CLV (Customer Lifetime Value) is one of the most important marketing metrics. It measures how much business a brand can expect from its customers. This allows the brand to make decisions such as how much to spend on customer acquisition, or how to provide differentiated products or services to high CLV customers, or how to create retention strategies for the high CLV segment, or how to increase the lifetime value of customers.
How is CLV calculated?
The simple calculation uses how much a customer spends on each transaction, how many transactions does a customer make in a year and what is the lifespan of that customer in years.
CLV = Average order value (amount spent per transaction) X
Purchase Frequency (Number of times a customer buys in a year) X
Average Customer Lifespan (How many years does a customer continue the relationship)
For example, if a customer spends $50 on an average on each purchase, and shops every other month i.e. 6 times a year and stays with a brand for 3 years. Then, the CLV is $50 X 6 X 3, which is $900.
The more complex calculation sums the profit from each customer (revenue in each year minus direct cost associated with that customer) discounted over time to calculate present value, for the period of the business relationship.
Historical CLV looks at past data to make judgements. Predictive CLV leverages past data to predict future value of CLV.
What is the difference between CLV and LTV?
While the terms CLV (Customer Lifetime Value) and LTV (Lifetime Value) are often used interchangeably, CLV is a metric of the value of an individual customer whereas LTV is more of an aggregate of all customers. CLV focuses on individual customers and segments and allows brands to make differentiated decisions, whereas LTV is an indicator of overall business profitability and growth.
How can AI models help in predicting CLV?
There are several AI models that can predict the CLV of a customer based on historical data of customer’s transactions and engagement with the brand.
A linear regression model can predict the value of the dependent variable CLV based on the value of independent variables such as average transaction, frequency of purchase, time since last purchase, number of returns, demographic variables such as age, gender, location. A logistic regression model on the other hand can give a binary outcome such as decide whether a customer is high value or low value, based on the model training data.
A brand can also use survival analysis to predict how long a customer will stay with the brand, by predicting the time of churn. This can go as an input to CLV prediction.
A decision tree uses a flowchart like structure and uses historical data to predict outcomes. Each node in the tree represents a test on an attribute, each brand represents outcome, and each leaf represents a label or the final decision. For example, a fashion e-commerce company may have average spending purchase more than $50 as the root node and purchase frequency more than 10 in a year as branches, finally leading to high CLV of $1200 if the frequency is more than 10 or low CLV of $500 if the frequency is less than 10. Further, another node for average spending less than $50 could be whether the customer is signed up for the loyalty program. The response yes or no will determine whether the CLV falls under medium or low.
Similarly, Gradient Boosting Machines (GBM) can be built using several decision trees to incrementally improve the prediction of CLV, where each new tree works on reducing residual error from the previous tree. GBM takes parameters like number of trees, depth (number of decisions each tree can make) and learning rate (how much each tree tries to correct errors of previous tree)
CLV estimation can also be framed as a clustering problem where the customers are clustered into low, medium or high CLV customers, using a model such as K-means clustering. This allows the brand to identify distinct groups and have separate marketing strategies for each.
Neural networks can be a powerful method to predict CLV identifying complex patterns and data relationships. The input layer will take in parameters that could impact the CLV. This could be demographics, purchase frequency, average order value, product categories, engagement scores etc. The hidden layers will detect complex patterns, such as the relationship between product categories, spending habits, and engagement level, and then compute an outcome and pass it on to the next layer. The output layer will give CLV as a single value. Over time, the model will adjust the weights of each of the features to give more accurate prediction.
Support Vector Machines (SVM) are another popular method to classify customers based on CLV. An SVM works by mapping data points in space and finding a surface or hyperplane that best fits the data points, based on historical data. Imagine plotting a customer Sara’s data point in a multi-dimensional space where each dimension is a feature (age, income, purchase frequency, preferred categories) The precise position of Sara’s data point relative to the surface will determine her predicted CLV.
Can packaged customer data platforms predict CLV?
The two commonly used platforms Salesforce and Adobe both have capabilities of CLV prediction through their in-built AI platforms, Einstein and Sensei respectively. Einstein aggregates data from various sources such as sales, service and marketing cloud and then using prediction builder and defining CLV as a target variable, it can forecast future value. Adobe Sensei can help identify which customer attributes correlate strongly with higher CLV and then create precise and dynamic customer segments through Audience manager.
What is the future of CLV prediction?
The future of CLV prediction will be characterized by more sophisticated models, more comprehensive data integration including unstructured data like social media reviews, real-time data processing, increasing the accuracy of the predictions. This will play a broader role in personalization at scale, helping with micro-segmentation which is creating highly specific small segments, and driving predictive personalization, by predicting what a customer needs at any given point of time and recommending just that.
Several companies are using CLV predictions today. Amazon and Netflix use CLV models to tailor recommendations. Starbucks uses CLV for its rewards program. However, there is still untapped potential in CLV prediction. With research suggesting that a 5% increase in retention can boost profitability by 75%, it's a worthwhile pursuit.
(Article originally published at: https://www.linkedin.com/pulse/unlocking-future-value-how-ai-predicts-customers-lifetime-neha-verma-gzzhc/ )
Posts
No comments:
Post a Comment