Business Friction and the Next Generation of Decision Management

Kelly Pennock, CEO

Decision Management (DM) can be defined in generic fashion as the application of math and logic to data to produce smarter decisions. It is sometimes called Enterprise Decision Management, though I've never seen an enterprise make a decision—it is instead a host of department heads that incrementally implement Decision Management, improving their processes and addressing business problems one at a time.

The value proposition for Decision Management sounds compelling and practical, even simple: If the accumulated "experience" of the company can be reliably harnessed through statistical means, then every relevant decision can be made more profitable. For some time, studies have borne this out. As far back as 2003, an IDC study showed a 145% median return on investment for predictive analytics used for DM, significantly higher than returns from other analytics investments such as business intelligence.*

Decision Management, however, has not been as widely adopted as its potential value would imply, despite the compelling story and high profile successes, including companies such as Amazon.com or Capital One. The basic reason is that Decision Management is difficult. Companies that build it from the ground up, that invest massively in people and tools, and that have incorporated it in their DNA throughout the enterprise have certainly been successful, but for the other 99% of us, adoption will be incremental and must be preceded by greater simplicity.

In this article, I address the three primary sources of business friction that have slowed the adoption of Decision Management technology—"analysis friction," "operational friction" and "command and control" friction. Each of these types of friction will be described in the paragraphs below, along with the solutions that in aggregate constitute what I believe to be the next generation of DM technology.

"Analysis Friction" a.k.a. analysis paralysis

The first source of friction that significantly slows the adoption of Decision Management is primarily an issue of resource availability and productivity. The gating technology for DM is predictive analytics—algorithms that transform yesterday's behavioral patterns into tomorrow's behavioral forecast—capable of predicting consumer activity such as fraud, delinquency or attrition. Today, the keepers of this gating technology are the statisticians. These quantitative professionals are scarce and in high demand. Compounding the resource scarcity is the state of current DM tools that do not promote great productivity. In a nutshell, relatively rare specialists using relatively low productivity tools has limited the application of DM solutions to only the highest profile problems, and at that, only for those companies that are "lucky" enough to find and employ quantitative talent.

This sort of squeeze—scarce and specialized talent on the one hand and limited productivity on the other—is exactly the type of problem that technical progress addresses, from 3rd generation programming languages to desktop publishing. Next generation analytics technology, in a manner analogous to other technical advances, can automate a good portion of the expertise and eliminate the most time consuming steps.

First generation predictive algorithms, such as linear or logistic regression, are still the dominant technology used today. While easy to understand, they are relatively inaccurate and rely on resource intensive manual intervention. Second generation analytics were adopted by some in the 1980's and 90's. Neural networks are an example of a second generation algorithm. While neural networks were a marketing success and were often more accurate, this technology did nothing to minimize reliance on scarce professionals, or to increase the productivity of those professionals; indeed, they relied on even scarcer resources and frequently took longer.

The third generation of analytics technology, collectively known as ensemble modeling, is empirically more accurate than neural networks and, even more importantly, significantly increases productivity by eliminating time consuming data analysis and feature pruning. These new techniques make it practical to add data such as temporal or textual data that previously had to be ignored due to the complexity. In short, today's state-of-the-art techniques can utilize more and different data, produce more accurate predictions, and substantially increase the productivity of scarce professionals.

Whether a business purchases analytics by the slice, through micro applications that solve specific business problems (e.g. debit transaction fraud or loyalty offers), or buys the full course meal with DM suites that enable them to roll their own, the next generation of DM technology should decrease the time to build and refresh the analytics that drive smart decisions. This will increase the number of decisions that can be addressed, increase the quality of those decisions and lower the total cost of ownership.

"Operational Friction"

The second primary source of friction that slows adoption of Decision Management technology is operational friction—the heat generated when extensive new software demands are placed on a perpetually stressed IT department. And DM can place a significant burden on IT because it is not a single, simple technology. The technologies that are part of effective DM include a data mining component, a segmentation or clustering component, a predictive modeling component, a reporting component, a business rules component, and a runtime environment. All must be linked and to some degree interoperable. For the current generation of DM, this is a heavy set of technologies, sold separately by separate vendors, requiring different sets of expertise to operate, and non-trivial IT support to link and maintain. Moreover, to deploy individual decisions for batch or real-time decisioning, the output from the analytics component must often be hand coded in production, along with the business rules that execute the strategies. And of course, it is not just a one-time deployment. New analytics must be built to capture the nuances of an ever-changing customer base, and new strategies crafted to continuously improve the efficacy of every customer interaction. Then IT must recode new analytics or rules. This is a maintenance challenge to say the least.

The next-generation DM solution will provide a single software platform that contains all the required components and can be offered in a hosted environment. (SaaS- software as a service) A single platform eliminates the need for IT to link components together, keep data resident for each of the applications, and increases the opportunity to collaborate between those responsible for the analytics and those responsible for the strategies or rules. A single platform actually reduces the need, given the automation discussed in the previous section, for multiple experts by providing a common interface across the elements. Offered on-demand, the platform even further reduces the requirements on IT, making it much easier to incrementally, problem by problem, apply DM technology to important decisions across the customer lifecycle.

The platform benefits accrue to runtime as well. Micro applications that solve very specific business problems, those produced either internally or purchased from a vendor, can be "published" from the design time components to the runtime component of the platform. This removes the error prone and resource intensive process of hand coding either analytics or rules-based strategies and therefore simplifies and speeds the best-practice of frequent revisions. For micro applications that are purchased, it becomes possible to buy an application that contains both pre-built analytics and "customizable" strategies—rules that can be adjusted as desired by the individual business without involving the vendor and triggering additional costs.

All in all, a single decision platform offered in a SaaS model represents a major step towards next generation DM technology, by minimizing the traditional IT friction and enhancing collaboration between various constituents.

"Command and Control Friction"

The final type of friction dampening the adoption rate for DM is often the most lethal. I label it "command and control friction." The business owner, as designated commander, must direct the forces (DM) to accomplish the mission (optimized customer interactions leading to happier, less risky, more profitable customers). The difficulty, and where the friction surfaces, is that most business owners don't understand analytics and almost never understand the likely costs and benefits. Commitment to running the business "by the numbers" requires a leap of faith. It should not be surprising that given the obscure, even incomprehensible nature of this quantitative "magic", business owners do not jump in with two feet, committing their business and careers to Decision Management technology.

The next generation of DM technology must provide visibility into the likely costs and benefits of strategies to be deployed, reliably simulating results before commitments are made to expensive deployments and the risk of execution. This capability, in effect, becomes an in-line "business case generator" that minimizes the guess work associated with DM.

A second and related benefit is that what-if analysis becomes possible. Business owners and operational strategists can test a variety of options: such as what's the likely difference in campaign costs and acceptance rates associated with four different loyalty offers, or the delta in dollars collected and call center costs between working debt internally and outsourcing it, etc. When these questions can be asked and answered by the business owner, they are able to "command" resources with full knowledge of the associated costs and benefits.

By reducing the primary business frictions that slow the adoption of DM technology—"analysis friction," "operational friction" and "command and control friction"—the next generation of Decision Management technology will be more readily adopted and the business benefit of smarter, data-driven decisions can be extended to the other 99% of the world, one decision at a time.

* Source: IDC, Predictive Analytics and ROI: Lessons from IDC's Financial Impact Study, 2003, September 2003, document number #30080