Proactive Supply Chain Performance Management with Predictive Analytics

Predictive Supply Chain Performance Management Model

Performance management of complex business networks such as supply chains requires a unified approach that comprises different management models, technologies, and tools. This section introduces an integrated supply chain PM model which incorporates the supply chain modelling method and business intelligence technologies such as data warehousing and data mining. It is based on the integrated supply chain intelligence model for collaborative planning, analysis, and monitoring.

The main elements and the structure of the supply chain PM model are shown in Figure 2.

Figure 2 Predictive supply chain performance management model.

The basis of the model is the supply chain modelling method which enables modelling of supply chain processes, relationships, metrics, best practices, and other relevant elements. Output of this stage is a supply chain process model that serves as input for data warehouse design.

First, based on the process model, data from various sources is extracted, cleaned, and transformed in order to accommodate requirements for KPI design, multidimensional analysis, and data mining models. The following step is construction of OLAP cubes with proper dimensions and measures. OLAP schema is the basis for design of supply chain KPIs which measure the progress toward predefined goals. KPI typically provides a visual representation of metrics over time, rather than just displaying the numbers.

The next step which includes data mining is the key step toward predictive performance management. Here, historical performance (KPI) data is used for making predictions about future performance. This information is then delivered to decision making via a special BI web portal in the form of web reports, charts, scorecards, dashboards, or notifications. Alternatively, prediction information can be used in supply chain simulation models for analyzing different scenarios and risks. Abukhousa et al. used simulation models as an analysis tool for predicting the effects of changes to existing healthcare supply chains and as a design tool to predict the performance of new systems under varying sets of input parameters or conditions.

The final step is taking appropriate actions to resolve problems and make adjustments to strategy and plans. These actions can be made based on more detailed reporting, data exploration, specific expert systems, simulation, and data mining models. One such intelligent software solution for integrated and interactive supply network design, evaluation, and improvement is developed. It consists of three modules designed for knowledge-based supply network modelling, rule-based simulation executions, and intelligent assessment and improvement.

In the next subsections, the main elements of the proposed supply chain predictive PM model will be described.

Supply Chain Process Model

The starting point is the SCOR process model which provides a library of the supply chain specific set of processes, relationships, metrics, and best practices. The SCOR process model contains a standard name for each process element, a notation for the process element, a standard definition for the process element, performance attributes that are associated with the process element, metrics that are associated with the performance attributes, and best practices that are associated with the process.

All process metrics are an aspect of the performance attribute. Performance attributes for any given process are characterized as either customer-facing (reliability, responsiveness, and flexibility) or internal-facing (cost and assets) metrics.

Top level metrics are the calculations by which an implementing company can measure how successful they are in achieving their desired positioning within the competitive market space. Lower level calculations (levels 2 and 3 metrics) are generally associated with a narrower subset of processes. For example, delivery performance is calculated as the total number of products delivered on time and in full, based on a committing date. Additionally, even lower level metrics (diagnostics) are used to diagnose variations in performance against plan. For example, a company may wish to examine the correlation between the request date and committing date.

Each process from the process model has its related metrics, best practices, and inputs and outputs. All metrics follow the same template which consists of the following elements:

(i) name,

(ii) definition,

(iii) hierarchical metric structure,

(iv) qualitative relationship description,

(v) quantitative relationship (optional, if calculable),

(vi) calculation,

(vii) data collection.

Based on the SCOR process model, we have created the SCM metamodel (Figure 3), which enables the creation of any supply chain configuration and is the basis for further modeling Metamodel is normalized and contains all SCM elements such as processes, metrics, best practices, inputs, and outputs. It also incorporates business logic through relationships, cardinality, and constrains.

Figure 3 SCM metamodel.

Metamodel is extended with additional entities to support supply network modelling. That way, processes, metrics, and best practices can be related to a specific node and tier in the supply network. With this metamodel, processes at different levels can be modelled, thus providing a more detailed view of supply chain processes and metrics. Metamodel database contains standard SCOR metrics but also enables defining of custom metrics, as well as metrics at lower levels (i.e., Level 4, workflows, or Level 5, transactions).

The developed SCM metamodel enables flexible modelling and creation of different supply chain configurations (models). These models are the basis for the construction of data warehouse (DW) metadata (measures, dimensions, hierarchies, and KPIs).

DW and OLAP KPI Modeling

A user who wants to retrieve information directly from a data source, such as an ERP database, faces several significant challenges.

(i) The contents of such data sources are frequently very hard to understand, being designed with systems and developers instead of users in mind.

(ii) Information of interest to the user is typically distributed among multiple heterogeneous data sources.

(iii) Whereas many data sources are oriented toward holding large quantities of transaction level detail, frequently the queries that support business decision making involve summary and aggregated information.

(iv) Business rules are generally not encapsulated in the data sources. Users are left to make their own interpretation of the data.

In order to overcome these problems, we have constructed the unified dimensional model (UDM). The role of a UDM is to provide a bridge between the user and the data sources. A UDM is constructed over one or more physical data sources. The user issues queries against the UDM using a variety of client tools.

Construction of the UDM as an additional layer over the data sources offers clearer data model, isolation from the heterogeneous data platforms and formats, and an improved performance for aggregated queries. UDM also allows business rules to be embedded in the model. Another advantage of this approach is that UDM does not require data warehouse or data mart. It is possible to construct UDM directly on top of OLTP (on-line transactional processing) systems and to combine OLTP and DW systems within a single UDM.

In the UDM it is possible to define cubes, measures, dimensions, hierarchies, and other OLAP elements, from the DW schemas or directly from the relational database. This enables providing the BI information to the business users even without previously built DW, which can be very useful having in mind the fact that within the supply chain there can be many nonintegrated data sources which require time to connect, integrate, and design the data warehouse.

Flexibility of UDM is also manifested in the fact that tables and fields can be given names and descriptions that are understandable to the end-user and hide unnecessary system fields. This metadata is further used throughout the UDM, so all the measures, dimensions, and hierarchies that are created based on these table fields will use these new names.

Definitions of all UDM elements are stored as XML (eXtensible markup language) files. Each data source, view, dimension, or cube definition is stored in a separate XML file. For dimensions, these files contain data about tables and fields which store dimension members. OLAP cube definition files also contain information on how the preprocessed aggregates will be managed. This metadata approach enables centralized management of the dimensional model for the entire supply chain and provides an option for model integration and metadata exchange.

Measures are one of the basic UDM elements. Measures are the primary information that business users require in order to make good decisions. Some of the measures that can be used for the global supply chain analysis and monitoring are as follows:

(i) reliability:

(ii) responsiveness:

(iii) agility:

(iv) cost:

(v) asset management efficiency:

Each of these measures is calculated by using the lower-level metrics from all supply chain participants. For example, the perfect order fulfillment measure is based on the performance of each level 2 component of the order line to be calculated (% of orders delivered in full, delivery performance to customer committing date, documentation accuracy, and perfect condition).

During the design, for each measure we need to define the following properties:

(i) name of the measure,

(ii) what OLTP field or fields should be used to supply the data,

(iii) data type (money, integer, or decimal),

(iv) formula used to calculate the measure (if there is one).

The next step is to cluster measures into measure groups. The measure groups are an integral part of the UDM and the cube. Each measure group in a cube corresponds to a table in the data source view. This table is the measure group's source for its measure data.

Supply chain process model (built using the SCM metamodel) can be used as the basis for defining measures and measure groups because it provides relationships between business processes and metrics, metrics hierarchies, definitions, quantitative and qualitative descriptions, and description of possible data sources that provide data for calculations.

Companies often define key performance indicators, which are important metrics used to measure the health of the business. An OLAP KPI is a server-side calculation meant to define company's most important metrics. These metrics, such as net profit, assets utilization, or inventory turnover, are frequently used in dashboards or other reporting tools for distribution at all levels throughout the supply chain.

The UDM allows such KPIs to be defined, enabling a much more understandable grouping and presentation of data. Key performance indicator is a collection of calculations that are associated with a measure group in a cube that is used to evaluate business success. Typically, these calculations are a combination of multidimensional expressions (MDX) or calculated members. KPIs also have additional metadata that provides information about how client applications should display the results of the KPI's calculations. The use of OLAP-based KPIs allows client tools to present related measures in a way that is much more readily understood by the user.

Table 1 lists common KPI elements and their definitions.

OLAP KPI structure.

Term	Definition
Goal	An MDX numeric expression that returns the target value of the KPI.
Value	An MDX numeric expression that returns the actual value of the KPI.
Status	An MDX expression that represents the state of the KPI at a specified point in time. The status MDX expression should return a normalized value between −1 and 1.
Trend	An MDX expression that evaluates the value of the KPI over time. The trend can be any time-based criterion that makes sense in a specific business context.
Status indicator	A visual element that provides a quick indication of the status for a KPI. The display of the element is determined by the value of the status MDX expression.
Trend indicator	A visual element that provides a quick indication of the trend for a KPI. The display of the element is determined by the value of the trend MDX expression.
Display folder	The folder in which the KPI will appear to the user when browsing the cube.
Parent KPI	A reference to an existing KPI that uses the value of the child KPI as part of the KPI's computation.
Current time member	An MDX expression that returns the member that identifies the temporal context of the KPI.
Weight	An MDX numeric expression that assigns a relative importance to a KPI. If the KPI is assigned to a parent KPI, the weight is used to proportionally adjust the results of the KPI value when calculating the value of the parent KPI.

Figure 4 shows a part of the return on supply chain fixed assets KPI defined in the OLAP server.

Figure 4 DX KPI definition.

Besides aforementioned benefits of the UDM model, it also serves as a basis for data mining model design because it provides a consolidated, integrated, aggregated, and preprocessed data store.