Why AI Needs Enterprise Context, Not Just Enterprise Data

A leading private bank asked us to build an AI capability to help relationship managers identify emerging risks across large portfolios. The system had access to market data, client information, historical transactions, risk indicators, and relevant policies. It was technically well designed. The retrieval architecture was sound. The recommendations it generated were consistently coherent and data supported.

The problem appeared several months into production when two clients received very different recommendations from the same underlying risk signal.

Both recommendations were technically defensible. Both were supported by available data. Both would have passed any reasonable audit of the system's outputs. Yet when the decisions were reviewed internally, the question that surfaced was not whether the system had the right information. It was whether the system understood the context in which that information mattered.

One client was approaching a major liquidity event with a short horizon and significant concentration risk. The other had a long-term investment mandate, a different risk tolerance, and a relationship history that made the same signal read very differently to anyone who understood the account. The underlying market signal was similar. The business context was entirely different. The appropriate action was not the same at all.

The data was the same type of input. The context was what made the difference.

This distinction becomes increasingly important as organizations move from AI experimentation toward embedded intelligence operating across real business decisions with real consequences. Enterprise AI systems are becoming remarkably capable at finding information, combining datasets, retrieving relevant documents, and generating recommendations at speed and scale that no human process could match. But information does not carry its meaning independently of the context in which it matters.

What was the business objective at the time a decision was made? What constraints applied, regulatory, contractual, or relationship based? Which policies were active, and had any exceptions been granted? What had already happened in this account, this workflow, or this market relationship that would change how the signal should be read? What alternatives had already been considered and rejected?

Without that context, an AI system can produce an answer that is factually supported and operationally wrong. Not wrong in the sense that it fails a technical audit. Wrong in the sense that it recommends the same action for two situations that any experienced practitioner would immediately recognize as requiring different responses.

This is why adding more enterprise data does not automatically make an AI system more intelligent. More data improves coverage, the system can find more information relevant to a query. More sophisticated retrieval improves relevance, the system can surface better matched content. More capable models improve prediction, the system can generate more coherent outputs. But none of these investments address the question of context, which determines how information should influence a specific decision in a specific situation at a specific point in time.

We observed this pattern repeatedly across transformation programs involving multiple business units, geographies, regulatory environments, and technology platforms. The organizations that achieved better outcomes were not necessarily those with the largest data estates or the most advanced retrieval architectures. They were the ones that had found ways to connect data to the business circumstances in which decisions were actually being made, preserving the context that made information actionable rather than merely available.

There is a layered way to think about what enterprise AI actually needs to function well. Data tells the system what is available. Memory tells it what the organization has learned from previous decisions, outcomes, and reasoning chains it has accumulated over time. Context tells it what matters now, the objectives, constraints, relationships, and circumstances surrounding this specific decision at this specific moment. Intelligence is what emerges when all three connect to a decision at the moment it needs to be made.

Most AI investments today are concentrated in the first layer. Larger data estates, faster retrieval, more capable models. The more consequential and considerably harder work is in building the other three, and most enterprises have not yet seriously started.

That is the architectural conversation that changes the question fundamentally. Not only: can the system find the right information? But: can it understand which information matters, in this context, for this decision, at this point in time? And beyond that: can the organization build the capability to preserve and connect all of that understanding as its people, systems, and business circumstances continue to change?

That is where the more interesting work begins.

Where do you see enterprise AI losing the most context today, in business objectives, regulatory constraints, historical decisions, or human judgment?

Source: Investment AI platform consulting engagements. Vikas Sharma, Senior AI and Digital Transformation Advisor | linkedin.com/in/sharma1vikas

Research: "From Agentic AI to RAG: A Framework for Responsible AI," BIGS 2025 | aisel.aisnet.org/bigs2025/1/