*Copyright 1995, Mustafa Uzumeri and David Nembhard*

Many business processes are inherently analog in nature. The key measurements from these processes (e.g., output rates, productivity, and defect rates) can be seen as "signals" that are generated by the internal activities of each process. A product's sales history, for example, signals the marketplace's satisfaction with the product. Similarly, a worker's hourly production is a signal of that worker's ability and motivation under current working conditions.

This view differs from the implicit perspective of most current MIS systems. While they are typically very good at collecting and processing real-time transaction data, their reliance on tabular data tends to obscure the dynamic behavior of the process.

These shortcomings have encouraged system designers to look for better ways to understand and visualize dynamic process behavior. To our way of thinking, the options really fall into three categories:

- Create classifications by applying logical decision rules.
- Analyze the data in its typical tabular form with the tools of linear algebra.
- Estimate a continuous mathematical function to approximate the pattern of the underlying data points. Use this function as the basis for subsequent analysis.

By taking an analog perspective, we are led to the third option.
The analog approach uses a mathematical function to *describe*
the observed pattern. Some of the variables in the mathematical
equation are adjusted to produce a curve that captures the
essential
shape of the underlying "signal". This produces a
"filtered"
summary of the process that is much more compact and offers two
important benefits over traditional representations and analysis:

**Reduced Dimensionality:**Large databases full of process information are almost impossible to comprehend. There are simply too many variables and too much noise in the data. By fitting mathematical curves that describe the underlying patterns of data, one can greatly reduce the number of variables in the analysis. By reducing the dimensionality, complex data can be displayed in ways that are easier for managers to visualize and comprehend.

**Implicit Mathematical Analysis:**Since the signal pattern is fully described by the variables in the best-fit equation, the analyst can use the fitted equation to directly perform a number of otherwise complex calculations. For example, the derivative to the fitted function is a direct estimate the signal slope at any point. Complex calculations involving integrals are equally easy to perform and store. These types of calculations are typically much more cumbersome when done with tabular data.

We believe that these advantages, combined with the growing need to understand process dynamics, justify the extra effort required to develop and apply an analog approach. Every large organization will have some strategic measures or operations that merit fuller description and closer analysis.

We expect that the analog approach will be particularly relevant
when managers need to understand the behavior of *populations*
of individuals or critical processes. Possible examples would
include the sales rates for products, delivery times for
shipments,
market prices for stocks, defect rates from processes, cycle times
through production cells, and productivity by operators.

By characterizing the shapes of these signals across a meaningful population, the analog approach can provide decision-makers with a powerful view of dynamic operations. Moreover, our initial explorations show that this view can be achieved with current hardware and software tools. To illustrate this, we offer an example that applies this approach with a specifically chosen function.

Some Questions and Answers regarding mining analog business data

*This research is being conducted jointly with David Nembhard,
a colleague at Auburn's College of Business. *