MIS771 Descriptive Analytics and Visualisation Management

on customer satisfaction and the possibility of a customer creating an alliance. Summaries of the collected data will be discussed and the models developed. The building process of the models and probabilities of having a strategic alliance with Auspaper will be discussed in this paper.

Descriptive Statistics

The information collected from the customers included their view on product and services that were rated on a scale of 0 to 10; with 0 as the lowest and 10 the highest. Some of this information include the length a customer has been purchasing from Auspaper, the type of industry (Newspaper or Magazine) and the number of employees employed by the firm. Also, the customer location, either within Australia and New Zealand or otherwise was also a point of concern that could affect the satisfaction levels. The mode of product distribution used and the perceived qualities were also extracted from the sampled customers. This information was among other including the view of the customers based on whether they would create an alliance with the company were included.

In a scale of 0 to 10, the average view and opinion on the quality of the product was 7.894, indicating that Auspaper was maximising its capabilities to produce quality products for their customers. Several customers reported a satisfaction rate of 9.9 out of 10 for the product quality. However, the product quality data was skewed to the right, indicating that most of the customers were very satisfied with the quality of papers they received. Auspaper did not perform very well in resolving complaints and the average score for their views was 5.36. The data for the complaint resolution was approximately normally distributed because the median is equal to the mean. In addition, depth and breadth of Auspaper product line, the image of Salesforce, a level of internal support and extend of developing & selling new products are the other factors that were averagely rated. Satisfaction levels and pricing were also rated above average with both having an average of 6.9 scores out of 10. Also, customers provided an average score of 6.04 on the basis which Auspaper supports claims and warranty.

There are some factors that were underrated because the customers were not satisfied. These variables include; perception of whether billing and ordering were handled efficiently, their delivery speed, an existence of e-commerce activities, price flexibility and advertising campaigns. Therefore, these are some of the main areas that Auspaper could focus on to improve the customer services, which would probably improve the possibility of the customers having a strategic alliance with Auspaper. On an average, the customers stated that they were satisfied with the products and services of Auspaper. 57% percentage of the sampled customers stated that they were not willing to have an alliance with Auspaper and 43% percent agreed. The firms purchasing paper from Auspaper were equally distributed between the newspaper and magazine industries. More customers did not originate from Australia and New Zealand, indicating that the demand was more outside than within these two countries. Also, the indirect method of supplying the products was more by 4% compared to the direct method.  The type of customers; either less than a year, between 1 and five and above five years, were almost equally distributed in the three groups.

Identify the significant variables for the model

There are assumptions which should be met when conducting a multiple regression. Firstly, the response variable should be drawn from a normal distribution. This assumption is tested by plotting a histogram, which shows the distribution of the data. After plotting a histogram, the distribution turned out to be normally distributed, which shows that the data was drawn from a normally distributed population. The second assumption is that the continuous predictors should have a linear relationship with the response variable. A correlation matrix was developed and it was observed that only 6 variables had an above moderate relationship with the response. Their scatter plots were developed and distribution of the data for the variables with moderate and high correlation with satisfaction score indicated the linear relationship.  

These variables include product quality, complaint response, product line, image, billing and delivery speed. These variables would then be included in the model focused on estimating the satisfaction levels for the customers. The frequencies for the categorical variables were obtained and it was found that the industry type did not show a significance difference between the industries because they were equal. The others including the location of customers, customer type, size of firm and distribution system shown a difference between the groups. Moreover, all the categorical variables will be included in the first model and those turn out to be insignificant will be removed for purposes of improvement.

Constructing the model to predict customer satisfaction

First model

The first model for estimating and predicting customer satisfaction levels for Auspaper products and services included the following set of predictors.

• Product quality
• Complaint resolution levels
• Depth and breadth of Auspaper’s product line
• Salesforce image
• Perception of ordering and billing
• The delivery speed
• Customer type
• Industry type
• Size of the firm
• Location of customer
• Distribution system

The model was fitted using the sample data, which produced an R-squared value of 84.96%. Therefore, the set of the predictors stated above explained 84.96% of the variation in customer satisfaction levels. Regardless of the high power of the model, there are still some variables that turned out to be insignificant at 95% confidence interval. These factors include industry type, the location of a customer, complaint resolution concern, billing and delivery speed. As a data analyst intern, I decided that the insignificant variables should be removed from the model to improve its reliability level.

This model was presented as shown below: -

Customer type was sliced into two dummy variables that would cater for the 3-level categories and type 1 was used as the base (reference) group. The other categorical variables were included into the model the same way and they can be interpreted according to the presented model above.

Interpretation

A customer who has been purchasing Auspaper product between 1 and five years had a higher satisfaction level by 0.665 than the others who were less than a year old. However, those who were more than 5 years had been satisfied more by 0.726 compared to those below a year. In addition, they were also more satisfied by 0.061 compared to those who have existed for 1 to 5 years. Customers producing magazines were satisfied more by 0.093 compared to the Newspapers industry. Also, the larger firm seems to have more satisfaction on Auspaper's products and services as compared to the small firms. Moreover, based on the model above, customers within Australia and New Zealand were more satisfied than the others by 0.127. Direct distribution had more effect on ensuring that the customers would continue doing business with Auspaper because they were more satisfied by 0.403.

An increase in one score about the quality of products by a customer would improve the satisfaction level by 0.273. Although complaint response seems not to affect business much, customers who reported a higher rate of complaint response were more satisfied. The depth and breadth of product line perceived improve customer need to be improved the customer satisfaction by 0.122 per score as it increased. An increment in Salesforce image opinion by one score improves the customer satisfaction by 0.346. In addition, billing and ordering improvement by a single score would improve the satisfaction level by 0.056 points. Finally, delivery speed has a positive correlation with satisfaction level and an improvement in a score would make a customer be satisfied more by 0.102.

Second model

The second model for the prediction of customer satisfaction level was based on removing the insignificant predictors from the model. After they were removed, the model has refitted again and it was observed that all the set of the remaining predictors were very significant because their p-values were less than 0.0001. The coefficients changed because of the removed predictors hence reshaping the model on different weights. The model changed as shown below and included the following variables.

• Product line
• Salesforce image
• Product quality
• Distribution system
• Type of customer
• Size of firm

Based on the second model stated above, increasing the product line by one score improves the customer level of satisfaction by 0.237. The view of the customer on the Salesforce image was directly related to the customer satisfaction and an improvement in the opinion by one score increases the rate by 0.353. The quality of the product is a very important factor to be considered and this model suggests that as customers perception on the quality improves by one point, the satisfaction level increases by 0.237. Customers who received via direction distribution system were more satisfied by 0.397 compared to those who received via a broker. A customer who had been buying from Auspaper for the time between 1 to 5 years was more satisfied by 0.907 compared to those who existed for less than a year. However, a customer who has been served by Auspaper for more than 5 years was more satisfied than those who have purchased for a period of 1 and 5 years by 0.3. In addition, comparing customer who has existed for less than a year and the other more than a year, the latter is more satisfied by 0.937 about Auspaper’s products and services. Larger firms were more comfortable working with Auspaper than the smaller firms.

Interaction effect for product line and location of a customer

Three models were plotted to check the effect and significance of the two predictor variables and their interaction. The first model generated an R-square of 41.78% and it included only product line predictor, which was a lot significant in the model. The second model included product line and customer location in the model as the predictors and the R-square improved to 45.23%. Both the product line and customer location were significant at 95% confidence level. Lastly, I fitted a model including the both predictors and their interaction. The interaction and the predictors were significant and 95% confidence level and R-square improve to 50.3%. Since the interaction effect was significant in the model, the location of the customer variable can be referred as an interacting variable in the model. The final fitted model for the interaction was displayed as shown below.

The increase in product line score would improve the customer satisfaction by 0.31 points based on the interaction model. However, customers who do not originate from Australia and New Zealand were less satisfied by 2.927. The interaction effect between product line and region helped in improving the customer satisfaction by 0.555.

Probability of having alliance for customers with neutral views on Salesforce image and product line

A logistic model was plotted to predict the probability of having a strategic alliance based on product line and Salesforce image. The model was less sensitive because its specificity was around 76.31% compared to a sensitivity level of 70.9%. Both the model predictors were significant at 95% confidence interval. The obtained model is shown below: -

The probability of having an alliance for neutral (5) product line and Salesforce image was obtained as 0.2258.

Probability for having strategic alliance on different levels of customers' perceptions of product quality and price flexibility

In the same manner as the above logistic model, product quality and price flexibility were rather used as predictors. The model below was obtained and used to calculate the probabilities at different levels.

Therefore, the probabilities of having a strategic alliance were calculated and tabled as shown below.

As shown in the table above, a customer who had rated price flexibility as zero and product quality as 1 had not chance of having a strategic alliance of having a strategic alliance. Therefore, as their levels of rating increased, the probabilities of having the alliance increased significantly; with those rating maximum for both having a probability very close to 1.

Time series for predicting turnover

A simple time series was fitted and a linear prediction model was used to estimate three periods, which include second third and fourth quarter in 2017. The obtained values include 5,701.2, 5,754.1 and 5,807 respectively measured in thousands of dollars. The equation used in predicting these values is as shown below.