Optimization

The selections and allocations on the Define tab determine which entities are candidates for which dimensions , and may return many possible solutions. If an entity is eligible for multiple dimensions , which do you choose? Optimization allows you to identify the best or most profitable dimension for each entity within certain contraints. In other words, IBM® Analytical Decision Management optimization defines the best way that entities can be allocated to dimensions.

The goal of optimization is to identify the solution that best meets a specific goal, such as maximizing the revenue from a marketing campaign, or minimizing the risk of fraud or churn. The result of optimization is a solution that answers this question across all possible combinations within your population, in the manner that best maximizes the value you want to optimize (such as profit).

Optimization supports a number of deployment scenarios. For example:

• Real-time optimization. Optimizing allocations for a single entity on a real-time interaction point (for example, best offers to make to a customer who is looking at the web site or who has called in to the call center).
• Batch optimization. Optimizing allocations across multiple entities on a batch interaction point (for example, best offers to make to each customer by the best outbound channel).
• Pre-scoring real-time offers ahead of time on a batch interaction point. For example, rather than performing a real-time optimization when the customer visits the web site, perhaps you would rather pre-calculate what offers should be made to the customer if they visit the web site, or if they call the call center, etc. If the customer does not arrive, then those precomputed offers could expire and can be recomputed on more recent data.

Optimization is helpful for solving problems such as allocating customers to marketing campaigns or offers, allocating claims investigations across a number of investigators, and allocating staff to tasks or shifts, for example.

Applications configured to use complex mathematical optimization (CPLEX), such as IBM Analytical Decision Management for Campaign Optimization, IBM Analytical Decision Management for Operations, and IBM Analytical Decision Management for Demand Optimization, have an Optimize tab.

Note that CPLEX optimization scales in a non-linear fashion. For example, the increase in time and resources required to optimize over 1000 records will more than double when 2000 records are used – possibly taking four times as long or more. This is because optimization is trying to find combinations of records, and the number of available combinations increases non-linearly.

1. Specify optimization parameters in the table as desired. You can simply type in values, or click the icon next to each input to choose a field, browse for an existing model, build a new model, or create an expression which can then be used as input to the optimization. If the application is configured to include more than one decision hierarchy element, each hierarchy will have its own tab containing a parameters table.

   You can also reorder the parameters in the table, but this has no impact on the optimization.

2. Use the Priority drop-down menu to give priority to certain items. For example, offers given high priority will be considered before those that are normal or low priority. The prioritization equation is still computed for all offers, but the selection is done in different groups based on priority order. In other words, all solutions of a given priority are considered before any solution of a lower priority. The priority will take precedence over the object value. For example, if Offer A has a high priority and Offer B has a lower priority but higher expected profit, then Offer A will be selected.
3. Select Customize table to choose at which level you wish to define each parameter. For example, whether to specify parameters at the level, or separately for each within a .
4. Optionally, you can use the same settings for all interaction points, if available. If not, you can deselect this option and choose interaction points separately from the drop-down list. For example, the cost of an offer may differ when made through a call center rather than a retail store location, while a web page might allow multiple offers to be displayed to the same customer. Note that interaction points must be configured by the administrator. See the topic Defining interaction points for more information.
5. The Optimize tab maximizes revenue or maximizes return on investment, based on the optimization equation used. Click Select equation and choose which optimization equation defines your goal. If the application is only configured to include one optimization equation, this option will not be available.

   Equations can be modified for different applications by the application designer. While the specific equation may change, many business problems may be modeled in this manner.

6. In the Constraints section you can select from available constraints to define the boundaries the optimization solution must work within, such as total budget. For example, you likely do not want the amount spent to exceed the total budget.

   The application designer configures the optimization equations and the constraints. For more information, see the Application Designer's Guide.

7. To experiment with different parameter combinations and see how each impacts your result, click the Create a new scenario based on current project settings icon.
8. If desired, click the Perform a test on the application for a selection of records icon.

For applications that use complex mathematical optimization (CPLEX), such as IBM Analytical Decision Management for Campaign Optimization, IBM Analytical Decision Management for Operations, and IBM Analytical Decision Management for Demand Optimization, advanced options are available to administrators. See Advanced options for more information.

Note that for applications with multiple dimension hierarchy elements, when users open the application they can choose which elements to include and reorder them. If an element has been excluded, any objective functions or constraints will also be excluded. See the topic Project configuration for more information.