Today, Business Intelligence (BI) is struggling to fulfill its early promise because it lacks fundamental components that are needed to support predictive analytics. BI has become too much about data analysis technology and not enough about modeling. Furthermore, BI application developers have built platform-oriented tools and have abandoned the end users who really needed domain specific solutions. For businesses that want to look backward, BI tools are great. They can help a business understand what happened in the past, but this backward-looking practice is insufficient to support true decision-making. Decision-makers need a system that provides a holistic representation of a business and models system constraints.
All decisions are really about the future. Analysis about the past is helpful, but what organizations need is the ability to model the future for true planning and business decision support. BI vendors understand this, as do the analysts covering the industry. BI technology is being modified to support predictive modeling (or predictive analytics). In essence, predictive modeling takes a past/current scenario and applies statistics to try and predict how a business would perform if it continued in the same circumstance. More advanced tools even allow a business to run rudimentary what-if analyses to examine potential remedies; however, the nature of this technology with its one size fits all approach limits predictive modeling.
Predictive modeling requires four essential ingredients:
- An accurate representation of the business situation that is holistic enough to provide the right context and understanding
- Domain specific knowledge
- Appropriate and clean data
- Support for managerial experimentation (e.g., what-ifs, simulation, optimization, root-cause analysis, etc.)
BI tools are adequate for supplying updated, clean data, but modeling is not their strength. Representing business constraints and integrating domain expertise into a BI-oriented tool is very difficult. Most importantly, BI tools will not support experimentation; their structure and the applications built on them are not suited for this.
OLAP (OnLine Analytical Processing) alone is not the answer. While OLAP systems are sufficient for analyzing significant amounts of incoming operational data to find patterns in past operations; these systems are not good at providing planning and decision support. OLAP tools have become generalized platforms for building and analyzing multi-dimensional data sets. They allow a user to build solutions, but do not represent data in a way that supports predictive modeling and decision support, nor do they have the built-in capabilities needed to check proposed solutions.
OLAP should be complemented with constraint-based modeling, which represents problems as a series of simultaneous equations that need to balance and close. Equations can be manipulated and solved using an LP solver. The solution represents the best possible outcome with the given constraints. Changing the constraints allows the user to predict and compare any number of possible futures. OLAP cannot manage this process alone because its focus is on data analysis and not on building a mathematical representation. The combination of BI and constraint modeling is ideal for predictive modeling. BI tools support the collection and transformation of data while domain and solution-specific knowledge is represented in the constraint-based modeling system. With a well-designed, constraint-based modeling system, many scenarios can quickly be represented, giving managers insights not possible before. Constraint modeling supports inquiries concerning particular events – it’s easier to understand why something is happening. Root cause analysis allows managers to truly understand the implications of their decisions.
Constraint-based modeling does two things that other technologies do not. First, it pulls all of the data together into a single representation. This is helpful to a business because for the first time it can look at the origin of data, how that data gets used, and how different data silos store identical data differently. All data is pulled together and forced to work together – there is no other way to provide a holistic check and balance on a company’s information. Secondly, constraint modeling reveals, often unexpectedly, how an organization really works. Besides pulling a whole data picture together, constraint-based modeling builds a mathematical representation of the data that supports analyses, in a manner that OLAP and data models cannot. For example, if a business can only make so many units of product X, what will its profit be for the year? If a business has a target profit, what are the production requirements for product X? These are two very different models that solve differently. Constraint-based modeling is the only technology that can properly represent and solve problems in a variety of ways.