Tom Khabaza, an early innovator in the field of data mining in the 1990’s, developed the “9 Laws of Data Mining” in his lectures in 2010. Since then an entire community of data miners has benefited from these sound principles, but they have not been widely used by GIS professionals. These principles are a suitable framework to guide location analytics and market analytics, which yield the same business value as data mining activities. As Khabaza points out, there is nothing really “new” in the laws themselves. It is in the explanation and application of these laws where we find innovation.
Those who are unfamiliar with the 9 laws would benefit from reading and listening to Khabaza explain these laws himself at http://khabaza.codimension.net/ before reading further. This series will review the 9 laws one by one, with this first post focusing on the application of the very first law.
Law #1: “Business Goals Law” – Business objectives are the origin of every data mining [market analytics] solution
Data mining does not start by looking at data; it starts with business objectives. Data mining is an iterative business process that is focused on solving problems and achieving goals. It is not simply a data “technology”; it is a process. Likewise, the GIS industry is often equated with a technology, but this is a critical misunderstanding. Anyone who has taken an introductory course in GIS will be very familiar with the five components of GIS: software, hardware, data, people and processes (methods). But even with that understanding, we sometimes miss the connection to business goals. At the start of any analysis project, we must tie our activity to a business goal because this can affect the entire structure of the analysis that follows.
The type of market analysis that I will do if my goal is related to marketing or customer retention for operations is going to be different than a project where my goal is to improve site selection for real estate. The long-term business goals and processes for different departments are inter-related for sure, but the specific short-term goals and analysis objectives will affect the analysis extent, the geographic summary levels, and the methods used to provide the business with the insight and answers they need to make better decisions.
In addition to guiding structure, the goal provides relevance to the analysis—the answer to the “So, What?” question. One way to formulate goals and objectives is to use the SMART acronym from George Doran, Arthur Miller, and James Cunningham (Management Review, Vol. 70, Issue 11, 1981). Goals should be Specific, Measurable, Action-oriented, Realistic and Time-specific. One example of a SMART business goal for a highly specialized service business would be to increase the number of customers by 10 (2%) each month as the result of a targeted marketing campaign so that you have 120 net new customers at the end of the year.
Once the goal is specified, market analytics can then be used to identify the proper message and medium for the marketing campaign based on customer data, demographics, psychographic and business data. If my goal is to increase the number of customers (both by retaining existing customers and adding new ones), I may decide to do a customer profile analysis. The customer profile will then help me create a better marketing message that will reach the customers through the most appropriate media channel.
In his book Visualizing Data (2008), Ben Fry states, “One of the most important (and least technical) skills in understanding data is asking good questions.” He points out that rather than starting with the data that we have, we need to start with the question and then work backward to the data that we need to answer the questions. The goal to increase the number of customers leads us down the path of answering a series of questions that we can turn into analysis objectives:
- Who are they? What is our target customer profile? (demographics and psychographics)
- Where do I have the highest concentrations of customers? Where do I have the lowest concentrations?
- Where is the center point of the customers? Is the pattern dispersed or clustered? Is there an orientation or direction to the cluster?
- Is there a correlation between customer density and the customer profile? (Core versus Developmental profiles?)
- Is there a relationship between density and proximity? What is the pattern of customers within specific trade areas? How many are within a specific drive time? Does a “donut” drive time study show any patterns that are related to proximity bands?
- Do the customers or stores that are within a specific distance from a shopping center (or a highway interchange) have different buying patterns than those who do not?
- Which customers spend more?
- Is there a differentiation between which products or services are purchased by different types of customers? Does the customer profile vary based on the product or service being purchased? If I map a subset of the customer data, do I see a different pattern or trend?
- Which customers come to the store more frequently?
- Is there a day of the week or time of day when I have more or less customers and how will that pattern affect operations? Is the profile different by time of day?
- How does the customer profile differ among various cities?
- How does the customer pattern change when I change the map scale or the level of aggregation?
- How does the pattern change when I group the customers into 6 classes versus 3 (low, medium, high)? What happens when I map just two classes—those above the mean and those below it?
- What has changed since the last customer profile and how much has it changed?
How do I know which goals and objectives are appropriate for my business? That question leads to the second law of data mining, which I will cover in my next post. It is only through business knowledge that we can learn to craft appropriate goals and those goals will allow us to formulate specific objectives (answers to questions) related to each variable that we are studying.