Customer input to improve your operation
1. Customer input to improve your operation
Often when you are running an operation, you need to continually improve it. Improvements can come from many places but often your own customers are a great place to start.2. Examine customer survey data
There are multiple ways to learn from customers. Qualitative interviews, online reviews, and transactional data to name a few. There are many ways to use this data to improve an operation. One of the easiest to implement is using a model to explain customer survey data.3. Survey inputs as a model
Imagine we asked the following questions through a customer survey to gather product input. We can then capture each customer's responses in a table, or modeling matrix, that can be used to build a model on.4. Survey inputs as a model
Supervised learning models have a dependent or target variable, which is the variable about which you want to gain a deeper understanding. The other variables are considered independent or explanatory variables and explain variations in the target variable. In the matrix, question one, how satisfied are you overall, is the target variable. Scores of 4 or 5 are considered to be success so are marked as 1 while scores 1,2 or 3 are failed customer interactions and coded as a 0. The responses to question 2, 3, and 4 are the explanatory variables.5. Explanatory models from customer data
Once the data is organized like this, a simple model, like logistic regression, can help explain the relationship to success in question 1 for each of the subsequent questions. The beta coefficients explain how much each of the explanatory variables explain the overall satisfaction. Examining these betas can really help identify customer-based improvements using this data.6. Sum of betas to understand impact
Once the model is built, we'll know the coefficient values. The coefficients are not that insightful by themselves. Instead, using a technique called sum of betas can provide detail for each of the corresponding question's impact. To perform a sum of betas on this model take the first coefficient, 0.25. Then add up all coefficients, 0.25 plus 0.25 plus 0.1 which equals 0.6. 0.6 represents the total amount of variation that is explained among the independent variables, which is the information captured in each survey question. Next, divide 0.25 by 0.6 which equals 0.42 then repeat for each question. For question two, the proportion of variation is 0.42. Thus, 42% of a positive customer interaction can be attributed to question 2's subject and so on.7. Adding context with frequency
In order to really understand the impact of a question, you also need to know how often your product performs well for that question.8. Adding context with frequency
For example, question 3 - How do you rate the product/service options - may score a 4 or 5 out of 5, 35% of the time. Similarly, the question - Do you agree that the offering is fairly priced - may score high among 15% of responses. Adding the overall frequency for each question adds operational context.9. Putting it altogether
Visualizing this data as a scatterplot helps identify where to focus. The x-axis represents how often good scores are earned. The y-axis is the sum of betas score. The red lines are averages for the axis making the chart easier to interpret. Question 2 pertains to quality, scoring well 80% of the time with a 42% impact on good customer interactions. This is the upper right point. The upper right quadrant represents operation aspects the customer cares about and that the business often does well. Question 4, price, in the lower left is an area for improvement since the frequency score is low. However, customers don't see price as important since the sum of beta value is low on the y-axis. The lower left section represents aspects the business isn't good at but customers don't mind. In contrast, product options, in the upper left, has a high sum of beta, but low frequency score. So customers care about having more options but the business doesn't score well. It's an opportunity for focused improvement.10. Off to it!
Don't worry if you don't understand the math behind logistic regression. The upcoming exercises just present the visualization, and model outputs. That's all you need to make a data-driven decision. See if you can spot the areas of opportunity to improve.Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.