This document describes some of the key terms used in LityxIQ Optimize.
Scenario – a combination of data, business metrics, and constraints based on which an optimal business decision is desired. Changing any of the inputs (data, metrics, constraints) can lead to a different result, allowing for what-if comparison of results across many scenarios.
Optimization Dataset - a dataset (created in Data Manager) that provides all the necessary data to support the determining of the optimal decision by Optimize. It often includes data such as model scores, forecasts, costs, segmentation data or demographic groups, and so on. Typically, there is a row in the dataset for each unique value of the Optimization Dimension Variable.
Optimization Dimension Variable - the result of an optimization problem in LityxIQ is to provide a recommended decision for each unique level (each value) of the selected optimization dimension. The Optimization Dimension variable is the variable that holds these unique values. Some examples:
- A single dimension ProspectID with 2,000,000 unique values (representing unique prospects). The decision to be made for each of the 2,000,000 is whether or not they should be targeted in order to optimize future value of a campaign.
- A dimension named ChannelMonth which is a string representing all possible combinations of 6 marketing channels and 12 months within a year. The decision to be made is how much of a fixed total budget to allocate to each Channel/Month combination in order to maximize sales for the upcoming year.
Attribute - a variable in the Optimization Dataset that is categorical (string or integer) that serves as a qualitative descriptor in the dataset. For example, in the optimizing prospects example, "Education Level" might be an Attribute in the dataset that provides information on each prospect. Attributes are used in the optimization problem to create constraints (e.g., all targeted prospects must have at least a college degree) or to report results (e.g., report response and sales by education level).
Data Element - a variable in the Optimization Dataset that is quantitative (integer or decimal) that serves as a quantitative descriptor in the dataset. For example, in the prospects example, "Predicted Response Rate" and "Prior Times Contacted" are examples of variables that could serve as Data Elements. Data Elements are used to define core optimization Metrics (see below) on which decisions will be based.
Metrics - a summary of a Data Element across the dataset or a subset of the dataset. Metrics are used in three ways in an optimization problem:
- As a basis for determining the optimal solution. In this case, they are also referred to as Objective Functions. For example, a metric in the prospects problem might be "Overall Cost Per Order", and the objective of the problem might be to find a targeting plan that minimizes this value in the face of other constraints.
- As a basis for constraints. For example, a metric in the prospects problem might be "Total Budget", and a constraint for the solution to the problem might be ensuring that the Total Budget of the recommended targeting plan be under $1,000,000.
- For reporting and analysis. All metrics defined in a problem will be computed and available for analysis. They can be compared and contrasted across scenarios, and within and across subsets of the dataset.
Constraints - business rules that restrict the possible solutions to the optimization problem. As referred to above, a common example of a constraint is a budget constraint in which the optimal solution must not go over a given budget amount.