The predictive enterprise is an IT infrastructure that supports business operations by delivering informed predictions about what is likely to happen in a business process. Implementing a predictive enterprise presents both cultural and technological challenges. However, many of the technologies required to build such an infrastructure are now mature, including enterprise data management (EDM) and predictive analytics. Also, new middleware methodologies such as service-oriented architecture (SOA) can be used to interface with the various technology components. Moving to a predictive enterprise and delivering results to business decision-makers across an organization, from the call center to the boardroom, is now achievable.

The central objective of implementing a predictive enterprise is to use the company’s knowledge about its past business activities, stored in data warehouses, data marts, log files, and other sources, and the current state of business operations to make predictions about future events. These predictions can be used to support the decision making of all individuals in an organization. Applications include the intelligent promotion of offers to specific users on your Web site, providing differential services to your best customers and detecting fraudulent transactions.

Many organizations have developed a business intelligence (BI) capability where raw data is cleansed, transformed and stored in a data warehouse using a format suitable for delivering commonly requested management reports. The traditional implementation maintains separate technology stacks, one for the operational (online transaction processing or OLTP) style of applications and another for BI and reporting applications. This method uses extract, transform and load (ETL) tools to move data in a single direction, from the operational system to the data warehouse. In a predictive enterprise, the goal is to close the loop in the flow of information, so that predictions that are derived from the historical data can be fed back into the OLTP applications.

Building a predictive enterprise solution that closes the loop between the OLTP and the BI domains can prove challenging, as there are many elements required to make it successful. The following steps illustrate the basic roadmap to the predictive enterprise.

Selection of Business Process

 

A predictive enterprise starts with understanding the business processes that are at the core of most OLTP systems. To deliver the most value, a predictive enterprise solution needs to be used at an operational level by making predictions about individual transactions, irrespective of whether this is a customer interaction, sales transaction or supplier transaction. The first step in developing a predictive enterprise is to identify business processes that can benefit from adding predictive elements. The best candidates are those processes that have the potential to improve the bottom line by achieving more with less. The process should also have a predictive element where patterns or behavior can be predicted such as those related to an individual customer’s needs, desires and behaviors or items related to specific financial transactions.

Once selected, the business process that will be implemented as part of the predictive enterprise must be redesigned to deliver variants of business processes, taking into account predictions and decisions inherent in the process. Without variations of the process, it is impossible to provide differentiated routes based on the nature of the decision. For example, an interaction with a customer who is predicted as likely to defect should be processed with a different variation of the call center process than a customer who is likely to remain loyal.

Definition of Predictive Services

 

The second phase of developing a predictive enterprise is to develop predictive services. The construction of a predictive service starts with defining the different decision points in the business process and determining what information can be predicted that can support those decisions. These predictions might include, for example, the likelihood of a customer defection to a competitor, the likelihood of a transaction being fraudulent, which offer a customer is likely to respond to, the best time to discount prices of perishable stock and the likely demand for a product.

Every predictive service is based on an archive of historical data. The primary source of data is likely to be a data warehouse, which might be supplemented with data from specific data marts, log files and purchased data. It is important to note that the data collected for management reporting might not be suitable for the development of predictive services, and so the data being archived as part of the ETL process may need to be altered.

Once the required data has been collected, advanced analytics methods such as statistics, data mining and predictive modeling can be applied to the data to develop the required predictions. To use the results of the analysis, different predictive systems must be deployed as predictive services. If these services are provided by the BI team, a culture change is often necessary. The operational characteristics of predictive services are more closely aligned with an OLTP system than a BI system. However, predictive services are also critically dependent on historical data, which is the cornerstone of a BI infrastructure. A predictive enterprise requires the BI platform to support these new requirements as well as its traditional requirements.

The key differences between predictive services and traditional BI capabilities are summarized in Figure 2.

In a traditional BI environment, the consumers of information use historical data to make strategic decisions. In most cases, the volume of queries is low since reports are typically generated daily. In a predictive enterprise, requests to the predictive services take place on a transactional level and may be used by many individuals at all levels in an organization.

Automation of Decision-Making

 

A critical element of a predictive enterprise is timing. The value of a prediction is dependent on when the prediction is made. For example, acting on the prediction that a transaction is fraudulen

t has no effect unless the intervention can take place before the transaction is completed. The action latency, or the time it takes to act once an event has happened, makes demands on the style of architecture used to build the predictive enterprise. In general, the shorter the action latency, the higher the value that the decision has to the business.

A predictive enterprise will fail if the action latency of the organization is slower than the required response time. Once the deadline for making the decision has passed, there is no value and sometimes even a penalty for not acting in time.

Decisioning services can also be implemented to provide automatic decision-making capabilities by applying a predefined set of business rules to reach an outcome. Business rules-based systems enable organizations to specify decision-making policies by groups of experts and then embed these rules into business processes – rather than relying on human operators.

There are a number of advantages to using business rules:

  • Decision-making policies can be centrally designed, implemented and changed;
  • Mistakes are reduced; and
  • The time required to make decisions is shortened.

Business rule-based decisions follow the same processes of deduction as a human decision-maker, and thus retain an element of predictive systems. However, every decision is made using the same logic.


Embedding Predictive Services into Business Processes

 

The final challenge of implementing a predictive enterprise is adding predictive and automated decision-making capabilities into the OLTP applications that run the daily business. The business processes that orchestrate the operational systems must be able to support different variations of the business process in order to respond to the decisions made using the predictions. This may require the process to be redesigned to support the requirements of the predictive enterprise. The process must then be integrated with the predictive services. One popular method of accomplishing this is to loosely couple the predictive services into the OLTP applications using an SOA. If the performance requirements are high, technologies such as enterprise service buses can be used to implement high throughput messaging services.

A predictive enterprise solution is complete once the loop has been closed so that predictions, based on historical data, can be used at an operational level to support or automate decision-making. In suitable cases, predictions can be combined with business rules to automate some of the operational decision-making, thus speeding up response time, reducing errors and standardizing organizational policy.

Once implemented, the performance of the predictive services, automated decisions and processes can be measured by analyzing transaction logs to determine how successful the process was. The business process intelligence derived from this analysis can then be used to optimize different elements of the predictive enterprise and enable all components to react to changes in the external business environment.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access