The missing piece in enterprise AI

Enterprise AI is currently at an awkward inflection point.

There has rarely been this degree of investment and executive attention around a technology, as is the case with AI. Companies have access to powerful LLMs, have built data foundations, deployed machine learning models, rolled out co-pilots, use GenAI in their work, and are now building agents.

And yet, if you ask how much has the business fundamentally changed, the answer is not as impressive as the excitement around AI. Barring exceptions, most enterprises have experimented with and deployed a breadth of useful solutions. But this is often not the same as building an intelligent enterprise.

I increasingly believe that the next phase of enterprise AI will be won by companies that reimagine how the business operates when agentic intelligence is abundant—and this is different from building use cases.

The gap between those two is, in my view, the missing piece.

Use Case Strategy

Over the past 15 years, I have seen the playbook for successful enterprise data & analytics initiatives:

Prioritise business problems using a value-complexity framework: Build point solutions with right data and ML models.

Measure the impact, and refine with feedback: This approach delivered value. It ensured that teams solved business problems of value. This logic has been expanded into the new age AI era as well. Prioritising use cases and solving them seems logical, but this approach has created a structural problem.

Forecasting systems understand demand. Pricing engines guide pricing. Marketing models understand customers and optimize spend. A GenAI tool understands a set of documents. A co-pilot helps an individual with specific tasks. An agent automates a workflow.

All looks good and each activity is valuable independently. When intelligence is niche, this is the right approach. But when intelligence is pervasive, this approach doesn’t unlock the full potential.

As intelligence starts to look like a commodity, we are discovering that the challenge is less about building intelligence. It is about connecting intelligence to the operating fabric of the business.

Understanding Businesses

How do individuals in an enterprise make decisions? They don’t start with a blank slate, or a blank prompt. They know the organisation, their customer base, the product offerings, the market conditions, the unit economics, the organisational constraints, regulatory compliance etc.

They know what matters more and what is optional. They know what changes to look for. They debate between alternatives. Then they act via a network of processes, approvals, systems, and controls.

Building intelligence via AI in an enterprise, means we need AI to understand all this—the business context. If AI is going to meaningfully enable companies to run, it needs a rich understanding of the business itself. It needs to know the entities that matter, how they relate, what metrics define success, what constraints exist, who is responsible, what are the incentives and penalties, which actions are permissible, and more.

AI needs business context, not just data.

The Missing Layer

Most enterprises today have two broad layers:

The technology foundation: data platforms, cloud infrastructure, models, AI platforms and governance.

Applications: dashboards, ML models, copilots, workflows, agents etc.

The missing piece is the layer between them—the layer that understands the business context. This enables enterprises the opportunity to develop an AI system that can continuously sense what is happening, decide what might need to happen next and increasingly operate through people, agents and enterprise systems.

This is not another application. It is closer to a business operating layer. And this will be far more important than the current bragging rights of having hundreds of disconnected agents.

Business Operating Systems

Imagine an underwriting function that does more than run a risk model. It assembles customer information, understands policy context, transaction history, existing and missing evidence, evaluates risk, recommends terms, and coordinates across people and systems. Or imagine a revenue growth function that monitors demand, sales, competitors, pricing, promotions and inventory, detects emerging issues, and helps execute an intervention.

Technically, this is much more than what one model or agent can accomplish. It needs an ontology of how customer, policy, product, transaction, people, process, metrics, outcomes etc are connected. It needs a memory of prior interactions and decisions. It needs multiple agents, which leverage reusable skills. It needs connectivity and integrations into a breadth of enterprise systems. And around all this are permissions, controls and decision rights.

This is not a collection of AI use cases. It is a business operating system. Achieving this will entail building the context, memory, ontology, skills and controls that allow many agents and many people to operate from a shared understanding of the business.

That is when AI starts operating as the enterprise fabric, rather than being yet another application.

The Next Phase of Enterprise AI

Strong data and AI foundations will be essential. Point solutions and use cases will continue to matter. Co-pilots and agents will become increasingly useful.

But this doesn’t automatically lead to the type of transformation that is possible. Transformation happens when data products, models, LLMs, agents, workflows and enterprise systems begin to operate through a shared layer of context, memory, orchestration and controls. This is how intelligence becomes part of the operating model of the business itself.

That is where the real value of enterprise AI will emerge.

(The author is co-founder and CEO-India, Tiger Analytics. Views are personal.)