Streamlining Decision Engine Production

	Bookmark This Page Email This Page Format for Printing Submit an Article Read More Articles
	Related Tools & Articles What Is Business Process Management and How Does It Work? BPM: The Cross-Industry Solution 9 Steps for Implementing Business Process Management
	Discussion Forum "BPM is a market and a methodology used to address a broad range of process issues - from workflow to integration, process modeling to business intelligence. Process modeling can exist apart from workflow...and vice versa." Contribute to this Discussion

By Aldo Mancini

If a company is not striving for operational excellence through business process management (BPM), it could be. BPM helps companies cut costs, respond faster to market changes and improve customer service. Much of the credit for these advances in business processes goes to BPM companies and their solutions. These companies have succeeded in integrating traditional applications, like customer relationship management (CRM) and enterprise resource planning (ERP) within existing enterprise infrastructures, facilitating business process management. Thanks to these companies, businesses are more agile than ever.

But BPM companies still have work to do. One area that needs improvement is their decision capabilities. As BPM applications have matured, BPM solutions have failed to address decision-making. As a result, companies can still take as long as four months to deploy a decision model into a production decision engine. If businesses are to survive, they must shorten their time to market. Fortunately, companies do not have to wait for new versions of today's BPM solutions to do that.

Barriers to Production

Companies today have all the necessary ingredients to facilitate decision engine production. Excellent software solutions exist that can process complex decision algorithms, and can store and retrieve data efficiently. Many companies feature large analytical groups composed of statisticians and econometricians that are adept at building advanced algorithms to create models and strategies. The software solutions these individuals use to score data are highly sophisticated and feature-rich. So what is the problem?

Some reasons it takes companies so long to produce immediate decision engines include:

1. Different vendors produce the development and the deployment software. These solutions do not communicate with each other.
2. The data used to create the model differs from that used to execute the model.
3. Companies lack the automated processes that can quickly move the model into production.
4. No versioning strategy of the model exists.
5. The data is not easily accessible to business analysts.
6. The overall process relies heavily on information technology (IT).

Unfortunately, there is no "silver bullet" that can eliminate these barriers. Nor are most BPM solutions robust enough to overcome them. In particular, BPM solutions today cannot bridge the gap between the development software and the production software, at least not enough to dramatically speed the deployment process.

Key Requirements Needed to Streamline BPM

Companies can do something to eliminate these barriers. They can customize existing business processes to facilitate production. But first they need to know what business analysts need, including the following factors:

• Ease of use: Comprehensive workbench (where all data and all models reside) that is designed for them and easy to use
• Agility: Ability to create/modify/deploy the model quickly
• Precision: Simulation capabilities to make more profitable and targeted decisions
• Consistency: Tools to eliminate the need to reinvent the wheel every time they create a new model
• Access to data: Access to clean data and the ability to get to it quickly. The better the data analysts have to work with, the faster they can produce their models.

Customizing Business Processes at Discover

Discover Financial Services, one of the world's largest credit card issuers and payments company, has been able to speed production through good change management. For example, the data used to develop a model was not the same as the data used to deploy it, because analytics is a "development" environment and deployment a "production" environment. That in turn meant the extract/transform/load (ETL) function happened independently for both environments.

At Discover, business practitioners created one ETL function with the same conditions for both. The data used in the development environment – which usually resided in different places but is now centralized – is the same as that used in the production environment. Thus, having the same quality of data in both environments reduced deployment time.

This approach has worked well in numerous situations. In one, a new customer contact strategy was created that involved what the best offer would be to customers the next time the company interacted with them – via phone, email or other channels. In the past, customers were scored once a month and the company determined what they would offer that customer. By improving the business process, the company was able to quickly develop a solution that would present the best offer to the customer, at the time of the interaction. This effort dramatically improved Discover's targeting efforts.

Four Steps to Customization

Below are four basic steps that a company can take to customize its business process to drive the production models:

1. IT and the business must agree on, and document, a common set of variables with definitions that will be used in creating models. This effort will take time and patience, since there could be as many as 10,000 variables.
2. IT must centralize this new set of common data and make it available for development and production, regardless of the tools used. The new data must be tool agnostic.
3. IT and business must define a streamlined process that will allow a model or a strategy to be converted from a development language to a production language. This effort can be further streamlined by quickly agreeing to use only one language.
4. IT must implement a versioning process when the same model or strategy is being recalibrated. There are companies that provide excellent versioning solutions.

If a company were to achieve only steps one and two they would speed the production process considerably. This would remove most of the unnecessary steps that IT must take to implement the decision model or strategy.

Figure 1: Current BPM for Decisioning

Current BPM for decisioning: Multiple data conversions create inconsistencies between analytic systems and production decision engines, which cause delays in model implementation and result in less accurate decisions.

Figure 2: Future BPM for Decisioning

Future BPM for decisioning:

• Comprehensively leverage the single source of truth for both analytics and production, which will eliminate data inconsistencies, enable faster model implementation and lead to more accurate decisions
• Decision ODS (operational data system) needs to be decision engine agnostic

Conclusion

Business process management is changing the speed at which business is conducted. Thanks to recent advancements in BPM, companies can process millions of statistical scoring decisions quickly and efficiently. But even as BPM solutions mature, they have failed to address one of business' most important areas – decision-making. Despite efficient BPM solutions and other technological achievements, companies still take far too long to deploy a model into a production decision engine.

BPM companies are making great strides in decisioning and are continuously improving their solutions. The situation, however, does not preclude companies from taking matters into their own hands. Taking the four steps discussed in this article will help a business customize a solution to expedite its decision-making efforts. Companies striving to become real-time enterprises will take these steps. Those that are reluctant to do so should remember that their competitors have probably taken them already.

About the Author: