From AI pilots to enterprise intelligence: Why Systemic AI will define the next decade

As artificial intelligence moves from experimentation to enterprise-wide adoption, organisations are discovering that successful AI transformation requires far more than deploying powerful models or launching isolated pilots. It demands a connected architecture, AI-ready data, strong governance, and a relentless focus on business outcomes. In an interaction with Express Computer, Sayandeb Banerjee, Co-Founder & CEO, MathCo, reflects on the company’s decade-long journey from an analytics startup to a global Enterprise AI partner. He discusses why enterprises must embrace “Systemic AI,” the growing importance of owning enterprise intelligence and intellectual property, the metrics that truly define AI success, and how business leaders can prepare for the next era of decision intelligence.

MathCo has completed a decade-long journey from being an analytics-focused startup to becoming a global Enterprise AI partner. Looking back, what have been the defining milestones, strategic pivots, and lessons that shaped this transformation?

When people say, “it’s been a decade,” it does make you pause for a moment. There’s certainly a sense of pride in what we have built and achieved together. But honestly, there are times when it feels like ten years is all it’s been. We have accomplished a great deal, but it also feels like we are only scratching the surface of what’s possible.

Before founding MathCo, my co-founders, Aditya Kumbakonam and Anuj Krishna, and I spent years in analytics, solving business challenges for large enterprises. We identified a core issue – organisations often lacked ownership and real use of their data and analytics solutions. One of our proudest achievements has been helping enterprises move beyond simply consuming analytics to building their own data and AI capabilities.

What began as a vision to democratise data-driven decision-making has evolved into helping organisations embed intelligence into the fabric of their business, empowering them to make faster, smarter, and more confident decisions.

Over the course of our journey, we have made many pivots both big and small based on what we see in the market and where we want to go. 2 years in we pivoted and built accelerators for capability growth that we were able to go to market with which resulted in growth for us. It was a lesson for us that we need to keep our focus on the ground, talking to our customers constantly and find ways to fulfil latent demand. Post that, as technology moved faster, we had to innovate constantly to ensure that our customers can get the right advice from us on how to maximise value.

Many organisations have experimented with AI through pilots and proofs of concept yet struggle to scale. What are the biggest barriers preventing enterprises from achieving enterprise-wide AI adoption, and how can they successfully transition from isolated AI initiatives to decision intelligence at scale?

We must understand that the issue is not with models failing. In isolated pilots and controlled environments, they often perform remarkably well. The real challenge is that enterprise intelligence remains fragmented. For example, there is a copilot in one function, a forecasting model in another, and a chatbot somewhere else, but these capabilities rarely connect to form a cohesive system. As a result, pilots demonstrate potential yet struggle to transform how enterprises operationalise decisions and execute actions at scale.

The problem is structural rather than technical. For years, organisations have approached AI as a collection of point solutions. What they truly need is an architectural foundation – a connected system where knowledge, decisions, and value accumulate over time instead of being recreated with every new use case.

At MathCo, we call this “Systemic AI”. We describe it as a supercharged brain for the enterprise, composed of four specialised lobes working together through a unified platform spine. This architecture enables intelligence to flow across the organisation, creating continuity rather than isolated pockets of capability.

This shift is critical because it fundamentally changes the economics of AI. Without an underlying architecture, every new initiative starts from scratch, causing costs and complexity to grow with each additional use case. With a systemic foundation in place, however, every agent, model, and data asset strengthens the next. Intelligence becomes cumulative, ensuring that what is built today accelerates the value delivered tomorrow.

MathCo has built a reputation around custom data products and proprietary intellectual property. How important is IP ownership in today’s AI-driven economy, and what competitive advantages does it create for enterprises seeking long-term differentiation?

Helping enterprises build and own their intellectual property has always been a foundational principle at MathCo. We have always believed that everything that matters should belong to the enterprise – the models, the context layer, the knowledge graphs, the decision logic, and the workflows that drive business outcomes. These are not just technology assets; they are enterprise IP that should remain within the organisation, continuously learning and compounding in value over time.

As AI becomes more ubiquitous, the differentiator won’t be who has access to the best models. It will be who has built the strongest enterprise intelligence – the proprietary data assets, decision frameworks, and contextual knowledge that competitors cannot easily replicate.

As Generative AI moves beyond the hype cycle, where are you seeing the most meaningful business applications emerge? What separates successful GenAI deployments that create tangible value from those that remain experimental?

What separates successful GenAI deployments from experimental ones is whether they are designed to create enterprise intelligence. So, instead of asking “how do we build an AI agent,” businesses should start with “what business problem are we solving, and is AI the right tool for it?” When purpose leads, technology follows, and that alignment must be deliberate from day one.

In practice, this means three things:

– Treat AI initiatives as enterprise transformation, not isolated pilots; embed them into workflows, governance structures, and decision rights rather than running side-of-desk experiments that never scale.

– Build the foundation before chasing flashy use cases: clean, connected, contextualised data and GenAI-ready systems and APIs are prerequisites, not afterthoughts. Without them, even well-intentioned initiatives stall at the proof-of-concept stage.

– Tie every initiative to measurable outcomes upfront, whether that’s faster drug discovery, optimised supply chains, or sharper customer insights, so ROI can be tracked rather than assumed.

MathCo has helped clients unlock more than $800 million in business value. When you evaluate AI and analytics investments, what metrics or outcomes matter most, and how should organisations measure success beyond traditional ROI calculations?

The first wave of AI adoption taught us that traditional ROI math misses the real story. Many proof-of-concepts never scaled because the technology was nascent, leaders assumed buying a platform solved the problem, architectures became obsolete within months, data and context readiness were overlooked, and quality was hard to measure, leading to unreliable outputs and poor user experience.

So, measuring success today means looking well beyond initial ROI calculations. Here are a few things that matter:

– Owning your AI blueprint: Don’t outsource your strategy, brainstorm with partners but design an architecture tailored to your specific use cases.

– Breaking the silver-bullet mindset: a multi-tech ecosystem is fine, and the right tool should serve the right purpose rather than betting on one platform to solve everything.

– Engineering AI-ready data: Good data isn’t automatically AI-ready, and clean, connected, accessible pipelines require upfront investment.

– Mastering context engineering: Poor context is one of the biggest causes of AI failure.

– Keeping implementation modular: Build reusable components, demonstrating value through early wins, and maintaining the flexibility to integrate new capabilities as the AI landscape evolves.

– Balancing build versus buys: Create what’s contextual to you and buy what’s already commoditised.

Together, these outcomes, not just dollar figures, define lasting value.

One of the persistent challenges in digital transformation is connecting data and technology investments to actual business outcomes. How can organisations better bridge this gap and ensure that AI initiatives are aligned with strategic business objectives from the outset?

If there’s one thing we have learnt is that technical expertise alone is not enough to bridge this gap. Building a great model or product isn’t enough; the real question is whether it gets used and drives impact. That means going beyond the build phase to actively understand business context, validate that the solution solves the right problem, and ensure end-users will actually adopt it.

A real example illustrates this: we once partnered with a manufacturing firm to implement predictive maintenance analytics. The model worked perfectly, but almost no one used it. We had to go back and address the human side by embedding analysts within operations teams, rethinking how insights were shared, and creating “analytics ambassadors” on the ground. Only then did real adoption happen.

At the heart of most solutions are human beings who need to see value in what’s been built. Whether developing talent or building technology, a people-first approach matter, always.

India has rapidly emerged as a global center for AI, data science, and analytics talent. What unique strengths does India bring to the global AI ecosystem, and how do you see the country’s role evolving over the next decade?

India’s strength has always been and will continue to be the workforce we have. We have accumulated knowledge over the years of the data, the business as well as the technology which is needed in this new world of AI for making real impact happen. The evolution will be in terms of how the impact is being made to happen and what skills are needed consequently.

Core skills that have always held value like front loaded thinking, depth and expertise in specific areas, ability to learn and adapt really quickly and ability to communicate effectively will continue to hold value with some new skills like the ability to iterate better, token management will become important. India can be the engine that powers AI impact and implementation due to our unique position over the years.

As enterprises navigate an increasingly AI-driven future, what do you believe will define the next phase of innovation in decision intelligence, and how should business leaders prepare their organisations to remain competitive in this new landscape?

The next phase of decision intelligence won’t be defined by chasing what’s new and exciting in tech; it’ll be defined by who actually closes the gap between building and using. The real challenge isn’t creating more advanced technology; it’s making sure the people it’s built for can use it to drive real outcomes.

At MathCo, we have learnt, sometimes the hard way, that success doesn’t come from powerful solutions alone. It comes from ensuring they are adopted, understood, and embedded into everyday decisions, and that’s the focus that will keep organisations relevant and future-ready, not chasing every new trend.

So how should leaders prepare? Here are a few suggestions from my side:

– Start with the decision, not the data: Before building anything, ask how decisions are made and what people need to know, and when.

– Define value in business terms: Value should not be defined in impressive models, but specific metrics that show clear value upfront.

– Think about the whole solution: Treat technology, process, and change management as one integrated system, since adoption never happens by accident. Build adoption into the analytics itself, fitting into how people already work rather than forcing wholesale change. And design for iteration, with feedback loops that improve both the analytics and how it’s applied over time.

The next competitive edge won’t come from features; it’ll come from fit, paired with talent that’s continuously updated to use new technologies effectively.