'Market segmentation' by Scandinavian Institute of Business Analytics SCANBA

Market Segmentationonline course script (without examples)

Scandinavian Institute of Business Analytics SCANBAhttp://online.scanba.org/courses/market-segmentation

Segmentation Definition

In the 1978, Paul Green and Donald Tull established the following four basic criteria for a successful market segmentation:

the segments must exist, be identifiable, be reasonably stable over time, and this segments can be reached. In general, the basic approach of segmentation is to divide a large, general market into smaller, specific segments with different needs, and different respond to marketing efforts.

Segmentation Advantages

Competitive advantage. Segmentation is one of the tools to succeed in the market despite formidable competitors. For example, car rental agencies succeed by targeting different segments of customers and adjusting rental locations for business and casual travelers.Niche marketing. With segmentation, companies can identify a niche market that desire a product or service not currently provided by existing suppliers. Customer satisfaction. Even a commodity products, like laundry detergent can have segments. For example, travelers prefer small detergent pack sizes for portability, while value -conscious people prefer larger sizes for cheaper per-ounce prices. Efficiency. With segmentation, companies can focus it core competencies on the relevant markets and utilize company resources with higher efficiency.Profitability. Segmentation acknowledges that different groups have different needs and value perceptions of the same service or product. We can grow profits by determining the maximum amounts the different groups of buyers will pay. Competitive advantage, Niche marketing, Customer satisfaction, Efficiency andProfitability are the advantages of market segmentation.

Segmentation VariablesWe denote two types of segmentation variables: response variables , or dependent variables, and identifier variables, or independent variables.Response variables describe how individuals respond to an offer. Response variables fall into several major categories: Functional. Specific variables in this category include performance, reliability, durability, quality, and so forth. Service and Convenience. For example, time savings, ease of use, ease of purchase, convenience and location of a retail store and so forth. Financial response variables focus on the monetary performance: cost savings, potential revenue gain, sensitivity to price-related promotions, liability avoidance, and so forth.

Usage response variables consider how customers will use the product or service: usage scenario or occasion, usage rate, usage frequency, application of the product or service, usage patterns, and so forth. Psychological response variables study how products and services affect psychological aspects: trust, esteem, status, and so forth. Identifier Variables categorize and describe the individuals in segments. Typical segments are: Demographics, Geographics, Psychographics for Consumer or Business Sample Consumer Demographics variables are Age, gender, family size, income, occupation, etcetera Sample Consumer Geographics variables are: Country, region, county, city, density, climate, and so on.

Sample Consumer Psychographics variables are: Personality, lifestyle, values, interests, attitudes, and other. For business markets identifier variables are different: For Demographics: Industries, company size, etc. For Geographics: Company location, proximity to customers, and so on.and for Psychographics: Specific applications, order size, and other. In this course, we treat response variables as the dependent variables, because they represent the behaviour in which we are interested, and the identifier variables as the independent variables, because they represent variables which help to explain this behavior.

Segmentation TypesMarkers use the so called, a priori segmentation, when they have existing knowledge from earlier experience on the segmentation variables.

In this approach, marketers specify the segmentation variables and the number of segments. Marketers usually design surveys and collect data in a systematic way.

In the so called, post hoc segmentation , marketers hold little knowledge about the type of segments in a particular market, or even quality.

The post hoc segmentation approach determines the segments after research is conducted and data is collected. For example, a new product or service could require post hoc segmentation techniques, because of lack of earlier experience.

Marketers use either descriptive or predictive segmentation techniques, depending on the objectives.

Descriptive segmentation techniques are applied if the objective is to describe the similarities and difference between customer groups. One common descriptive segmentation approach is cross-tabulation analysis.

Conversely, marketers apply predictive segmentation techniques if the objective is to predict how changes in independent variables affect the values of dependent variables. For example, we might wish to find how advertising spending affects sales revenue. One common predictive segmentation technique is regression analysis.

Clustering analysis techniques are fairly popular for descriptive post hoc segmentation. Clustering can be hierarchical or partitioning. Both have two aspects in common. First they all use numerically-based index to indicate the degree of similarity of two individuals. Second, all use the index to

group individuals into homogenous segments using a defined clustering process.

Conjoint analysis based segmentation is a predictive post hoc segmentation technique which reduces preferences for certain goods and services into the worth that particular attributes hold for individuals, called part-worths. We then use the part-worths to segment the market. Conjoint-based segmentation holds the strong advantage that the resulting segments represent demonstrated preferences held by different groups in the market.

In our course, we cover details of the market segmentation with cross-tabulation, regression, clustering, and conjoint-based analysis.

Cross-tabulation Segmentation

Cross-tabulation, or “cross-tab”, is a procedure that cross-tabulates two variables, expressing their relationship in a table. We demonstrate cross-tabulation segmentation using an example.

Olle wants to discover the market segments existing in their local area for his Restaurant. Olle is interested how the segments vary with respect to the number of times they dine out per month. Olle plans to focus its marketing budget on diners who like to dine out often.

Gather Market Data

Olle survey the local community during a local town fair, asking individuals how frequently they dined out per month over the past year. He also asked questions for identifying demographic variables, such as annual income, age, and occupation. The table shows a small excerpt of the resulting data. Actual surveys can have hundreds, or even thousands, of respondents.

Construct Cross-tabulation Table

In this example, on the behalf of Olle, we will construct a cross-tabulation table using IBM SPSS software. A quick look at the data reveal some market insights. Clearly the dining frequency vary with annual income. The age remained quite consistent. The occupation variable appears to be redundant because the stated annual incomes correlate with the typical incomes expected from those occupations. For large data sets we

recommend to sort the data by the variable of interest to more readily predict the underlying patterns.

We will focus on dining frequency and income.

At first, we will split the income variable into bins using Visual Binning menu.

For this, we will select the variable Income and click Continue

We will create three bins : a bin with income below thirty thousands euro, a bin with income between thirty thousands and fifty thousands, and a bin with income above fifty thousands. Do not forget to set the new name for the binned variable. We will use the name “Income Binned” with the label “Annual Income”.

Now our table contains a new column with binned income values.

To create a cross-table, we will choose Crosstabs under Descriptive Statistics in the Analyze menu.

We will set Dining Frequency for the Rows and binned Annual Income for the columns. Then we will add some more details in the “Cells” dialog window.

We want to display row values as percentages.

Upon completion, we will see the cross-tabulation table in the Output window.

It is time to interpret the results.

Interpret Cross Tabulation Table

The cross-tabulation chart shows three distinct segments. The first segment consists of relatively low income individuals who dine out rarely. Olle is not looking to target this segment due to its low revenue potential.

The second group consists of mid-income individuals dining out about once per month.

The third group appears to be Olle’s most attractive segment: individuals who dine out more than one time a month, with relatively high income.

With this knowledge, Olle will seek to increase the number of diners eating at his restaurant by executing campaign targeting hi-income individuals

Regression-based Segmentation

The goal of regression-based segmentation is to group different customers together based on the similarity they have in their relationships between the independent and dependent variables.

In regression, we seek to determine the relationship between response variable (dependent) and identifier variable (independent). Many commercially available software packages, including Microsoft Excel, include tools for linear regression.

Regression Equation

The relationship between the independent and dependent variables can be expressed in a regression equation. Here is a typical regression equation using demographic variables.

In the equation, “a” is the y-intercept and “b” is the coefficient for the variable income.

By calculating coefficients “a” and “b”, we can determine how income affects spending. We can assess if we can categorize certain groupings of people at certain income levels as segments. In this case, we estimate spending based on the single variable Income. Regression on one variable is called simple regression. We refer to regression involving more than one variable as multiple regression.

Market Data

In this example, a random group of individuals replied how much they intend to spend on their next car. The survey included age of responders and personal income. Table shows the market data sorted by the dependent variable “Spending”

To gain insight into the possible segments of the data, we review how the independent variable affects the dependent variable. In this case, our independent variable is Income. Just looking at the data we noticed that there are gaps in Spending between eight thousand and twenty one thousand, and again between twenty eight and forty one thousand.

We therefore declare three potential segments, based on the intended spending levels: Segment 1, Segment 2, and Segment 3.

More sophisticated segmentation techniques, like clustering, can identify these gaps in an automated manner. We will cover clustering methods later but now, we will conduct the regression-based segmentation.

Linear Regression

We will use Microsoft Excel to demonstrate basic regression-based segmentation.

At first, we will build a scatter plot of the Spending variable over Income variable in a range of spending below eight thousand. The first segment that we identified.

After the plot is ready, we need to right click on the markers and choose an Add Trendline dialog.

We choose Linear regression type in the Trendline options and check the boxes to display equation on chart and display r-squared value on chart.

The software will calculate the regression coefficients “a” and “b” and display the regression equation next to the regression line as well as the R-squared value.

R-squared represents the degree to which the independent values explain the estimated function (in this case the line through the data). If the data lie directly on the regression line then R-squared is equal to 1. If no relationship exists and the data points look randomly scattered then R-squared is equal to zero.

So, what R-squared value is OK for us ?

R-squared value

The R-squared value is often called the coefficient of determination.

The table shows typical values of R-squared in various disciplines. In most marketing research applications, values of R-squared fall in between the two extremes, generally near 0.6.

If the value of R-squared is lower than the typical value from the table, the case could be that the existing variables are insufficient to explain the phenomenon, and that more variables are needed. Ultimately, we seek to explain behavior using as few variables as possible.

Estimation of Spending

We conduct the regression fitting for the all three segments and record the regression coefficients in the Table. From the data, we observe that spending behavior increases with the personal income in all three segments, but at different rates. Segmenting the market into the three segments allows us to pinpoint estimated spending levels more accurately than if we had aggregated all buyers into one mass market.

Using the regression results, we can estimate the anticipated spend level for an imaginary buyer with the income of fifty three thousand euro. This buyer falls into Segment two. We already know the income coefficient and the intercept of the buyers from the segment two. Now we can estimate the spending level for the buyer using the regression equation we developed With our analysis, we expect that the buyer with income of fifty three thousand will spend between twenty four and twenty five thousand euro. The regression-based segmentation allow us predicting the spending more accurately.

Usually, market researchers use several segmentation techniques in one study. In this example, we defined the segment borders manually. In the next section, we will study clustering techniques that can identify the segments automatically.

Cluster Analysis

The great thing about clustering is that you can find the segments in your market data after the data was collected. Even if you know nothing about the variables in advance.

In cross-tables and regression, we had to assign cluster borders manually. With the clustering the segmentation can be done automatically and we will show you how.

Cluster Analysis Methods

There are two types of clustering methods: hierarchical and partitioning

We will show two most popular hierarchical clustering methods: TwoStep and Ward's.

Next we will demonstrate partitioning clustering. The most popular partitioning clustering method is K-means.

TwoStep Clustering MethodOverall, the hierarchical methods are slower than partitioning except TwoStep cluster method.

The TwoStep clustering method is a method that can create clusters in a single scan of data. The method has certain limitations, like it can be sensitive to the order of cases in the data, but it is very easy to use.

Let return to the Olle's survey data and use IBM SPSS to conduct the analysis.

At first, we need to open the Choose Technique dialog in the Direct Marketing Menu.

The we click on the icon Segment my contacts into clusters and click continue. This is a shortcut to the TwoStep clustering option.

In the new dialog window, we choose Income and Dining Frequency as segmentation variables ans run the analysis.

The Model Viewer shows the results of TwoStep clustering. As we expected from our previous studies, the algorithm found three clusters in the data set. Frequent diners with an average income of sixty six thousands a year, rare diners with the average annual income almost two times lower, and occasional diners with average income of roughly forty five thousands euros a year.

Ward's Clustering Method

Ward’s method is a popular example of agglomerative hierarchical clustering. The result of the clustering is a hierarchical structure similar to a family tree, called a dendrogram. In dendrograms, root branch off into two branches, which in turn branch off into further branches, and so on.

In Ward's we start with individual elements and merge them together into clusters. During the merging process, our goal is to lose as little information as possible. For example, if we group two individuals into a cluster, the cluster will not be as precise as each individual on their own. The information lost during clustering is sometimes referred to as the merging cost. To reduce the merging cost, Ward's method minimizes the error sum of squares.

K-Means

K-Means is a popular example of partitioning clustering. In K-Means, we specify K, the number of final clusters to expect. Next, we run the K-Means algorithm. The algorithm consists of steps, which continue to iterate until a stable solution is reached. A stable solution is reached when individuals converge into specific groups, and cease to change groups.

The first step is to determine the centroid coordinates. The centroid coordinates define the location of the center of the system, as calculated using a weighted mean, similar to finding a center of mass in physics.

Second, we calculate the distance of each individual object to the centroid and form groups based on the shortest distance from the individual object to the centroid.

Conjoint Analysis

These statements are really important for the conjoint analysis understanding.

Conjoint Analysis is a statistical market research technique to examine the trade-offs consumers make between two or more attributes.

By examining the trade-offs consumer make, we can infer the value they place on individual attributes.

Conjoint analysis is appropriate for situations where we need to quantify customer preferences for attributes. As a result, organizations often use conjoint analysis in product and service development and for market segmentation.

New Business Idea

Olle met Tove at the market fair during the survey for his restaurant.

Tove turned out to be a gourmet coffee expert. She had an idea of developing a brand new coffee machine. Olle liked the idea and they started a project “Tove Coffee”.

At first, they decided to study the market with the conjoint-based segmentation.

Conjoint-based Segmentation

The conjoint analysis process reduces preferences for certain goods and services into the worth that particular attributes hold for individual, called part worth.

Part-worth shows willingness to pay for the certain attribute by the customer. We then use the part-worths to segment the market.

Conjoint-based segmentation has an advantage of representing demonstrated preferences held by different groups in the market.

To explain the conjoint-based segmentation, we start with terminology.

TerminologyConjoint analysis uses the following terms, applicable for both products and services.

Attributes : color, size, speed

Attribute levels :large or small for size; fast or slow for speed

Bundles:Different combination of attributes

Part-worths:Values placed on particular attributes

Profiles:Specific bundles preferred by segments

As a whole, Conjoint Analysis is a technique to examine trade-offs consumer make to understand their preferences.

Conjoint Analysis Essentials On behalf of Olle and Tove, we conduct a conjoint analysis to segment the market of people wanting new coffee makers.

We start by defining attributes. In our example we declare the attributes as speed, capacity, and price. We then combine attributes in different ways to form different coffee machine bundles. Using conjoint analysis, we can learn the values people place on each attribute based on the bundle they prefer.

With an overall goal to increase sales of Tove coffee machines, we seek to understand how we can segment the market, based on the types of features. For example, we might find that casual coffee drinkers prefer machines emphasizing price, whereas coffee fans prefer high performance machines.

Conjoint Analysis ProcessWe prepare for the conjoint analysis study by defining the attributes, assigning different levels to those attributes, and forming bundles using combination of the different attributes at the various levels.

We ask consumers to state their preferences for each bundle. We prepare the data for analysis by coding it in a special form.

We then calculate the preference for customers have for each attribute , called a part-worth.

We can apply that knowledge to execute marketing campaigns.

Attributes and Levels

In order to reduce the number of variables for the conjoint analysis, we asked customers to rank several attributes: speed, capacity, price, and a cord length. All responders marked cord length as a not important variable. We will omit the cord length from the conjoint analysis.

We assigned two levels to each attribute.

Fast and Slow for Speed, Small and Large for Capacity, Budget and Premium for Price.

BundlesWe now form bundles, which represent candidate “products” for us to test. Marketing researchers often refer to the bundles as cards, because researches in the past used cards to represent individual bundles.

Our example includes three attributes of two levels each, so we will need 8 cards (2 in a power of 3). This forms a complete set of cards, which is called “full-factorial”. A typical analysis with many attributes and many levels could result in hundreds or even thousands of combinations. To reduce the number of cards, researchers apply fractional factorial techniques, for example, Taguchi orthogonal arrays. The rule of thumb is to keep manageable amount of cards. On practice, people are redundant answering more than 15 survey questions, so, we recommend using less than 15 cards.

Data Collection TechniquesThere are three commonly used techniques for data collection in conjoint analysis.

Pairwise comparison

In the pairwise comparison, respondent compare two different cards and tell us which one they prefer. Some responders find pairwise comparison easier than rank ordering. However, the number of comparisons can be too high.

Rank ordering

In rank ordering, we provide all cards at once to the responder, and ask to rank the cards in order of preference, from the first choice to last choice. The advantage of the rank ordering is

speed. In our example, the ranking will take less time than three separate pairwise comparisons. The disadvantage of the ranking is the complexity the responders face. In general, many responders quickly establish their most and least preferred choices, but find it difficult to to rank the choices in the middle.

Rating scale

The third data collection technique is the rating scale. In the rating scale technique, we ask respondents to rate each choice on an absolute scale. The rating scale can be easier for respondents than raking but some of them might find difficulty in assigning ratings for fine rating scales. To counter the disadvantage, and to

improve the consistency of results, we recommend providing guidance to responders on how they should assign rating to choices. One of the best practices is to have five choices: Outstanding. “Definitely would consider buying”Good. “The quality is good and it would be one of several units to be considered for purchase.”Neutral. “Neither good nor bad”,Fair. or “Flawed in one or more important areas”Poor , which means “Unacceptable in multiple areas”

Survey Answers

In our example, we asked respondents to rate each card on a scale from one to five stars, using the rating scale described before.

To assist in segmenting the market into accessible groups, we also asked for respondent data.

For example, for our Tove Coffee example, we would ask about demographic data like age and income, geographic data, behavioral data - intended usage for use the products, and psychographic data, like favorite interests or activities.

The tables show the result of our data collection effort for one individual.

Coding Data

To perform our analysis in a computer based tools, we need to prepare data for the analysis by coding it into digits. Some software tools, like IBM SPSS, do the transformation automatically, some does not. For example, if you intend to do the analysis in Microsoft Excel, you have to code the data manually.

To demonstrate the process, we will assign numbers to the variable levels manually.

We recommend coding two level variables as ones and zeros . If your variable has more than two levels, you will need to introduce extra variables to accommodate the information into a binary form. Most of the statistical software packages will do that for you.

Calculate Attribute Part-Worths

We now calculate the preference consumers have for each attribute, called a part-worth, using multiple regression analysis. The multiple regression process fits a function that approximates the data for multiple variables, in our case, speed, capacity, and price.

In the equation, A1, A2, and A3 are coefficients expressing the contribution for the three attributes Speed, Capacity, and Price into the coffee machine Preference in the eyes of the responder..

Apply Conjoint Results

Olle made a great job of doing conjoint analysis. Now he is going to apply the the results to complete the conjoint-based market segmentation.

Olle decides to take a close look at the cross-table with the result summary. The table sums up part-worths of female and male survey respondents for work and home use of the new Tove's coffe machine.

Olle notes relatively high average part-worth value female respondents place on speed of the coffee machine. At the same time, the Price part-worth of the coffee machine is low. With this Olle identify his first market segment: fast and premium coffee machines for women at work.

For male respondents, the average part-worths are different. The Capacity part-worth for home use is very high. The second highest part-worth is the Price part-worth for home use. With this, Olle identify his second market segment: high capacity and budget coffee machines for men at home.

Now, Olle and Tove are going to develop two coffee machines for these market segments and adjust the marketing campaign to the segment target audiences.

Pass the exam at the end of the online course and receive your certificate fromScandinavian Institute of Business Analytics SCANBA

1. Please, correctly answer at least 6 of 10 exam questions

2. Send the course title and the answers to [email protected]

Please provide your name as you want it to appear on the certificate

and your email.

To receive your SCANBA Certificate:

Anti-Crisis Analytics: Business Analytics That Helps Before, During, and After a Crisis

http://www.amazon.com/Anti-Crisis-Analytics-Business-Before-During/dp/151200930X/

Evolution forces companies to search for new means of competition. Information technology, management science, and process knowledge are no longer enough to differentiate your business and stay competitive. Business executives often suffer from poor forecasts, insufficient data, and misleading advice that directs them towards imprudent decisions. Eventually, they find their businesses unprepared for yet another crisis and they ultimately fail.

This book is about business analytics - a new competitive advantage for companies. In it, Kuandykov describes the methods that will help your business stay ahead of its rivals, foresee future crises, survive them, and move on after them. The book is written in a plain language and is aimed at a broad audience. Readers will enjoy its fresh view on business analytics methods, its real world examples, and its useful hints and ideas for their work. The book is sure to be an invaluable resource for a broad range of business disciplines.