The Power and The Pitfalls: Understanding Descriptive, Predictive, And Prescriptive Models

Descriptive, predictive, and prescriptive models are the three types of analytics models used in business planning. Descriptive models mine data to show what happened. Predictive models use historical data, statistics, and AI to project what will happen. Prescriptive models, also called optimization models, recommend the best course of action. Each has distinct strengths and limits.

Descriptive, Predictive, and Prescriptive Models At A Glance

Why Does Trust in Models Require Transparency?

Trust in a model requires knowing its purpose, its strategy, and what it leaves out. It does not require knowing the math. The clearest illustration comes from aviation, not supply chain.

On January 15, 2009, US Airways Flight 1549 struck a flock of Canada geese shortly after takeoff from New York’s LaGuardia Airport and lost thrust in both engines. Captain Chesley “Sully” Sullenberger and First Officer Jeffrey Skiles glided the Airbus A320 onto the Hudson River, and all 155 people aboard survived.

During the investigation, the National Transportation Safety Board (NTSB) ran flight simulations. Pilots who turned back toward LaGuardia immediately after the bird strike could reach a runway. When investigators added a 35-second delay to account for the time real pilots need to recognize the emergency, assess it, and decide, the return attempts failed. The NTSB’s final report (AAR-10/03, adopted May 4, 2010) concluded that the decision to ditch on the Hudson gave the highest probability of a survivable outcome.

The 2016 film Sully dramatized that moment: simulated flights that ignored the human factor made a return to the airport look easy, and the conclusion reversed once the human factor was added. The simulation was not wrong about the physics. It was wrong about the scope. A model that leaves out a real-world constraint can give a confident answer to a different question than the one being asked.

The same risk applies to every forecast and optimizer used in planning. Planners and executives do not need to read the equations, but they do need to know what the model is for, how it reaches its answer, and what it assumes.

What Are the Three Types Of Mathematical Models?

Business analytics is commonly divided into three types: descriptive, predictive, and prescriptive. INFORMS, the largest professional society for operations research and analytics, uses this three-part framework. Gartner’s Analytic Ascendancy Model adds a fourth stage, diagnostic analytics, which explains why something happened.

What are descriptive models?

Descriptive models summarize past or current events by aggregating and mining data to reveal patterns. They answer the question “What happened?” Business reporting is the most common example: graphs, charts, dashboards, and KPI scorecards that show shipments, fill rate, inventory turns, or forecast error by product and region.

Descriptive models can be complex, such as mining millions of transactions to segment customers or classify SKUs by volume and variability. Because they describe data that already exists, their results are usually close to 100% accurate, provided the underlying data is correct.

What are predictive models?

Predictive models analyze historical data and apply statistical or machine learning algorithms to estimate what is likely to happen. Demand forecasting is the classic supply chain example. Two families of methods dominate:

• Statistical models, such as moving averages, exponential smoothing, and ARIMA, rest on explicit assumptions like stable seasonality or a continuing trend, and are often fit on limited history. When those assumptions break because of a new product, a lost customer, or a market shock, the forecast degrades.
• Machine learning models are more flexible and can learn from many demand drivers at once, but they are data-hungry. Their accuracy depends on careful data selection and preparation: cleansing outliers, separating promotions from baseline demand, and choosing inputs that will still be available when the forecast is made.

No predictive model is infallible. Every forecast carries error, which is why good planning teams measure it with metrics such as root mean square error and forecast bias, then plan buffers around it.

What are prescriptive models?

Prescriptive models, also called optimization models, recommend the best course of action. They use mathematical solvers to trade off many complicated options against defined criteria, such as cost, service level, capacity, and margin, and find the best feasible plan. Examples include production and campaign plan optimizers, material-cutting optimizers that minimize scrap, and scenario engines that compare plans under different assumptions.

Their key limitation is computing time. Many planning problems are combinatorial: the number of possible plans grows exponentially as products, plants, and time periods are added, and some belong to a class of problems for which no known algorithm guarantees a fast optimal answer why some planning problems are hard.

The practical risk is a gap between compute time and decision time. If leadership wants five scenarios by Friday and each scenario takes seven days to solve to optimality, the model cannot support the decision as designed. The usual remedies are to cap solve time and accept a near-optimal answer within a stated gap, simplify the model, or break the problem into smaller pieces, and then tell decision-makers which trade-off was made.

How Do Descriptive, Predictive, And Prescriptive Models Compare?

Descriptive, predictive, and prescriptive models answer different questions, use different techniques, and fail in different ways. The table below compares them side by side.

How Are These Models Used In Supply Chain Planning?

In supply chain planning, each model type maps to a distinct part of the planning cycle, and the output of one feeds the next.

• Descriptive models handle performance reporting and root-cause analysis: service level, inventory turns, forecast accuracy, and the exceptions planners need to investigate.
• Predictive models drive demand forecasting with statistical methods or machine learning, usually inside demand planning software.
• Prescriptive models power supply planning, production scheduling, multi-echelon inventory optimization, and scenario optimization, turning the forecast into a feasible plan.

The cycle closes in sales and operations planning (S&OP): descriptive reports show how the last plan performed, the predictive forecast sets the demand picture, and prescriptive scenarios give leadership options to choose from.

How Do You Use Models Safely for Critical Decisions?

No model beyond the simplest is infallible. Using descriptive, predictive, and prescriptive models safely for critical decisions means understanding, in business terms, what each model does, what it assumes, and where it stops being reliable.

Explainable AI (XAI) is the set of methods that make a model’s outputs understandable to the people who rely on them. The U.S. National Institute of Standards and Technology (NIST) defines four principles of explainable AI in NISTIR 8312 (September 2021):

• Explanation: the system delivers evidence or reasons for its outputs.
• Meaningful: the explanation is understandable to its intended users.
• Explanation accuracy: the explanation correctly reflects how the system produced the output.
• Knowledge limits: the system operates only under the conditions it was designed for and signals when it is outside them.

The knowledge limits principle is the Flight 1549 lesson in formal terms. Before acting on any model, planners should be able to answer six questions:

• What question does the model answer, and what decision does it support?
• What data, time horizon, and level of detail does it use?
• What does it assume, and what does it leave out?
• How has it performed against actual results in backtesting?
• How long does it take to run, and does that fit the decision deadline?
• Who owns the model, and who can override it?

Consultants and software implementers share this responsibility. Their job is not only to configure the model but to teach users what it can and cannot do, so that trust in its output is earned rather than assumed.

Frequently Asked Questions

Plan With Models You Can Explain

Planning software should make its models easier to trust, not harder. Arkieva Demand Planning and Arkieva Supply Planning pair statistical, machine learning, and optimization models with the transparency planners need to see why a forecast or plan says what it says. You don’t need another platform. You need a partner who understands the realities you manage every day.