Early Training and the Move Into Business Analytics

How did moving from computer science in India to business analytics in the United States change the way you thought about technology?

I studied Computer Science in India, which gave me a technical foundation and a structured approach to problem-solving. Over time, I found myself more interested in what happens after a system produces an output: how people interpret information, determine whether a decision is correct, measure its consequences, and choose what action to take.

My Master’s in Business Analytics at Arizona State University expanded that focus into statistics, experimentation, predictive analytics, business intelligence, and evidence-based decision-making. That training shifted my attention from simply asking whether a system could perform a task toward asking whether its outputs could lead to useful, validated, and accountable decisions.

Lessons From Different Sectors

What did working across operations, large-scale technology platforms, financial services, and account security teach you about automated decision-making?

Across Better Being Co., Meta, Discover, Capital One, and now Google, a common theme emerged: analytics and automation can improve decisions, but only when accompanied by validation, sensible metrics, and human context.

At Better Being Co., I learned that more information does not automatically produce better choices; analytical outputs have to be translated into clear, actionable insights.

At Meta, supporting commerce analytics at global scale reinforced why experimentation matters. Correlations in dashboards are not proof of causation, and teams need well-defined success metrics, appropriate experiments, and mechanisms for identifying unintended effects.

In financial services, the stakes can become personal because analytical errors may directly affect consumers. Validation, governance, and mechanisms for detecting and correcting errors therefore become particularly important.

Today, in Account Security and Trust & Safety at Google, those threads converge. Large-scale pattern detection is powerful, but security decisions still require context, investigation, validation, and clear human responsibility.

The scale of the environments in which I have worked has also shaped how I think about these questions. Better Being distributes its brands and products in more than 65 countries; Capital One says it serves more than 100 million customers; Meta says approximately 3.5 billion people use its apps and services each day; and Google says it operates 13 products with more than one billion users each, including five with more than three billion users. These figures describe the scale of the organizations and platforms-not the number of people personally affected by my individual wor-but they illustrate why reliability, experimentation, and safeguards become increasingly important when decision systems operate at scale.

Human Oversight and the Limits of Approval

Why do you believe human oversight needs to mean more than simply having a person approve an AI recommendation?

Approval alone can be a brittle form of oversight.

My independent research argues that automation should augment human judgment by reducing repetitive work while increasing the importance of contextual interpretation, governance, and higher-level analytical judgment.

In papers such as Data Science, Analytics, and the AI Revolution: Creating More Jobs and Empowering Careers and Why AI Cannot Replace Business Intelligence, A/B Testing, and Human-Driven Insights Across Industries, I examine why experimentation, traceability, validation, and contextual interpretation remain necessary even as AI becomes more capable.

Later work, including Integrating Human Oversight with AI-Based Planning Systems: A Framework for Ethical Business Intelligence and Adaptive Experimentation in Nationally Critical Sectors, proposes a more concrete approach combining AI planning, experimentation loops, monitoring, and human-in-the-loop governance so that people do more than simply rubber-stamp automated recommendations.

Separately, serving as a judge and reviewer of others’ work has reinforced for me that strong methodology, transparent assumptions, evidence, and validation matter at least as much as the sophistication of a model or technique.

Policy Context and Broader U.S. Technology Priorities

How do current U.S. priorities around AI, cybersecurity, validation, and human control connect with the problems you have been studying?

Recent U.S. strategic guidance overlaps closely with several of the technical questions I have encountered in industry and research.

The White House’s 2026 National Security Science and Technology Strategy identifies artificial intelligence and autonomy among its critical and emerging technology areas and specifically includes planning, reasoning and decision-making, interpretability and control, and adversarial robustness and AI security. The strategy separately identifies information management and cybersecurity, including AI-enabled cyber capabilities and digital identity technologies.

NIST’s Artificial Intelligence Risk Management Framework similarly emphasizes characteristics such as validity and reliability, security and resilience, accountability and transparency, and explainability and interpretability.

Those documents do not imply that I work for the U.S. government or that government agencies have endorsed my individual research. What they show is that areas such as AI reliability, security, validation, interpretability, and control are receiving explicit federal strategic attention. Those are many of the same technical challenges that have emerged repeatedly across my professional experience and research.

Research, Practice, and What Comes Next

What do you want to continue researching and building in the United States over the next several years?

My professional direction is broader than any single employer.

I want to continue developing reusable approaches to AI validation, experimentation, security, monitoring, and human oversight that can apply across high-impact environments. Practically, that means methods and processes for validating AI-assisted decisions, measuring whether interventions actually improve outcomes, identifying systemic errors earlier, monitoring changing behavior, and preserving clear human decision ownership.

I am particularly interested in the recurring problems that appear across otherwise different industries. Financial services, large-scale digital platforms, account security, and other high-impact environments involve different technologies and objectives, but they repeatedly raise similar questions: What is the system optimizing? How do we know whether an intervention actually worked? What happens when conditions change? How are unintended consequences identified? When should humans intervene, and who ultimately owns the decision?

I plan to pursue these questions through a combination of industry practice, independent research and publication, reviewing and judging peer work, and interdisciplinary collaboration.

The goal is to develop methods and operating practices whose underlying principles can transfer across organizations and sectors rather than remain useful only inside one company or product.

Closing Thoughts

My career began in computer science and moved through business analytics, product analytics, financial-services analytics, experimentation, and account security and Trust & Safety.

Across those experiences, the same question has increasingly guided my work: as organizations rely more on data, automation, and AI, how do we keep experimentation, validation, security, human judgment, and accountability at the center of decision-making?

The answer, as I see it, lies in combining careful measurement and experimentation with monitoring, adversarial thinking, and governance so that automated systems amplify human responsibility rather than obscure it.

I intend to continue building and studying those approaches in the United States, publishing what I learn, and developing ideas that other teams and organizations can evaluate and adapt to their own high-impact environments.

Read the full article →