Bruno Billy, President & CEO of APGAR North America, advises organizations on the operational reality of data and enterprise transformation
getty
Over the past decade, companies have invested massively in cloud platforms, data lakes, data pipelines and increasingly sophisticated analytics environments. From a technology standpoint, many organizations appear to have the foundations required to support AI, but AI is exposing a different reality.
The issue is not whether organizations can store, move or access data. Most organizations can. The issue is whether people, and now AI systems, can understand, trust and use that data consistently in the right business context.
What Actually Happens Inside An Organization
Before an executive meeting, finance controllers reconcile reports that are supposed to match. Sales operations correct customer deals that don't show in the pipeline. Supply chain analysts adjust the metrics because one warehouse always reports late.
The context that makes those corrections possible lives in people’s heads. It almost never appears in process diagrams or documentation. The hidden layer of human interpretation is what keeps the enterprise running, and it is also one reason AI pilots often look so good.
In a pilot, the scope is narrow. The subject matter experts are in the room. They curate the inputs, correct the outputs and adjust the business rules when edge cases appear. They know which exceptions matter and which can be ignored. They can tell the model, directly or indirectly, what counts as signal and what is just noise.
The pilot succeeds because the model is being surrounded by human judgment.
The problem starts when organizations try to roll that pilot out across the enterprise. The AI product now has to navigate the business across systems, teams, workflows and definitions that were never fully reconciled.
When AI Meets Organizational Reality
Consider two recent client examples.
At one client, an AI chatbot pilot worked well in a contained environment but struggled when extended across sales, billing, customer success and marketing. The issue was not just that those teams had different definitions of “customer.” It was also that each team interacted with the customer in a different context: Support managed issues, sales managed opportunities, customer success managed expansion and billing managed payment. Human representatives could move between those contexts almost without noticing. A bot could not.
At another client, AI was used to generate a dashboard of the “top 10 customers at risk.” The system gathered the data and produced a ranked list. Then leaders reviewed it, and the conversation stalled. No one could quite agree on why those customers were there. Was “at risk” about churn, revenue exposure, service issues, payment problems, declining engagement or something else? The model had produced an answer, but the organization had not agreed on the question.
In both cases, the AI appeared to work until the organization needed to act on the output. The service bot struggled because it could not move between customer contexts. The dashboard stalled because “customer at risk” meant different things to different parts of the business.
These are context problems.
They reveal two things organizations often leave implicit: how work actually happens across functions and what business terms must mean before they can drive action.
What Business Leaders Should Be Doing Now
This is why the discussion around AI readiness needs to move beyond infrastructure. Too much of the conversation is still focused on the technology: which models to use, which copilots to deploy, which data platforms to modernize and even how to measure token usage as a proxy for AI productivity.
Those questions matter, but they do not get you very far if the business cannot agree on the meaning of the output. You can measure token usage down to the decimal point and still have no shared answer to the basic question, "What does 'customer at risk' actually mean?"
AI readiness is not just a question of infrastructure, tooling or usage metrics. It is also a question of whether the organization has an operating model for context.
That operating model has to answer three practical questions.
• Who owns the data? Someone has to be accountable for the quality, reliability and usage of critical data domains.
• Who owns the meaning? Definitions like “customer,” “product,” “active,” “complete,” “eligible” or “at risk” cannot live only in systems or reports. They need business owners who can define what those terms mean in the workflows where decisions are made.
• Who mediates between contexts? Sales, billing, support, finance and operations will not always use the same concept in the same way. Nor should they; they operate in different parts of the value chain. The operating model has to define when those differences should be reconciled, when they should remain distinct and who decides when there is contention.
The Bottom Line
The organizations that get this right will not simply have better platforms or more AI activity. They will have a clearer understanding of how work actually happens, where meaning breaks down and what human judgment has been holding together.
AI does not always need more data, but it does need organizations to be clearer about who owns meaning, who resolves conflict and how context is applied at scale.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
>> Home