The Missing Layer in Enterprise AI

For the past few years, enterprise technology has been animated by a seductive promise: connect powerful AI models to corporate data, and the organization itself becomes smarter.

It is easy to see why the idea took hold. Give an AI access to financial reports, customer records, operational metrics, market data, contracts, emails, and forecasts, and it should be able to answer almost anything a business might ask. In the imagination of many executives, this is the beginning of a new operating model: faster analysis, better decisions, fewer silos, less friction.

But as AI moves from impressive demonstrations into the machinery of real business, a less glamorous problem is becoming harder to ignore. Many companies now have AI systems that can answer questions, yet far fewer have systems that can make confident decisions.

The difference matters. A model can explain why revenue fell in a region or summarize a forecast. It can detect a pattern in sales, inventory, or margin. It can even recommend a course of action. But in an enterprise organization, a recommendation is not the same as a decision.

For instance, a forecast is less of a prediction than a commitment between finance, operations, supply chain and commercial teams for what the business believes will happen and what it will do in response. A merchandising plan is not a spreadsheet with better math. It is a web of choices about inventory, pricing, margin, customer demand, supplier constraints, and risk. A supply chain decision can ripple through factories, logistics networks, service levels, working capital and financial guidance.

These are organization-spanning challenges that chip away at the dream of “AI on top of everything.”

The theory sounds reasonable enough: if the model can see everything, surely it can understand everything. But in practice, the more the AI sees, the more ambiguity it inherits.

Revenue may mean one thing to finance and another to sales. Product hierarchies may differ between planning, merchandising, and supply chain systems. Forecasts may exist in several versions, only one of which is approved. Ownership may change by region, business unit, channel, or process. Some assumptions are live. Some are locked. Some are politically sensitive. Some are simply wrong, but everyone knows not to use them because of their company’s tribal knowledge.

Humans navigate this corporate mess remarkably well through experience. They know which numbers matter, which exceptions can be ignored, which rule overrides another, and which decision needs escalation before it becomes a problem.

AI can’t know these unwritten rules without being given the context, which is precisely why many enterprise AI initiatives stumble.

The technology industry has begun to recognize at least part of this. The rise of semantic layers, knowledge graphs, ontologies and governed data products reflects a growing awareness that raw data is not enough. AI needs meaning, not just access. Still, data semantics are not decision semantics.

A data model can describe a customer. A decision model has to describe what happens when that customer changes behavior, demand shifts, supply tightens, margin erodes or a forecast no longer holds.

Taking that a step further, a data platform can help explain what happened, and a decision system can help an organization decide what should happen next.

That difference becomes especially important in enterprise planning, one of the areas most likely to be transformed by AI and least likely to be replaced by it. For years, enterprise planning has been criticized for being slow, periodic, and spreadsheet-heavy. Much of that criticism is fair—annual plans age quickly, monthly forecasts arrive too late, and functional silos produce conflicting versions of reality. By the time a plan is approved, the market may already have moved.

AI is already changing this. Enterprise planning is becoming more conversational, continuous, and signal-driven. Finance teams will ask for variance explanations in plain language. Retail planners will simulate allocation and assortment changes faster. Supply chain teams will model disruption scenarios in real time. Executives will receive AI-generated narratives that connect external signals to financial impact.

While none of this removes the need for a planning system, it does change what that system must become. The future planning platform becomes the place where business context is made explicit: definitions, assumptions, rules, scenarios, approvals, ownership, constraints, and accountability.

This contextual decision layer is the missing piece in much of enterprise AI. Without it, AI risks becoming a brilliant commentator on the business: articulate, fast, occasionally insightful, but not truly embedded in how the enterprise operates. With it, AI can become something more useful: a participant in coordinated decision-making.

That does not mean autonomous decision-making everywhere. But it is a meaningful step toward enabling the autonomous-ready enterprise. The future of planning is not humans versus AI. It is humans and AI working through shared systems of context.

AI will monitor more signals than any team could track and detect anomalies earlier. It will generate scenarios faster, surface risks, explain drivers, and recommend actions.

Humans will still decide what matters, weighing trade-offs, interpreting ambiguity, managing politics, accepting risk, and owning outcomes.

Planning is changing accordingly: less time gathering data, more time exercising judgment; less time producing static plans, more time deciding how to respond as conditions change.

This is the real shift—AI does not make planning disappear. It makes planning more continuous, more connected, and more dependent on trusted context.

If the the first phase of enterprise AI was about intelligence, the current phase has been about data. The next will be about decision context.

We have spent the past few years teaching machines how to reason. The next challenge is teaching them how organizations decide. The companies that solve that problem won't just deploy better AI. They will redefine how decisions are made.

For more information, visit board.com.

By David Marmer, Chief Product Officer, Board