India’s AI opportunity will be built on data infrastructure, not just models 

By Supratik Shankar, Co-founder, Dview

Walk into any boardroom in Mumbai or Bengaluru today. You’ll hear the exact same question: “When are we deploying AI?” Every executive is under pressure. Leaders are carving out massive budgets in a scramble to keep up. But under the hood, a structural failure threatens the entire transition. The bottleneck isn’t our talent. And it isn’t a lack of ambition, either. It is simply that our internal data and knowledge around workflows is fragmented, undocumented, and unprepared for AI workloads.

We should stop blaming the models

The models work just fine and are, in fact, incredibly capable. The real issue is that most Indian enterprises cannot actually use them yet because their databases are in no shape to support AI workloads. The problem goes far beyond minor glitches and stems from siloed systems, mismatched records, and missing labels, which remain the default state of affairs across the board. Feeding broken, chaotic files into a powerful AI model does not automatically produce magic. Instead, it results in highly confident but incorrect decisions. 

This isn’t a knock on Indian business. It is just history. US companies spent 30 years cleaning up their data habits by building a rock-solid data culture. They did it because strict regulators forced their hand, and investors demanded clean audits. They built a solid foundation over decades. Now, they are placing AI on top of it. On the other hand, many Indian enterprises are trying to skip those thirty years. The focus is on building the roof before even pouring the concrete for the walls. 

Compliance is not data culture in our country

Yes, India has strong financial regulators. The RBI, SEBI, and IRDAI pass down massive data-handling rules. But let’s be honest, compliance is just paperwork. Most firms do the bare minimum to be in the good books of auditors. Nobody treats this data like actual capital. They check the boxes, make changes in the excel files, print the compliance reports, and ignore the actual systems holding that data. You cannot run a smart predictive model on data that only exists to satisfy regulatory guidelines.

The regulatory framework for AI in India is still evolving, with no comprehensive playbook in place yet. In the absence of clear guidelines, many organisations are drawing inspiration from AI use cases and operating models that have proven successful in the US. However, the context is fundamentally different. Indian enterprises often operate with more constrained research budgets and a lower appetite for experimentation and risk. Replicating approaches developed for vastly different market conditions, without adapting them to local realities, can limit the effectiveness and scalability of AI initiatives. 

The fear of making mistakes

Real innovation requires a fail-fast approach, which in turn needs an appetite for experimental budgets and visionaries who are patient while sponsoring with capital.

In Silicon Valley, a tech team can burn through five million dollars testing a wild hypothesis, realise it’s a dead end, and move on with a shrug. Indian enterprises, however, do not always have that luxury. Budgets are tight, the window for experimentation is short, and a single mistake can damage hard-earned reputations.

As a result, many Indian enterprise leaders play it safe. They find a proven US application, retool it slightly for the Indian market, and call it innovation. It feels safe, but not cheap from a TCO standpoint when scale hits the ceiling. It also means losing the opportunity to build an AI ecosystem that actually fits the unique realities of the Indian market.

What actually works

Let’s get practical. AI is not a math problem- it’s an infrastructure and data awareness problem. The smartest companies in India have already figured this out. Instead of chasing flashy, surface-level pilots that lead nowhere, they are putting their heads down to fix the core data infrastructure.

Executing this shift comes down to three clear priorities:

Building a unified knowledge layer to pull information out of isolated department silos. So that the AI models actually understands the real-world context of your enterprise data, not just a bunch of random numbers.

Cleaning and labelling the raw data so that AI can consistently give accurate answers. We can’t expect AI to clean the messy datasets with a magic wand. If the sales team cannot rely on the client transaction database, AI will not deliver any value. 

Shifting the company culture so that maintaining clean and governed data is someone’s actual day job with real bonuses tied to its accuracy, not just delivery speed. Clean datasets are easily monetised, they create a faster route to solving business problems which are ever evolving but not at the cost of compliance.

None of this actually makes for a flashy press release. But it is the only way to build AI that actually generates revenue, saves time, and builds efficiency instead of just burning compute and precious time.

The hard choice ahead

Look at the sheer scale of what we are producing. India is generating transactional data faster than almost anywhere on earth. Digital micro-loans, instant deliveries, health diagnostics, farm logistics- the raw value is there. It is just buried under a mountain of unorganised systems.

The winners of this decade won’t be the loudest brands launching generic chatbots. It will be the disciplined ones. The companies that quietly invest in rock-solid data infrastructure, hire great data architects and engineers, and treat accuracy like a matter of survival.

The boardroom conversation needs to change starting today. Instead of asking which new model to buy, organisations should start asking whether their data is even clean enough to feed into it. In almost every company, the honest answer is “not yet.” But admitting that? That’s the exact pivot needed to build an AI economy on solid concrete instead of shifting sand. 

India does not need to wait for a smarter model to come along. It needs better, cleaner datasets right now. That is the gap the next generation of data infrastructure companies will need to close, one enterprise at a time.