The Art and Science of Customer Lifetime Value (CLV) Analytics

The Art and Science of Customer Lifetime Value (CLV) Analytics

Reading time: 6 mins

Meet the Experts

CLV- The core metric of your marketing campaign strategy

Formulating a winning marketing campaign is more critical now than it ever was. As organizations become more and more customer focused, developing marketing campaigns that help address the needs of demanding customers is becoming more and more challenging. While there are several aspects that go into formulating a winning marketing strategy, common sense dictates that one parameter of success is the revenue you will make from a specific marketing campaign. And no other metric helps you capture that better than the Customer Lifetime Value (CLV). In this article, we will explore the basics of this metric but specifically how you can leverage data science to fine tune this metric better.

Understanding the definition of CLV

At the core, understanding how much revenue you can make from a customer, and then “treat” them accordingly is at the core of CLV. Let us simplify it further with an example. Say you are a sales person of an AI solution company and you still need to meet your sales quota for the month. And due to some reason, you can now cold call only one prospect. You have two names in front of you- one is the COO of a mid-size company and other is a COO of a Fortune 100 company. Assuming all other factors are the same (like they both will be equally willing to entertain your sales call, will have enough budget etc.), the propensity will be to cold call the COO of the F100 company, because you expect them to spend more throughout the relationship. And this hypothetical example sums up the concept of CLV-you decide to invest your limited resources on customers who you believe will help you generate more revenue during their lifetime (of relationship with you).

And this is why CLV is one of the key metrics to monitor for marketing campaign managers. To formally define it:

The CLV metric measures a customer’s total worth to your business over the course of their lifetime relationship with your business.

And hence, it is not difficult to contemplate that this metric is very critical to keep track of, when it comes to evaluate the cost of acquiring new customers. The general principle in the world of customer retention is that it is generally more expensive to acquire new customers than to keep existing customers. Due to this, knowing the lifetime value and the costs associated with acquiring new customers is critical to strategize profitable marketing campaigns. For example, if the average CLV of your customer is $500 and it only costs $50 to acquire a new customer, then your business will be generating more revenue as you acquire new customers.

However, if it costs $550 to acquire a new customer and the average CLV of your customer is still $500, then you are losing money with each acquisition. Simply put, if your marketing spend for new customer acquisition exceeds the CLV, you will be losing money.

The challenging aspect of CLV calculation

Now the key question is- how do we calculate CLV ? There are multiple ways to calculate CLV but at the core, the components of the calculation are simple. To understand the components, let us use another example. Think of a hypothetical case, where a customer’s average purchase amount is $500 and they makes purchases four times every month on average. Then this customer’s average value per month is $2,000, which is simply multiplying the average purchase amount with the average purchase frequency

The CLV calculation example shared above was pretty simple-wasn’t it ? Ponder a bit and you will realize that a key challenge that you can come across when doing this calculation is to determine customer’s lifetime span. In our example, we did the calculation for a period of one month. But to calculate true CLV, we also need an estimate of the cutomer’s total lifetime span. And that is the challenging part of CLV- to determine, based on certain attributes, how long will the customer stay with us (lifetime). Because we do not typically know the lifetime span of customers, we often try to estimate CLV over the course of a certain period. It could be a customer’s 12-month CLV, 24-month CLV, or can also be a 3-month CLV.

Using predictive analytics to calculate CLV- incorporating the art of feature engineering

One way to estimate a customer’s lifetime span is to look at the average monthly churn rate, which is the percentage of customers leaving and terminating the relationship with your business. You can estimate a customer’s lifetime span by dividing one by the churn rate. Assuming 5% of the churn rate in our hypothetical case, the estimated customer’s lifetime span is 20 years. Given the customer’s average value per month of $500 and lifetime span of 20 years, the CLV of this customer turns out to be $120,000. This final CLV amount is calculated by multiplying $500, the average value per month, by 12 months and the lifetime span of 20 years. If you are not comfortable using historical churn data, you can use logistics regression algorithms for churn forecasting and then calculate an average monthly churn rate based off that.

Another approach is to use linear regression algorithms on historical data. And this is where the “art” aspect comes in. The art aspect here is to understand which variables and features are relevant and will help you develop a model that aligns with business realities.

In his book “The Art of Feature Engineering”, Pablo Duboue states:

“When machine learning engineers work with data sets, they may find the results aren’t as good as they need. Instead of improving the model or collecting more data, they can use the feature engineering process to help improve results by modifying the data’s features to better capture the nature of the problem.”

And this applies to every Machine Learning (ML) and Artificial Intelligence (AI) algorithm. It does not matter what problem your algorithm intends to solve. When I see articles stating failures of AI, they make me cringe since those failures are still human failures. AI algorithms fail to deliver value due to bad choices made by humans (in algorithm selection, design, features and training approaches). And the fact is, poor feature engineering is a major driver behind failures or unsatisfactory model performance. So when leveraging regression analysis for determining CLV,  feature selection and feature  engineering should be your key focus area. Remember that actually running the algorithm has become pretty easy. After I have an input data file ready, I can teach my 8 year old to import that data, import Python libraries like scikit-learn and run the algorithm, all in a Jupyter notebook environment, by copy pasting existing codes. The “soul” of an algorithm gets defined prior to funneling the data through a code.

What does this mean for SAPinsiders

Marketing has always taken a lead among all functions as far as leveraging analytics goes. With the explosion of applications of AI and ML algorithms due to advances in technology and computing power, there may be opportunities to leverage these algorithms for analytics that has been leveraged  for decades in the marketing world, like CLV, churn, personalization etc. There are certain aspects that SAPinsider needf to know to successfuly embark on their marketing analytics journey:

  • Leverage an “advanced” algorithm only when it makes a difference. As the  MIT Sloan Review article Is Deep Learning a game changer for Marketing Analytics states, the improvements in accuracy and insight will justify the investment in deep learning technology and data. https://sloanreview.mit.edu/article/is-deep-learning-a-game-changer-for-marketing-analytics/ As an example, if an advanced and more intricate algorithm is improving your forecasting percentage by 1-2%, it may not be worth the execrise to upgrade to the new algorithm. however, this does not mean that you do not need to re-evaluate algorithms after a certain timeline.
  • Remember that the “heart” of a solution lies outside the algorithm. In this case, it lies in how you select features, create new features- which require that you have the “business intuition” to understand what aspects are critical, what variable is missing in the data, how you can create a feature to compensate for that missing variable and to have “user empathy”, to evaluate whether the model results align with what the end users expect.
  • Look for simple solutions for complex problems. While this statement itself may seem contradictory, it is the new reality of analytics tools today. Since marketing analytics is more mature compared to other functions, there are off -the-shelf tools available that allow you “citizen data scientists” to leverage advanced algorithms. If one is available for your need, it makes sense to go with it rather than build a solution in-house.

More Resources