearly eight million patients annually, Manipal Hospitals holds vast longitudinal datasets, spanning imaging, lab reports, and even family histories over a decade. That data, Jose admitted, is powerful. But its use raises uncomfortable questions.

“Should we use the clinical course that a 20-year-old elder sibling went through and use that to sort of alert a younger sibling of a likely challenge he or she may face?” he asked. “Where do we draw the line between our interest as a commercial organization and the general interest of the patient?”

As India’s Digital Personal Data Protection (DPDP) Act comes into force, compliance and consent are no longer optional. “When we have to train a model, when we have to use datasets, I think the consent process… is something to be followed,” he said, adding that some dilemmas still lack clear solutions.

Data and the perennial enterprise battle
While healthcare grapples with life-and-death stakes, enterprise AI faces a different but equally foundational challenge: data quality.

Vijayvergiya cited a widely discussed study showing that the majority of enterprise AI projects fail largely due to poor data feeding the system.

“If poor data is the input to the systems, you’ll have a garbage in, garbage out problem,” he said. “If you don’t deliver precision… You have no trust in the system. And if you have no trust in the system, adoption takes a hit. And if there’s no adoption, you’ll not get ROI.”

For Nektar.ai, the solution lies in unlocking unstructured historical business communication, emails, sales conversations, and interactions that never made it into formal CRM systems. By extracting insights and feeding clean signals back into enterprise systems, AI becomes more accurate and scalable.

“Now, when you have this AI-ready data in your CRM or your data warehouse, then if you deploy AI agents on top of it, they are far better,” he said.

However, precision here also means strict guardrails. “We have to make sure that we have the guardrails in place and the filters in place to not take in any of the data which is not contextual, which is not related to the business,” he said, referring to sensitive communications like salary slips, board discussions, or private appointments embedded in emails.

Learning from Healthcare: Consent and context
Interestingly, Emmanuel suggested that other industries could borrow a page from healthcare’s playbook.

“When we had to get our data, we had to take informed consent from every patient, where we had to make sure that the patient understood what we were taking the data for,” she said. Only after obtaining comprehensive consent could her team build the AI models they use today.

That discipline, she implied, should not be unique to medicine.

Reflecting on BrainSightAI’s journey since 2019, before AI became a buzzword, she described six years of “plodding work and a lot of hard work.” While faster iteration cycles today may shorten that timeline for others, she welcomed the influx of builders.

“There are so many unsolved problems in healthcare that the more people come into it and solve it, the better it is for all of us,” she said.

The 1% that keeps leaders awake
As the session drew to a close, the moderator posed a final question: What is the 1% error that keeps each of them awake at night?

For Emmanuel, the answer lay in proactive risk management. “Every sprint… the team actually think[s] of all the possible risks that could arise… and then mitigates those risks,” she said, describing a guardrail-first culture before any hospital deployment.

For Jose, the fear was existential. “Are we deploying it without thinking it through?” he asked. AI must be embedded into clinical workflows as a “force multiplier,” not a standalone efficiency tool.

Yet he acknowledged the promise on the operational side, from compressing hospital discharge times from “four to five hours… to maybe one, one and a half hours now,” to streamlining scheduling and billing queries through chatbots.

For Vijayvergiya, the anxiety centres on precision within sensitive enterprise datasets. “It does keep us worried to make sure that the enterprise precision is important,” he said. A missed forecast or exposed confidential communication may not equal a lost life — but it can erode trust irreparably.

At the inaugural ET AI Conclave & Awards 2025, the message was clear: the future of AI may be intelligent, but it must first be accountable.

The ET AI Conclave & Awards 2025 has L&T Finance as the NBFC partner, Snowflake as the AI Data Cloud Centre, EY as the Evaluation Partner, and T-Hub as the Ecosystem Partner, and is driven by BYD, with Indri as the celebration partner and Zoho as the Technology Partner. Vahdam India, Vaaree, Natch, and Andamen are the event’s gifting partners.