The AI race is now about execution

AI performs brilliantly in a controlled environment. Give it clean data, a focused task and a contained demo, and it can be impressive. But an enterprise is not a controlled environment. The real test begins when AI meets live systems, fragmented data, legacy processes and human-led throughput. This is where many promising initiatives lose momentum, long before they deliver meaningful business value.

Across Indian enterprises I work with, the pattern is familiar. Boards are engaged, budgets are being approved and pilots are multiplying. Yet relatively few scale. Globally the picture isn’t very different, with Deloitte’s research finding that only a quarter of organisations have moved 40% or more of their AI pilots into production. The challenge, in most cases, is not the model. It is the infrastructure underpinning it.

In my conversations with business leaders, the excitement around an early AI pilot often masks a more challenging reality: how to build a robust data architecture that allows AI to produce reliable inference repeatedly at scale. And this is where many enterprises still have work to do, starting by recognising that reliable AI output is less a function of the model and more about strong data foundations. For consistent outcomes, data must be organised, connected and accessible through a strong data architecture.

Yet today, most organisations have information trapped in silos across applications, locations and environments. Breaking those silos is not simply a technology exercise. It is a prerequisite for building AI that employees can trust and scale beyond the pilot. Because once AI output becomes unreliable, people start checking everything manually. The productivity benefit disappears, and employees lose confidence in the technology. Therefore, trust in AI starts well before the model is selected. It starts with the right data foundations.

Additionally, adoption is not the same as value. This distinction matters particularly in India, where AI adoption is moving quickly, with innumerable ongoing AI pilots. Among enterprises, the number of pilots launched tells us little about the value being created. A proof of concept can work beautifully in isolation and still struggle when it encounters live data, legacy systems and the complexity of real business workflows.

The success of countless pilots means Indian enterprises are no longer simply asking the question, “Can we make AI work?”. They can. Instead, they’re asking themselves, “Can we make it work reliably, repeatedly and at scale?” Answering this question requires thinking differently about AI.

I find it useful for enterprises to ask themselves two questions: How much of the model is uniquely ours? And how much of the data feeding it is uniquely ours? Together, these questions create four broad operating models where organisations land.

Some organisations will rely largely on external models and external data, where governance and centralised oversight become critical. Others will combine external models with proprietary data, making reliable data pipelines the priority. Organisations building custom models on their own data will need scalable, high-performance infrastructure. And where sensitive data is involved,

that infrastructure must also support strong security, compliance and India’s DPDP requirements.

Most large enterprises will operate across several of these models at the same time. What matters across all of them is bringing data, models and compute together without creating another layer of fragmentation. For Indian organisations, this is where a unified edge-to-cloud approach can make a practical difference: allowing data to remain where it needs to be, namely on premises, at the edge or in the cloud, while still being securely accessed, governed and used.

EPS India, which operates ATM and fintech services, has found success by doing so. Its operating environment needs to support real-time financial transactions, high availability and stringent security and compliance requirements, making a unified edge-to-cloud approach the ideal solution. The company’s success highlights the difference between demonstrating that AI can work and building an environment where the technology is relied upon every day by millions.

But AI readiness is also about more than data. It is about creating an environment where data can move securely, models can be deployed where they make the most sense, and compute can scale with the needs of the business. To be AI ready, enterprises must seamlessly connect these pieces without adding complexity.

That, I tell Indian leaders, is how they will win the next phase of enterprise AI. Those who succeed will not be the ones chasing every new model or running the most pilots. The winners will be the ones building the architecture that allows data to move securely, models to be deployed optimally, and compute to scale seamlessly alongside the business.

These enterprises will see pilots moving from experiment to production, from production to scale, and from scale to measurable business outcomes.

As I share with Indian leaders, that is the opportunity ahead: not simply to adopt AI faster, but to build the data foundations, infrastructure and operating discipline to turn it into sustained business value.