What is customer lifetime value (CLV)?
Customer lifetime value is a metric that is used to quantify the value that a customer brings to the company. While CLV is commonly used and mentioned everywhere, there is no fixed or pre-defined or one formula that analysts could use. It depends a lot of the business domain and its application. From my experience, often a combination of metrics are used to describe a customer’s value.
There are many articles that computes CLV in the literal sense of LIFETIME value and these articles use churn rate to determine customer lifespan, which I rarely find it helpful and have not used them in practice. This is because the computation of 1/churn rate = # years/months assumes linear churn rate, which is a strong and often inaccurate assumption.
What to consider before deciding on CLV?
- Measure to use
- Is the company concerned with the ultimate revenue amount? Some other measures that can be considered are spendings/expenses. While revenue is mostly tied with spendings/expenses, for companies with complex product offerings and pricing structure and a different strategy, the key measure may not necessarily be revenue.
- You may also include value drivers, which are factors that you know drives revenue/profitability. This accounts for all dimensions of value across a customer’s lifetime. For instance, in a B2C company which relies heavily on word-of-mouth, a customer who is an advocate and refers a lot of his friends, can be considered as a high value customer.
- We also need to consider what is the statistic to be used — is it average or median? In the examples below, I use average revenue per user (ARPU) as it is the most common CLV metric. For businesses where revenue is heavily skewed / long tail, median revenue per user can be considered.
- ARPU = Total revenue in a month/year / Total paying customers in the same period
2. Use case / Level of granularity
- Understanding the use case of CLV helps to determine the level of granularity required for CLV.
- [Broad metric] Reporting purposes
- If stakeholders would like just a single metric to measure the revenue generated by a single user on the platform monthly/yearly, the average revenue per user (ARPU) is a common metric used.
- CLV is also measured against Customer Acquisition Cost (CAC) to determine the payback duration (i.e. time taken for the CAC to be earned back). - [Cohort based] Inform business decisions about past customer behaviour/trends
- Most marketing campaigns are run on a monthly/quarterly basis. The effectiveness and type of marketing campaign can determine the quality of acquired customers. As such, it is help helpful to look at cohort based revenue generated. This helps to track the performance of customers acquired from certain campaigns in specific months. - [Individual] Customer segmentation for promotions
- Often, CLV is computed at an individual level so that the marketing team could design segmented and differentiated promotions to customers of different value. For instance, a customer CLV is likely to spend more and can be incentivised with a higher promotion amount.
3. Forward or Backward looking
- [Backward looking] In the above use cases, I’ve illustrated them with the use of ARPU. Even using the ARPU metric, we decide what is the optimal time period to measure a customer’s ARPU. Is it life-to-date, monthly rolling x-day or rolling x-transactions basis? Why? Would this metric be easy for the end user to interpret and understand the logic of computation?
- [Forward looking] Sometimes machine learning can be used to predict a customer’s future value, though the use of machine learning in CLV model may result in added complexity to interpretation as the end user (e.g. marketing team) cannot easily reason out why a customer has a particular value. With a forward looking metric, we can better embed leading indicators of high future LTV into the model and enable us to make better judgement on the value of customer today. Some examples of leading indicators include: Product usage behaviour, demographics, source of acquisition.
4. Ease of computation
- As mentioned above, besides computing CLV based on historical data, another way to measure CLV is to predict a customer’s future value. The use of machine learning or some form of modelling may require more significant computation and engineering resources.
Approaches to Evaluate CLV
Historical approach
Historical approach utilises past transactional/revenue data from customers to report and inform business decisions. This approach is evaluated at the broad level and can be applied at cohort and individual level.
- For broad based metric, it is common in business reporting to report ARPU within the month.
- For cohort based, it is commonly viewed at a monthly level. Sometimes, it can also be seen weekly for business with shorter lifecycle.
- At an individual level, this is trickier as we need to decide what is the optimal time period to measure a customer’s ARPU for the particular use case. Is it life-to-date, monthly, rolling x-day or rolling x-transactions basis? Why?
- What is a customer’s natural tendency to purchase? Is it once a week, once a month, once a year? Plot a histogram of number of purchases made within the past 30 days among users who are more than X days old. If it peaks at 1, we know that the natural frequency is once a month. Any consideration of interval shorter than a month would not be valid.
- Is there seasonality in the dataset? Plot number of customers who purchased by month, week, daily. Observe for seasonal trends. If there exists seasonality in data, the time period to be considered should include the seasonal time period. For instance, in e-commerce where purchases are heavily influenced by Black Friday sales and festive season in December, it is perhaps more ideal to compute a yearly ARPU value to include data points by customers who make one-time big purchases during such period
- If there is any reason to be more biased towards recency of purchase, we could use a time-decay model to allocate a higher weightage to recent purchases.
Predictive approach
The predictive approach takes into account transactional data to predict their future value. This approach works well when there is a sizeable amount of data and there is a use case for predictive future value e.g. marketing team would like to create some campaigns to incentivise existing customers based on their future value.
RFM metrics are commonly used in CLV modelling.
- Recency represents the time taken between now and the customer’s latest purchase.
- Frequency represents how frequent the customer makes a purchase on the platform.
- Monetary Value represents the average value of a given customer’s purchases. This is equal to the sum of all a customer’s purchases divided by the total number of purchases.
There are 2 ways to model CLV:
- Machine learning model
Models can be trained with RFM and other customer demographics features. - Probabilistic model
While traditional machine learning approach works well and can be interpretable, I read an interesting approach to CLV model using probabilistic model with Poisson distribution → link here. In general, the steps taken are summarised below:
- It uses BG/NBD model to predict the expected number of purchases in the next 6 months.
- Then it uses a Gamma-Gamma model to predict the most likely value for each transaction.
- Both models will predict the 6 months CLV. The customers are then segmented into groups for marketing to target differently.
Application of CLV
The CLV metric used should be defined based on your use case. Here are some of the use cases of individual CLV that can be helpful.
- CAC to CLV ratio
- May have room to increase acquisition cost to gain more users
- LTV value is a benchmark for business decision on acquisition e.g. how much can marketing team spend to acquire a customer? It also helps to trigger discussions around healthy payback duration for the company.
2. Segmentation of CLV
- Offer products to specific segment of certain traits. Differentiated promotion can be given to those with higher CLV.
- Promotion campaigns to increase CLV for lower segment
- Focus on more valuable segments. We can also look into the demographics / behaviour of customers with higher CLV against those of lower CLV, similar to below chart done by McKinsey → this can inform marketing team to strategise their content and branding to target the higher valued clients
I hope this helps! This is a pretty general and broad topic. Feel free to leave me any feedback or any angles that I have not considered!
