India’s AI Problem is Product Thinking: Hoonartek’s CEO Peeyoosh Pandey

In a conversation with NervNow, Peeyoosh Pandey, CEO of Hoonartek, argues that the race to deploy AI models is the wrong race, and that the enterprises winning by 2030 will be the ones that built decision infrastructure, not the ones that bought the best APIs.

Peeyoosh Pandey has spent more than fifteen years building data systems for some of India’s most complex enterprises: financial institutions, insurers, and large-scale BFSI operations where the cost of a wrong decision is regulatory, and not limited to operational. As CEO of Hoonartek, he sits at the intersection of where enterprise data infrastructure meets the current wave of AI adoption. His view is unambiguous: most of what is being called AI transformation is infrastructure work that has not been done yet. During an in-depth conversation with NervNow, Pandey laid out what separates cosmetic transformation from the real thing, and what Indian enterprises need to build now to be competitive by 2030.

Excerpts

NervNow: What structural gaps in data architecture are most preventing organizations from realizing measurable ROI from AI?

Peeyoosh Pandey: The problem is not the AI. Enterprises built their data infrastructure to bring data from systems of record like ERP, core banking, and claims platforms, and not to provide intelligence. When you try to run AI on top of infrastructure that was designed to store and retrieve transactions, you end up with models that are technically impressive and operationally less valuable.

The specific gaps I see consistently are three. First, decision logic is buried inside platforms rather than being treated as an enterprise asset. You cannot build intelligent systems on top of intelligence that is already locked inside SAP configurations and vendor rule engines. Second, data is organized around systems rather than around decisions. An enterprise that wants to automate credit decisions needs data organized around the credit decision, and not around the core banking system that processes the loan. Third, there is no governance layer between the AI output and the business action. Insights without governed execution paths create liability, not value.

NN: What differentiates true data modernization from surface-level transformation?

PP: True data modernization has one defining characteristic: it changes how decisions are made, not just where data is stored. If your business leaders are making the same decisions in the same time with the same confidence after your transformation as they were before it, you have upgraded your plumbing. You have not transformed your business.

The test I apply is simple. Before the transformation, how long did it take to reach a decision for a given business situation? If the answer after transformation is still measured in weeks or months because the logic is still embedded in platform configurations, the transformation was cosmetic. If the answer is hours or days because decision logic now lives in a governed, independent layer that business teams can actually control, you have done something real. Surface-level transformation gives you better dashboards. True modernisation gives you platforms built for decisions.

“If your business leaders are making the same decisions in the same way after transformation, you have upgraded your plumbing. You have not transformed your business.”

NN: How can enterprises modernize mission-critical systems without compromising risk, compliance, or operational continuity?

PP: The answer is logical separation of architecture, not a big-bang migration. The mistake most enterprises make is treating modernisation as a replacement programme. That approach concentrates risk, extends timelines, and almost always compromises operational continuity at some point.

What actually works is introducing an intelligence layer above the existing systems rather than inside them. Core banking continues to process transactions. Claims platforms continue to manage policies. ERP continues to run supply chain. But the decision intelligence: the rules, strategies, and judgment calls that those systems have been executing, gets extracted into a governed layer that can evolve independently. This delivers three things simultaneously: systems of record remain stable and auditable; decision logic becomes visible and changeable; and modernisation becomes incremental, which maintains operational continuity.

NN: With AI models becoming commoditized, is competitive advantage shifting to data engineering and integration capability?

PP: The commoditization of models is real and accelerating. Any enterprise can access a world-class model through an API for a few cents per inference. The model is no longer the moat.

What has not been commoditized is the ability to ground those models in enterprise-specific decision context, govern their outputs within regulatory constraints, and connect their intelligence to operational execution. That requires deep data engineering, but more specifically it requires understanding how decisions are made in a specific industry, what data those decisions depend on, and how to structure that data as an asset rather than a by-product. The enterprises that will win are not the ones with the best models. They are the ones that have built the most sophisticated decision infrastructure: the data pipelines, semantic layers, governance frameworks, and integration architectures that turn commodity AI into proprietary intelligence. That is genuinely hard to build and genuinely hard to replicate.

The enterprises that will win are not the ones with the best models. They are the ones that have built the most sophisticated decision infrastructure.

NN: Are enterprises architecturally and culturally prepared for AI that makes decisions rather than just provides insights?

PP: The architectural gap is real but solvable. The core issue is that decision logic is embedded in systems rather than owned by the enterprise. Extracting that logic, structuring it, and connecting it to agentic execution is an engineering problem; hard, but tractable. Enterprises that invest in the right data architecture can close this gap in 18 to 24 months.

The cultural gap is more complex. Autonomous AI requires something most organisations have never explicitly done: define what a good decision looks like. When a human makes a credit decision, the definition of good is implicit in their experience and judgment. When an AI agent makes that decision, the definition must be explicit, stated as policy constraints, success metrics, and escalation criteria. Most enterprises do not have this clarity. The enterprises I see succeeding with agentic AI are the ones that have done this cultural work first, bringing risk, compliance, operations, and technology into the same room and answering the hard questions: what decisions are we comfortable automating, at what confidence threshold, with what human oversight, and how do we measure success?

NN: Where are Indian enterprises leading globally in data innovation, and where are they structurally behind?

PP: India’s BFSI sector, particularly payments infrastructure through UPI, credit bureau ecosystems through CIBIL and Experian, and the Account Aggregator framework, has built data processing capabilities that most Western markets are still catching up to. The scale at which Indian banks process real-time transaction data, the sophistication of India’s credit decisioning infrastructure, and the speed at which regulatory data frameworks like ONDC are being adopted are genuinely world-leading. The depth of engineering talent reinforces this: India produces more sophisticated data engineers per year than any other market, and the complexity of problems Indian teams solve routinely, multi-system integrations, legacy modernization at scale, real-time processing for hundreds of millions of users, is not matched elsewhere.

Where we are structurally behind is product thinking in data and AI. Indian enterprises and technology firms are world-class at building and delivering. We are significantly behind in defining categories, building platforms, and commercializing products. The gap between the sophistication of what we build and the value we extract from what we build is the central challenge for Indian data and AI firms over the next decade.

“Where we are structurally behind is product thinking in data and AI.”

NN: How should Indian enterprises architect their data ecosystems given DPDP, ONDC, Account Aggregator, and evolving digital public infrastructure?

PP: Three architectural principles matter most. First, consent-first data architecture. The Account Aggregator framework has made consent-based data flows a regulatory reality in financial services. Enterprises that architect their systems around consent, where every data flow is explicitly authorised, traceable, and revocable, are positioning ahead of regulation, not just complying with it.

Second, encrypted data layers. DPDP requires that certain data is stored, processed, and governed within Indian jurisdiction. Enterprises need architectures that enforce this at the data layer, not just at the infrastructure layer.

Third, treat digital public infrastructure as infrastructure, not competition. Many enterprises still view ONDC, UPI, and the Account Aggregator framework as compliance burdens. The smarter framing is that India has built public data infrastructure that no enterprise could afford to build independently. Enterprises that architect their systems to participate in this, contributing data through appropriate consent frameworks, consuming DPI-enabled data flows for decisioning, will have access to population-scale intelligence that their global competitors cannot replicate.

This interview was conducted by NervNow with direct responses from Peeyoosh Pandey, CEO of Hoonartek. The submission has been lightly edited for formatting only. No responses have been altered in substance.

MORE AI INTERVIEWS
AI Governance Must Match AI Authority: FSS’s Rishi Verma on Fraud, Reconciliation, and the Next Phase of Intelligent Payments
Jamie Dimon’s Powerful AI Warning for Future Jobs
Women Must Understand How AI is Sold, Procured, and Regulated: Chaitra Vedullapalli, Women In Cloud
Medicine is Not About Algorithms; Responsibility Must Stay With a Human: Surjeet Thakur, TrioTree CEO
Without System Maturity, AI Remains Just a Tool: Hector’s Meher Patel on What Brands Keep Getting Wrong