As the name itself explains, a model-driven decision support system utilizes a model to solve problems or aid in decision making. A model can be statistical, financial, mathematical, analytical, simulation or optimization. A model-driven DSS may employ a single model or a combination of two or more models, depending upon the specific needs of its users. Simple models provide basic functionality while combination of two or more models lets users analyze complex data.
Model-driven DSS are generally not data intensive. Rather they use parameters entered by decision makers and help them analyze a situation. They generate optimal solutions that are consistent with time and resource constraints. The scope of model-driven DSS is huge and can be further enhanced by integrating web-based applications.
When developing proprietary MDSS, it’s important to understand modeling and analytical tools, their working and scope. Building model-driven DSS requires a considerable level of expertise. Managers and DSS analysts need to work closely to develop an efficient system, which is scalable, versatile and easy to integrate and use.
Modeling Decision Situations
Model-driven DSS can be used to aid decision making in a variety of situations. It can assist managers in making:
* Credit and lending decisions
* Product demand forecasting
* Budgeting decisions
* Marketing decisions
* Production forecasting decisions
* Resource allocation decisions
* Project planning
* Investment decisions
Each MDSS has a clear objective and specific purpose. It deploys a model. Consequently, a lot of thought goes into deciding what models should be included in a model-driven DSS. MDSS usually carries out sensitivity analysis or ‘what if’ analysis. However, the users must remember that the system doesn’t make a decision. It only generates alternatives that are to be analyzed and assessed by decision-makers.
How to Build a Model-Driven DSS ?
The most important aspect of a model-driven DSS is the model it uses for decision making. This means that the selection of a model is the most crucial step in building an MDSS. So, how you go about it? Let’s understand:
Modeling is the process of identifying an appropriate model for a prospective model-driven decision support system. It goes through following phases in a chronological manner, beginning from problem identification:
Once modeling is done, it’s vital to validate the selected model, to ensure it works well and generates appropriate results. Model validation is done by comparing model’s output and the actual behavior of the event.
Assumptions & Forecasts
Assumptions are predictions or best guesses. Each model has certain assumptions about the time and risk involved in a particular situation. These results are tested through sensitivity or what if analysis.
Assumptions play an important role in defining a problem and identifying and dealing with uncertainty. Decision makers form a hypothesis and attempt to predict results. Basis the outcome, a hypothesis is either accepted or rejected. Model-driven DSS are designed assuming any of the analyses – static and dynamic.
1. Static Analysis: This type of analysis doesn’t take into consideration the long term response of a system. It takes a single snapshot of a situation and assumes that it will remain stable all through and won’t change. Static analysis is done when a situation in which company makes a decision is static in nature.
3. Dynamic Analysis: Dynamic analysis is testing a program or a software system in real-time. This method considers that the situation changes over time, due to any reason, such as cost, rules and regulations, time, etc.
What kind of analysis needs to be conducted depends upon the situation. Decision makers and DSS analysts must identify whether it is appropriate to assume certainty, uncertainty or risk in a situation.
Certainty * When adequate information about a situation is available.
* Models based on certainty/static analysis tend to yield optimal solutions.
Uncertainty * When information available is vague, unpredictable or unreliable.
* It’s important to acquire more information to find an appropriate model.
Risk * When information is missing.
* What if analysis is carried to aid decision making.
As mentioned earlier, each model-driven DSS works on some kind of model or a combination of models. Therefore, knowing about various models pays off. A DSS deploys one or combination of below models:
1. Explanatory/Descriptive Model: Describes and explains why something is the way it is and why and how it works.
3. Contemplative Model: Forecasts results or outcomes that may be produced from a specific set of parameters.
5. Algebraic Model: A high-level modeling system for solving complex equations. It is employed to optimize a variable or equation. The best part is that it can handle several simultaneous equations.
A DSS with any one of above models performs a single function whichever it is meant to do while a DSS with multiple models is a complete system to perform all three tasks, including:
* Identifying relationships between variables
* Forecasting results based on changes or parameters
* Deciding to what extent a variable can be manipulated
1. Accounting and Financial Models
These model-driven decision support systems aid in decision making in various situations related to accounting and financial management. The examples include: * Break-Even Analysis: A DSS with break-even analysis model aids managers in determining a break-even point for a product. It helps establish a what-if selling price and analyzing the relationship between various related components – prices, marketing spend and profits. The process begins by assuming fixed and variable costs. Profit is set at zero. It helps determine a break even cost of a product at which the company is neither loss nor makes profit.
* * Budget Financial Model: DSS with budgeting model is typically an enterprise-wide application. Many companies use such systems for budget planning and forecasting.
* * Pro Forma Financial Statements: A DSS with this model summarizes the anticipated financial results for a specific time period in future. Costs are estimated based on past data, gross sales are predicted and profit or loss is then calculated on these relationships.
* * Ratio Analysis: This helps a business in evaluating its financial statements. Ratio analysis makes financial data more meaningful, by showing logical relationships between data.
2. Decision Analysis Models
The main job of decision analysis models is to identify and evaluate alternatives with their respective pros and cons. The decision makers then evaluate all the alternatives and pick the one that they think is the best. The aim of decision analysis techniques is to: * Decompose and restructure the problems
* Help decision makers gain in-depth understanding of the problem
* Separate facts and figures from preferences and priorities
* Help users study the performance of decision alternatives
* Avoid citing priorities that don’t help in decision making The following are various types of decision analysis models: . Analytical Hierarchy Process (AHP): It’s a multi-criteria decision technique that combines quantitative and qualitative factors when evaluating alternatives. The analytical hierarchy process begins with developing a hierarchical representation of a problem, with the overall objective on the top, decision alternatives at the bottom and relevant attributes and selection criteria in between. . . After you write decision alternatives at the bottom, you need to compare the alternatives by generating relational data. Consistency ratio is calculated after comparing relative priority of each attribute. The alternatives with the highest priorities and topmost objectives are then displayed. . Decision Trees: As the name suggests, a decision tree uses a tree-like flowchart of decisions, draw from left to right, with further branches explaining their consequences, cost involved, event outcomes and utility. The aim is to identify the most appropriate strategy to reach a goal. A decision tree has three types of nodes . * Choice node: represented by a square * Chance node: represented by a circle * End node: represented by a triangle . The nodes and decision rules are the building blocks of decision trees. The decision trees are simple to understand, offer valuable insights, determine the best and worst scenarios and can be combined easily with other decision techniques. . Multi-Attribute Utility Analysis (MAUA): Multi-attribute utility analysis gives much importance to attribute weights. The information is provided about each decision choice on each attribute. A decision maker then perceives the utility of usefulness of a decision alternative in terms of its attributes. This method is generally used when the attributes of an alternative are certain. . . Influence Diagrams: It’s a diagrammatic representation of a decision situation, to express the precise nature of relationships between variables. It uses geometric shapes to represent various elements. . * A decision variable is represented by a rectangle. * An intermediate variable is represented by a circle. * A result or outcome variable is represented by an oval.
3. Forecasting Models
Forecasting models form an integral part of a large number of decision support systems. Their main job is to predict the value of interrelated variables at some point of time in future. The two main types of forecasts are: * Short run forecasts: where the prediction will be used anytime soon mainly in deterministic models
* * Long run forecasts: where the prediction is used for long term investment/planning decisions
* Forecasting may include ambiguity as factors on which decisions depend are uncontrollable and dynamic in nature. This means that the accuracy of data and time taken in making near-perfect predictions matter a lot. The following are various types of forecasting models: . Naive Exploration: As is explained by the name itself, naive exploration is not a sophisticated prediction. Rather it is simple forecasting that provides limited accuracy. The technique is implemented using a spreadsheet. . . Judgment Methods: The predictions or forecasting are based on the perceptions and opinions of experts instead on hard data. It’s a subjective estimate used for long-run forecasts where external environment plays a critical role. The results are not very accurate. . . Moving Average: Used for short-run forecasts, the predictions are based upon the historical values. DSS with this model is inexpensive and easy to use. . . Exponential Smoothing: Used for short-term forecasts, it alters the historical data mathematically to better reflect the assumptions of a decision maker. Similar to moving average model but claims to obtain better results using exponential smoothing. . . Time Series Extrapolation: This method takes into account the economic variables that are measured at consecutive intervals of time. It is believed that the knowledge of past behavior of the variable at successive intervals of time will help understand the behavior of the variables in future better. . . Regression and Econometric Models: These types of forecasting models make use of linear and multiple regressions to establish cause and effect relationships. These methods are considered more powerful than time-series but also complex at the same time. They are complex because they use sophisticated models and include more variables. The results obtained are more accurate. .
4. Network and Optimization Models
Network and optimization models are integrated into a DSS when decisions regarding resource allocation, project control, location, scheduling, transportation, distribution, size, shortages, multinational cash flow management, inventory management and distribution and network need to be made. For example: * The best location for an operation or manufacturing
* The resources needed to carry out the operations
* Most suitable aircraft route to transport products Network and optimization models typically use linear regression technique, which falls in the class of mathematical programming tool. Using this technique, problem solvers can find the best set of values that minimizes or maximizes a specified calculated formula. A linear programming situation consists of six elements, including: * Decision variables, the value of which we try to find by applying the model
* * Objective function, a mathematical expression showing linear relationship between the goal and decision variables
* * Coefficients of objective function, the variables that express the pace at which the value of the objective function alters (increases or decreases) when the values are included in the equation
* * Constraints, the linear inequalities reflecting the fact that the resources are limited
* * I/O (Input-Output) Coefficients, the coefficients of constraints which indicate the pace at which a given resource is utilized/depleted
* * Capacities, which express the minimum resources needed
* Remember that it’s the managers who determine what ‘best’ means for them.
5. Simulation Models
DSS with simulation models conduct experiments to identify conditions or situations that approximate the actual conditions. These models are utilized to solve a number of problems, including * Manpower planning and assignment
* Inventory control
* Reliability and replacement
* Sequencing and scheduling
* Stock-in and stock-out
* Queuing and congestion Simulation models: * Try to imitate reality
* Perform what-if analysis
* Are descriptive tools for forecasting
* Repeat experiments to obtain an optimized estimate of impact of certain actions
* Aid in solving extremely complex problems
* Form elementary relationships and interdependencies among variables
* Are made for one problem and aren’t suitable for another problems
* Reduce the time taken in decision making Simulation Methodology The process goes through a number of steps, beginning from problem identification and ending at evaluating the results. Simulation models are of following types: . Probabilistic: In this method, experts conceptualize one or more independent variables as a probability distribution of values . . Time dependent: Also known as discrete simulation, it takes into account the exact time of the occurrence of an event . . Visual Simulation: This method uses visuals and animations of results to foster quick and deeper understanding. .
Modeling Languages and Spreadsheets
As models are computerized software program, a number of programming languages can be used for coding. Typically the languages used are C++ and Java. Moreover, the decision support systems make use of spreadsheets, allowing users to
* Write values
* Manipulate data
* Apply mathematical and statistical formulas
* Create graphs and visuals
* Prepare, consolidate and sort reports
There are numerous software packages available for model-driven decision support systems. However, you need to carefully select a package. You must ensure that it meets all your specific needs. Reputable packages allow you to create your own models and manipulate the existing ones.
Building a customized model-driven DSS is a complex, time consuming and expensive process. However, the end decision of buying a package or develop a DSS lies with you.