Guest Post: P-Decisions, C-Decisions, and Why the Difference Matters More Than Ever

One of the most persistent sources of confusion we see in organizations that want to improve decision making using AI and/or data science is a quiet category error: treating all decisions as if they were the same kind of problem. They aren’t. In Decision Intelligence, it’s useful to distinguish between pattern-based decisions (P-decisions) and causal, action-to-outcome decisions (C-decisions). This distinction isn’t academic. It shapes what methods work, what risks we take, and whether AI actually improves decision quality.

P-decisions are decisions made by recognizing patterns. Given some inputs, we produce an output—often quickly, often repeatedly—without explicitly reasoning about how that output will change the world. Fire commanders recognizing a familiar incident pattern, call centers routing tickets based on rules, credit models scoring applications, or classifiers filtering spam: all are P-decisions. Some are made rapidly and automatically, without human intervention. Some of these patterns are encoded explicitly as rules or guidelines. Some are learned empirically from historical data using statistical or machine-learning methods. And still others live implicitly in human heads as “experience” or “gut feel”.

What unifies P-decisions is not how the pattern is created, but what it does: it maps observed situations to responses without modeling downstream consequences. P-decisions don’t ask “what happens if we do this instead?” They ask, “what usually works in situations like this?”

That’s not a weakness—it’s a feature. P-decisions scale expertise, enforce consistency, and operate well under time pressure. They are the backbone of modern operations. But problems arise when the pattern non longer works: maybe the situation has changed (so our experience or data is no longer relevant). Or maybe we’ve never seen this particular situation in the past. In these situations, need causal reasoning.

That brings us to C-decisions.

C-decisions are about choosing actions in the real world and reasoning about their consequences over time. The decision is not just an output or conclusion—it’s an intervention. Pricing strategies, infrastructure investments, policy choices, market entry, capacity planning, reputation management, and many operational decisions all fall into this category. These decisions are forward-looking, counterfactual, and often irreversible. They require an explicit representation of what can be influenced, what cannot, what is observed, what is latent, and how effects unfold.

Here’s a little video about that.

C-decisions depend on world models—sometimes formalized as simulations, optimizations, or digital twins, and often organized around Causal Decision Diagrams (CDDs). These models don’t replace human judgment; they structure it. They make assumptions visible, trade-offs discussable, and uncertainty explicit. They also explicitly model the decision context or environment—the external factors like weather, the global economy, or competitor behavior that can influence decision outcomes. Once decisions leave the digital system and touch the physical, economic, or social world, this explicit modeling of how events play out in reality matters.

P-decisions encapsulate wisdom from the past. They make the implicit assumption that the world or the decision context is stable and consistent. But external changes can undermine expertise and rule systems, as we have recently seen as new tariffs force a rethink of supply chain decision making. On the other hand, C-decisions, especially when simulated, let us explore multiple possible futures to observe outcome sensitivity to various external changes.

The P/C distinction is critical as agentic AI moves into the physical world. Many agentic AI success stories showcase internal P-decisions: what to search next, how to summarize a document, how to decompose a task. These are real decisions, but they occur within a stable IT environment. As AI agents begin to make decisions whose outcomes depend on and whose consequences unfold in the world outside the system, we cross into C-decision territory—and pattern matching alone is no longer enough.

The good news is that P-decisions and C-decisions are not rivals. They support each other. Predictive (P-decision) models can live inside causal decision models as components. And repeated execution of causal models can generate synthetic experience that is distilled into new P-decision rules. This mirrors how humans learn: we reason carefully when stakes are high and time allows, and we rely on patterns when speed matters.

The mistake isn’t using patterns. The mistake is using them without acknowledging when causal modeling is a better fit. Recognizing that difference is one of the most important steps organizations can take as they bring AI and data science into consequential decisions.

Read more about Causal decisions in The Decision Intelligence Handbook.