Speaker 1: Hi welcome to the channel Data Science Demonstrated. In this channel you will learn about data science using examples and visualization. There is no theoretical power points neither there is any programming screens. So whatever your profession you can just sit back relax and enjoy the videos. Now in this specific video I will be showing you on how to use data science to do some cutting edge customer analytics. The success of your business depends upon how well you understand your customer. So in this video you will see on how to use data science to gain more customer, to sell more products to your customers and to retain your customers. So let's get started but before that do not forget to subscribe to the channel. Customer life cycle generally starts by acquiring the customer then you need to understand the customer based on various data, communicate with the customer and sell additional products as well as retain the customer. Data science can help you in each of these areas and here are some of the customer analytics which can be used during the customer life cycle. I will be going through each one of these using examples and visualization. So let's get started with the channel attribution which helps you acquire your customers efficiently. Nowadays you can acquire customers through various channels such as emails, social media, telephone calls or page search. Channel attribution helps you understand which of these channels are effective. So here is the data on marketing channels which were used while acquiring the customer. So you have the customer ID, you have the marketing channel on which the marketing message was shown to the customer, you have the date and time and you also have the information whether the customer bought something or not. So if he just saw the marketing message it is shown as an impression but if he bought something it is shown as a conversion. So if you take all the customer lines which are marked in yellow, you will see that various channels were used for marketing such as the online video, page search, Instagram, email and Facebook between 7th of July and 14th of July. Now the customer bought a product after seeing the advertisement on Facebook and the value of purchase was 50 US dollars. Now even if the purchase was made using Facebook it is possible that all other marketing messages on different channels also contributed to the customer's purchase decision. So data science algorithms can be used to attribute the purchase value equally to all the preceding channels. So here we have a value of 50 and we are equally splitting it across different channels. Now in order to find out which channels are effective you can use a transition matrix which is shown on the screen. So on the x-axis you have all the marketing channels and on the y-axis you have the similar marketing channels but in addition you also have one line for conversion. So the color of each cell represents the number of customers. So higher the number of customers more red is the cell. So as you can see that most of the conversions was due to Facebook, due to online video and due to page search. You can also use the transition matrix to find out how the customers go from one channel to another and as you can see that most of the transitions are between Facebook and Instagram. So even if Instagram does not have a high conversion value you can still use it and follow it up with a Facebook advertisement which would then lead to a conversion. Let us move to the next customer analytics called the customer behavior analysis. A customer can perform various activities such as visit your shop or send emails, call the call center or visit your website. Now all these activities they generate data and data science can help you understand the customer behavior. So for example you can understand if a customer is a frequent visitor or is the customer satisfied or angry or is the customer digital savvy or is the customer price conscious. Let me show you with an example on how you can use the website data to understand the customer behavior. So for example here is a simplistic view of data generated while customers are browsing a website. So you have the customer ID, you have the date and you have the activity. You can visualize and analyze this data using path analysis which is shown over here. The visualization shows the start of the customer activity and the end of the customer activity and all the paths which are taken between the start and the end. So in this example you see that one of the main paths which are taken is customers they browse the product, then they add the product to the cart, after that they check out and after that they either proceed to the payment or they remove the product from the cart. So customers who are adding the product to the cart and then removing it are possibly doing a price simulation in order to know the total price including the shipping cost. So this indicates a price sensitive customer behavior and one would possibly target these customers with some discount offers. On the other hand you have some customers who are returning the products and they are checking the return policy and after that they are calling the call center. Now this indicates that either the return policy is not very clear and the customers need to talk to a call center to know more about the return policy or the customers are not digitally savvy and they prefer getting in touch with the call center. So path analysis can help convert data into some very powerful customer behavior insights. Now let us move to the third customer analytics called the customer segmentation. Now once you have acquired the customer and you have understood the behavior you can now start communicating with the customer. Now let us say you want to send mails for any promotional offers. Now one option is to send a generic mail to all your customers. Now as we have seen in the previous section that different customers have different behavior. One common message to everyone is not a very efficient thing to do. The second option is to send a custom mail to each and every customer. However this may not be also very efficient and it is also not very productive as it requires a lot of effort to create custom mail for each and every customer. Now the third option is to make groups of customers who are similar based on some similar customer behavior and then you can send a mail specifically for each group. So this process of making groups of similar customers is called customer segmentation. Now let us see this in action using an example from the banking industry. So here is a data on banking customers where you have information such as the customer ID, the age, the job, the marital status, education, whether the customer is a defaulter or not, the bank balance and other information. Now one way of making groups or segments is to use heat map visualization. You can select any two columns based on your intuition for making this heat map. So here we have selected job as the x-axis and the marital status as the y-axis and each cell in the heat map is a segment and the color of the cell indicates the number of customers in that cell or segment. So more number of customers means more red is the color of the segment. So you can see that most of the customers they fall into the segment of the job as the blue color job and marital status as married. So now you can tailor your marketing message very specific to these segments. So the heat map approach is a good approach but it is based on intuition because we had to manually select two columns for the heat map. Now the other approach is to use data science algorithms to automatically determine the segments and the result of such algorithm is shown here in terms of a scatter plot. On the x and the y-axis you see the columns which the data science algorithms have automatically determined to be optimal for making the segments. So we see that the age of the customer and the bank balance have been chosen by these algorithms as the most optimal fields to do the segmentation. Now each of this point over here they indicate a customer and the color of the point indicates the segment or the group. So we have three customer segments we have the blue segment, we have the red segment and we have the green segment. So the blue segment over here it corresponds to people who have less age and they have relatively low bank balance and the red segment over here indicates people who have a higher age or which are older and they have a low bank balance. And then you have the green segment over here which is customers who have relatively high amount of bank balance. Now with these segments one can now create very targeted marketing messages and communicate very effectively with the customers. Let us move to the fourth customer analytics called the product recommendation which will help you sell more products to your customers. Now let me give you an example based on Hollywood films. So here is the data on what films customers have liked watching. So you have the customer ID, you have the movie name and you have the rating which was given by the customer. Using data science algorithms you can convert this data into a network of films that you see on the screen. This is also called as association analysis. Now each circle or the node here represents a film and the link or the line between the two nodes indicates that the two films have been liked by the customer. So for example the customers who liked Beauty and Beast they also liked Lion King and they also liked Aladdin. We have some other examples over here where customers who have liked Pulp Fiction they have also liked Memento and they have also liked Fight Club. So you can use this network of films to make a recommendation to the customers. So for example if a customer has liked Goodfellas you can recommend also Godfather to the customer. So in this way you give the customer what he wants and thus you are increasing your sales. 5th customer analytics called the churn prediction which will help you retain your customers. Let me give you an example based on the telecom industry. So here you have some data on telecom customers, you have some demographic information and you have the information about the services which the customer has and you have some billing information and this last column over here titled as churn indicates if the customer has churned or not. So churn means the customer has left the telecom operator and all the records which are highlighted over here in light blue this indicates that the customer has left the telecom operator and all these records over here indicate that the customer is still there but we do not know that if this customer will churn or no in the future. So we would like to use data science algorithms to predict whether this customers will churn or no. The magic which makes this prediction possible is called machine learning. There are many ways to do machine learning and one of the most efficient way is called decision tree which you see on the screen. On the right hand side of this decision tree you see the decisions which have been taken whether the customer will churn or no indicated as churn equal to no and churn equal to yes and then you have the path towards this decision. For example if a customer does not have a month to month contract neither the customer has got fiber optic services and if the tenure is less than 30.5 months this customer is most likely to churn and leave the company. You also have the number of customers that fall under this decision path. So in this case there are 832 customers which fall under this decision path and most likely predicted to churn and leave the telecom operator. So if you want to retain these customers you can start contacting them, start offering some discounted services or discounted prices and prevent them from leaving. So machine learning is a very powerful way to predict any future outcomes or events so that current actions can be taken based on it. So hope you have liked the video. Do not forget to subscribe to the channel and give a thumbs up to the video.