Data Based Decision Making Mining Your Business Data for More Informed Decision Making
What Is Data Mining? Data mining is sorting through data to identify patterns and establish relationships. Automatic discovery of patterns Prediction of likely outcomes Creation of actionable information Focus on large data sets and databases for valuable insight Machine learning – software runs to apply proven algorithms Classification as well as predictive modeling Business decisions based on model results Larger data sets yield statistically confident results
What sort of business issues can it help? Marketing Customer retention Basket analysis – e.g. what’s the next best offer Sales forecasting Improve return on marketing spend – targeting – identify cross/upsell Find prospects where there’s the best chance of a quick win Reduce acquisition costs Merchandise planning Claims analysis Credit scoring Fraud detection
Example: Telecom Churn Analysis The Business Pain: Most telecom companies suffer from voluntary churn. Churn rate has a strong impact on the life time value of the customer because it affects the length of service and the future revenue of the company. For example if a company has 25% churn rate then the average customer lifetime is 4 years; similarly a company with a churn rate of 50%, has an average customer lifetime of 2 years. It is estimated that 75 percent of subscribers signing up with a mobile carrier every year are coming from another mobile provider, which means they are churners. Telecom companies spend hundreds of pounds to acquire a new customer and when that customer leaves, the company not only loses the future revenue from that customer, but also the resources spent to acquire that customer. Churn erodes profitability.
What We Need To Know To Fight Churn Who is churning? Why are they churning? What are the alarm bells we can look for? What strategies can we employ to limit churning?
How Can Data Mining Help? Historic customer data – demographics, transactions, interactions Every customer has a footprint within this data Each footprint will identify a behavior Mining will identify patterns in behavior Patterns are tested and validated for accuracy Valid patterns produce models that can be applied to predict behavior.
Churn – the solution The Dataset Data mining requires large datasets to ensure statistically valid results Data may come from many sources as its important to include all aspects of the business that may influence churn. CRM, accounting, customer service, surveys
Churn – the solution Preparation Flag records that we know already churned Identify variables that may be predictive – for example we would include Number of service call but exclude phone number Build decision tree – the decision tree will identify the variables or combination of variable that influence churn within the sample dataset.
Churn – the decision tree
Understanding the results The first split is on Day Minutes so lets look at this variable
Understanding the results We can examine the Day Minutes variable to understand what the typical usage is. In this case the median is around 175 min
Understanding the results Darker colour indicates higher correlation to churn
Understanding the results – the key nodes Higher than median Day time usage without Intl plan Higher than median Day time & evening usage without VM Low day usage with high number of customer service call High day usage with no VM
Retention strategies High usage valuable customers - provide Intnl plan to this group High usage valuable customers - Provide VM service to this group Lower value customers with high service cost. Maybe too high maintenance to be profitable High usage high value customer so provide VM service
Deploy retention strategies Each node of the decision tree that we wish to action provides the criteria that we can apply to identify new and existing records that fall within that node Day Mins >= 210.480 and < 245.560 and Eve Mins >= 254.590 and VMail Plan not = 1 These criteria are used to create rules within the organisations CRM to identify records that fall within each actionable node, thus allowing retention strategies to be deployed on an on-going basis, as customers move in and out of actionable groups.
Summary Retention strategies based on actual historic data Deployable to live systems actionable in real time Models easily re-evaluated as new products and services added Efficient allocation of marketing/retention budget Identify systemic issues causing churn e.g. poor customer service