Decision support Concepts

Decision Support and Business Analytics

The lecture introduces how decision-support technologies affect business processes and explains the main types of models and data analysis used to support managerial decisions.

Decision-Support Systems

A decision-support system (DSS) provides interactive, often ad hoc support for managers and other business professionals.

Decisions may be:

• Structured: Sufficiently well-defined to automate.
• Semi-structured: Some parts can be automated, while others require human judgment.
• Unstructured: Rely substantially on human interpretation.

DSS approaches can be broadly divided into:

• Model-driven systems, which use analytical models to evaluate choices.
• Data-driven systems, which use available data to inform decisions.

A typical DSS includes:

• Data gathered from sources such as operational, sales, market, and customer systems.
• A database or data warehouse, which may also hold unstructured data.
• A model base containing models or programs that process the data.
• A user interface through which people access and interpret the results.

The interface might be provided through existing software, a web application, or a separate application.

Why Decision Support Is Changing

New technologies generate data in greater volume and variety. Examples include sensor and GPS data, machine-to-machine communication, and customer contributions such as online feedback and blogs. As customers respond and create content, business data is no longer limited to information that companies send or collect through traditional channels.

These changes have expanded the role of data-driven decision support alongside established model-driven methods.

Main Analytical Model Types

Optimization Models

Optimization models seek values for decision variables that maximize or minimize an objective function while satisfying constraints. The objective may involve one or more goals. Linear and integer programming are examples.

Business applications include procurement, inventory replenishment, and production scheduling. In a reverse auction, for example, optimization can help select among competing suppliers.

Exact methods can become impractical for large problems because of their time and computational requirements. When exact solutions are too costly, analysts can use heuristics or metaheuristics. Heuristics are often designed for a particular problem, while metaheuristics—such as genetic algorithms and ant-colony optimization—provide more general approaches inspired by natural processes.

Simulation Models

Simulation models recreate an existing or anticipated situation so decision-makers can examine how it might behave. For example, an airport can simulate flight arrivals and departures, landing order, and the time people spend waiting or at the airport.

Models are validated against real situations where possible, then used to test different scenarios and compare outcomes.

Probabilistic and Uncertainty-Aware Models

These models estimate the likelihood of events and help identify suitable actions under different outcomes. Models may be:

• Static or dynamic, depending on whether they account for time.
• Deterministic or stochastic, depending on whether they represent uncertainty.

For example, uncertain demand in a procurement decision can be treated as a random variable.

When model parameters are estimated from historical data or decision-maker experience, analysts can vary those parameters to see how results change. This is known as sensitivity analysis.

Statistical Models

Descriptive statistics help analysts understand data. Examples include mean, variance, standard deviation, skewness, and kurtosis. For multiple variables, analysts may use correlation and covariance.

Analysts should check whether data meet the assumptions of the methods they plan to use. If necessary, data may be transformed or outliers examined. Outliers may be removed when they distort an analysis, but they may also represent a meaningful group and should not be discarded automatically.

Predictive statistical methods include univariate and multivariate regression.

Data Mining and Machine Learning

These methods can uncover patterns or knowledge that is not specified in advance. Examples include:

• Classification: May be supervised or unsupervised; neural networks and support vector machines are examples.
• Clustering: Groups data, typically without predefined labels.
• Association-rule mining: Identifies relationships among items or events.
• Collaborative filtering: A predictive approach used in recommender systems.

Text data can also be analyzed. Natural-language processing can help extract features, after which data-mining methods can search for patterns.

Practical Considerations

Developing a decision-support model takes more than choosing an analytical technique. It also requires effort to build, code, implement, and provide an interface that managers can use.

The lecture suggests two situations in which investment may be worthwhile:

1. The decision must be made repeatedly, allowing the initial development effort to pay off over time.
2. The decision is especially important to the business, making accuracy and cost-effectiveness valuable enough to justify the investment.

Evolution of Decision-Support Terminology

The lecture describes a progression in the language used for data-based business support:

• Decision-support systems combined data analysis with analytical support for decisions.
• Executive-support systems brought together information from different sources to help senior executives assess situations rather than dictate a solution.
• OLAP emphasized multidimensional analysis beyond conventional two-dimensional database tables.
• Business intelligence referred to toolsets using data analysis and data-mining methods.
• Analytics highlighted the use and interpretation of statistical and mathematical models.
• Big data refers to working with large volumes of data that may be unstructured or rapidly changing. Data from multiple sources may also be combined.

The lecture concludes by previewing a follow-up session on applying these tools and techniques to business situations shaped by information and communication technologies.

Speakers and Sources Featured

• Speaker: An unnamed instructor delivering the lecture.
• Source mentioned: Thomas Davenport, whose book on big data is referenced in connection with using big data in organizations.