by Rohit Kapoor
In a world where knowledge is commoditized, what really distinguishes your business from everyone else’s?
That can be a jarring question to ask the CEO of a Fortune 500 insurance, healthcare or banking company that’s spent the last century as a leader in information arbitrage. But now that many of these leaders have seen the power of AI to instantly analyze massive corpuses of data, produce fully formed business rationales and outperform armies of analysts, the idea that general knowledge will soon be universally accessible is not hard to imagine. Accordingly, businesses need to shift their focus on AI away from chasing the most powerful large language model (LLM) and toward developing models that think like their business.
Resisting technology commoditization
We’ve seen this phenomenon before. Cloud infrastructure was once a strategic differentiator before it became a utility. SaaS solutions, data platforms, and mobile apps all started as strategic investments that gave leaders a first mover advantage, and then they quickly became table stakes. The same thing is happening now with enterprise AI adoption, where the race to adopt the latest, greatest frontier model is rapidly ceding ground to a focus on how businesses can differentiate when everyone has access to the same tools.
Put simply, business leaders need to find a way to preserve their special sauce – their unique approach to problem solving, client experience and overarching value proposition – in a world where technology is increasingly commoditizing the execution of core operational tasks.
This is part of the enterprise AI adoption journey that’s creating the biggest challenges for many business leaders today. EXL’s 2026 Enterprise AI Study found that while 76% of companies believe they’re ahead of competitors on AI, only 10% actually qualify as leaders with measurable business impact. The AI tools have proven capable in a vacuum, but when it comes to integrating them into specialized enterprise models – the systems that carry a company’s encoded institutional judgement: its expertise, decision patterns, operational know-how and domain knowledge – they fail to deliver.
Decision data: A new strategic asset
Overcoming that hurdle requires a fundamental rethink of the AI value equation away from basic productivity and accuracy goals, toward a focus on better judgement, encoded and scaled based on each business’s unique operating model.
Building this type of model does not mean training a foundation model from scratch. The foundation already exists in the form of the business’s existing operations. The objective needs to be customizing a general-purpose model based on each enterprise’s unique data, domain expertise, business rules and institutional judgment to produce a model that understands not just isolated functions or a basic business logic, but the manner in which those pieces intersect in a real-world business context.
In insurance, for example, AI models need to understand not only what a claims denial is, but why it occurred, when exceptions apply, and how experienced practitioners have decided in the past. Similarly in banking, models need to incorporate the specific criteria applied to lending decisions, the macro and micro economic variables at play and the larger portfolio risk context at any given moment in time.
Specialized models become a strategic asset class
Until recently, most industry analysts and corporate technology teams have suggested that the central limiting factor to scalable AI adoption has been structural – that fragmented data, legacy systems and overly rigid governance restrictions have made it difficult to fully embed AI into complex workflows. While those issues are real, continued innovation around data migration and API integration make it much easier to break down the artificial silos between datasets and business functions.
The bigger AI challenge today is the cultural one. Before any AI model can be fully integrated into enterprise workflows, it needs to be able to think like an employee.
Based on my experience working with dozens of financial services, insurance and healthcare leaders, I’m seeing a clear set of best practices emerge around how to build that type of strategic advantage with AI. It’s about taking a portfolio approach to enterprise AI models, treating each like a strategic asset class that delivers on a prescribed set of goals unique to each business.
Why now and why waiting is a mistake
The urgency to figure this out quickly is growing. We’re at a critical juncture in the evolution of enterprise AI where three key things have changed. First, foundation models have become basic infrastructure for many companies. They are no longer a novelty being embraced only by the most technologically advanced businesses – everyone from the start-up to the multinational enterprise has access to the same foundational tools.
Second, the economics of specialization have collapsed. In the early days of generative AI, there was a great deal of hand-wringing and programming time spent trying to fine tune LLMs for specialized tasks. Now, as the AI focus shifts from answering questions to making decisions, judgment matters more than raw knowledge, which is why Gartner expects small, task-specific models to be used three times more than general-purpose LLMs by 2027.
Third, there is the realization that today’s breakthrough new LLM is tomorrow’s outdated artifact. Every model available today will soon be superseded, but the workflows and processes built on top of those models will compound. The decision data, company-specific ontologies and unique approaches to problem solving learned by models are becoming the specialized weights that will be applied to all future models.
A checklist for business leaders
To get over those hurdles, leaders need to start asking three critical questions about how their companies are using AI:
- Are you capturing your workflow decision data: not just the answer, but the reasoning behind it? That decision data is the critical component to training a model a competitor cannot replicate.
- Is your AI strategy focused on building a durable layer, including the underlying data, unique ontologies, evaluation criteria and governance controls, or are you just trying to gain access to frontier models to do things faster and more inexpensively? If the latter, what will happen when access is re-priced or pulled overnight?
- Which of your decisions are so proprietary you’d never let a general model make them? That is exactly where the leaders will emerge in the next phase of AI-driven innovation. Those businesses that can develop specialized models to work reliably within their most complex, proprietary workflows will pull ahead of their competitors who are still treating AI as an efficiency tool.
Building today for what’s next
As advanced AI functionality becomes the baseline, the playing field will level dramatically — smaller, nimbler start-ups will be able to challenge incumbent industry leaders on speed and functionality alone. What won’t be so easy to replicate is the accumulated expertise, insight and culture that have always given industry leaders their edge. As AI becomes a bigger part of enterprise workflows, true leaders will be the ones who find a way to build that DNA into their models. Their futures depend on it.
About the author
Rohit Kapoor is chairman and CEO of EXL, the global data and AI company.
