From Prediction to Prescription: The Next Leap in Data Science

Moving from foresight to action, how Decision Intelligence and agentic AI are redefining the value chain of analytics

For much of the past two decades, the story of data science has been a story of prediction. We built models that could tell us what was likely to happen, who might churn, which customer would buy, what product would fail. And for a long time, that was enough. Businesses could plan, optimize, and prepare.

But as anyone working in analytics knows, foresight without follow-through is an incomplete promise. Knowing what will happen is one thing. Knowing what to do about it and doing it at the right time, is another matter entirely.

We’re now entering a new phase in the evolution of data science, one that moves beyond prediction toward prescription: the ability not only to forecast outcomes but to recommend and autonomously execute the best course of action. This is the domain of Decision Intelligence and agentic AI, the systems that don’t stop when the model says “X will happen” but instead respond, “Here’s what you should do next.”

From Insight to Impact

To understand why this shift matters, it helps to look back. Predictive analytics marked a major milestone in the data science journey. It was the move from descriptive (“what happened?”) to predictive (“what will happen?”). But in many organizations, insights from predictive models remained trapped in dashboards, waiting for human translation into decisions.

In financial services, for instance, a risk model might flag customers with a 70% chance of defaulting. Useful information, yes, but unless that insight is linked to a defined set of actions (adjusting limits, contacting the customer, or offering restructuring), its value is capped. The same is true in retail, healthcare, and manufacturing. Predictions abound, but actions often lag behind.

The next leap, prescriptive analytics, bridges this gap. It connects the model output directly to decision pathways, often using reinforcement learning, optimization, and causal inference to simulate possible futures and recommend the most effective interventions. In this world, data science no longer ends with insight; it culminates in impact.

Prediction informs. Prescription transforms

Accuracy is only potential energy; the graph below shows why, great AUC can still deliver middling ROI when insights aren’t wired to decisions:

The Rise of Decision Intelligence

This is where Decision Intelligence (DI) enters the scene, a framework that unites data science, behavioural science, and decision theory. At its core, DI asks not only “what’s likely to happen?” but “how should we respond to achieve our objectives?”

Imagine an airline trying to minimize disruption during adverse weather. Traditional forecasting models predict delays and cancellations. But a decision-intelligent system goes further: it simulates alternate crew schedules, passenger re-routing options, and fuel requirements, then acts on the optimal scenario.

When we optimised the decision policy, not just the score, the cumulative incremental conversions surged, see how the prescriptive curve outpaces the predictive-only approach.

The power of DI lies in its integration of human and machine intelligence. While AI models generate possible actions, humans define objectives, ethical boundaries, and risk tolerance. Together, they form a decision ecosystem that’s transparent, auditable, and adaptable, precisely what regulators and executives are now demanding.

Decision Intelligence is where prediction meets purpose

From Static Models to Agentic AI

Enter the next frontier: agentic AI, systems capable of autonomously executing multi-step decisions aligned with human intent. These are not your static machine learning models but dynamic agents that observe, plan, act, and learn continuously.

Consider an e-commerce platform managing thousands of price and inventory combinations. Traditional analytics might suggest optimal price points once a day. An agentic system, however, can adjust in near real time, responding to competitor changes, customer behaviour, and stock levels, while adhering to pre-defined governance and fairness rules.

In effect, we move from “AI as adviser” to “AI as collaborator.” Models no longer whisper predictions into dashboards; they participate in the decision loop. This is where analytics becomes operational, actionable, and most importantly, value-driving.

This is also where trust becomes paramount. Agentic systems that act autonomously require governance by design, clear lineage, explainability, and oversight. After all, if an AI system can make decisions on behalf of a business, then accountability must be equally intelligent.

The question is no longer can AI decide, but should it?

The Role of Causality and Simulation

As we move toward prescriptive and agentic systems, one discipline quietly reclaims the spotlight: causal inference. Predictive models can tell us correlations, what tends to happen together, but only causal models can explain why something happens and how interventions might alter that outcome.

This distinction is critical. If a model predicts that customers who receive push notifications are more likely to purchase, it doesn’t necessarily mean the notification caused the purchase. Acting on that insight blindly could lead to over-messaging and fatigue.

Causal AI and simulation-based modelling, through techniques such as counterfactual reasoning, digital twins, and synthetic control groups, allow organizations to explore “what-if” scenarios safely. This transforms decision-making from reactive to proactive, enabling businesses to test, learn, and optimize strategies in virtual environments before committing resources in the real world.

Operationalising Intelligence

Moving from prediction to prescription isn’t purely a technical transformation; it’s an organizational one. It demands closer collaboration between data scientists, domain experts, and decision-makers. It requires data-to-decision pipelines that integrate model outputs directly into business processes.

At SAS, we’ve seen this in practice across sectors.

• In banking, prescriptive models drive next-best-action strategies in credit and fraud, dynamically adapting to changing risk profiles.
• In healthcare, decision intelligence optimizes treatment pathways by balancing patient outcomes with resource constraints.
• In manufacturing, digital twins and reinforcement learning agents continuously fine-tune production parameters to maximize yield and minimize downtime.

In each case, the technology is not the bottleneck, adoption is. The most successful organizations treat decisioning as a capability, not a feature. They build closed feedback loops, where every action taken generates new data, refining future predictions and prescriptions.

Intelligent decisioning is not a one-off act, it’s a living system

Human-in-the-Loop, Not Human-out-of-the-Loop

The allure of autonomous AI should never eclipse the human factor. Prescriptive systems may optimize outcomes, but they cannot replace contextual judgment, ethical discernment, or empathy. Decision Intelligence thrives when humans define values and boundaries, and machines execute within those parameters.

In the age of prescriptive analytics, the role of the data scientist evolves too, from model builder to decision architect. Their mandate expands beyond accuracy metrics to encompass interpretability, fairness, and impact. The new success measure isn’t just AUC or R², it’s real-world consequence.

That’s the beauty of this next chapter in data science: it’s as much about philosophy as technology. It challenges us to design systems that not only predict the future but help shape a better one.

The Next Leap

So, what happens after the model says “X will happen”?

It’s where analytics becomes intelligence, and intelligence becomes action. It’s where the lines between data, decision, and delivery blur into a continuous feedback system. It’s where data science transcends its analytical roots to become a strategic co-pilot for enterprise decision-making.

The journey from prediction to prescription is not a simple upgrade, it’s a paradigm shift. It requires us to think differently about what models are for, how decisions are made, and where humans fit in. But the reward is worth it: a future where AI doesn’t merely describe or predict, but actively helps us choose, wisely, responsibly, and in real time.

The future of data science isn’t in predicting what will happen. It’s in deciding what should happen, and making it so

Follow “The Data Science Decoder” for more reflections on AI, analytics, and the evolving human-machine partnership shaping our digital world.