Predictive Marketing 1.4: Life Time Value

“Customer Life Time Value” Also Known As “CLV”

In this blog series in previous posts, we have reviewed the importance of Hyper-targeting, Channel attribution, Personas, and contexts. Today, let’s review how CLV plays a key role and what are the considerations in deriving a predictive CLV for subscription-based and non-subscription-based, e-Commerce/Retail businesses.

What is CLV?

Perhaps, this is the most talked-about item, in the area of Marketing. Every Customer spends a certain amount of dollars to buy your products during their active engagement period with you. The predictive value of each Customer’s spend can be termed as the Customer Lifetime Value (CLV).

Is this a new concept?

No. CLV calculations can take the most basic form by utilizing all the average values at the organization level and deriving some individual Customer’s average CPA, average order value, average retention value, average product costs, average profitability, etc.,

Why is CLV not accurate in most cases?

Most often than not, your CLV is not predictive and not calculated based on individual Customers’ facts and figures. More often, it is basic math applied based on various average values based on aggregates at the Organization level applied at the individual Customer level.

There are two major issues with this.

• Your organization level averages can not be accurate for every individual Customer. You might be satisfied that you have some sort of CLV number to look at, but it is not good when you have to rely and act on it for any strategic or tactical marketing purposes.
• Even any advanced math-based CLV is based on historical numbers and tells you what happened but what good is it, if it can not tell you what a predictive CLV would be, in their entire engagement with your Organization?

Now, we know where the complexity lies in deriving CLV.

1. Every customer is unique in their engagement in terms of their interests, tastes, lifestyle, motivations, affordability, need, periodicity, seasonality, special occasions in their life, etc., So, if we apply average numbers for a CLV calculation for every Customer, you are oversimplifying the CLV number so much, that it almost becomes useless for gaining deeper insights.
2. If we all agree that, every individual Customer takes his/her own unique style or path in the buying patterns (while at large a persona might be relevant for precise targeting), we would be pretty convinced that applying averages is the most naive method for deriving CLV.

Is there a different formula for calculating CLV that will fix the above gaps?

As we now know the complexity and the origination of the issue with CLV, we can safely infer that it is not a one-step formula that can give us both predictive CLValso an accurate, precisely calculated number for each individual Customer.

Let’s break this down into two components.

Individual Customer’s data:

• Individual Customer’ purchase history needs to be factored in. Not aggregated numbers and averages at the Organization level.
• What entails this component:
  • An understanding of individual Customer’s recency, frequency, monetary value.
  • If you sell all similar priced product categories, CLV at only the Customer level might be sufficient.
  • If you sell multiple product categories, widely ranging in pricing also, you must consider a CLV at both Customer + Product category level.  

Predictive Customer Life Time Value:

What entails this component?

• Data assumptions play a key role in all these calculations. Assumptions are an output of industry research study in the domain in question. So, for instance, data assumptions about a subscription model are not the same as that of non-subscription or non-contractual business models like E-Commerce and Retail.
• A subscription-based model assumes that, if your customer does not renew the membership, they can be counted out. In the case of a non-contractual business model like E-Commerce, there is no easy, simple, definitive way of saying which of your Customer is active/inactive at any given point in time.
  
• 
• A Customer might have purchased an item a year ago but he/she might buy again in the future.
  • How do we predict the likelihood that this instance of the Customer is truly active/inactive etc.,?
    • This is dependent on the data assumptions understood and applied.
    • There are quite a few assumptions that are technical in nature but one of the top two research studies in the area of CLV differs in an assumption that a Customer dropout occurs immediately after purchase and that it is not independent of the occurrence of actual purchases. You may check out the reference link given at the bottom of the post.
    • A precisely tailored algorithmic approach that honors the industry research study assumptions that are translated into mathematical and statistical formulas.
    • Apply them in conjunction with the individual Customer’s facts and figures discussed in the section above.

The key takeaways are:

• CLV is an important insight for every E-Commerce and Retail business.
• Having an accurate CLV number requires a thorough understanding of what model (Subscription or Non-Subscription) is being analyzed.
• CLV algorithms developed for Subscription-based business are no good for non-subscription-based business models as they diverge at the most fundamental levels of data assumptions. It gives some dollar value for CLV but it won’t be any good if not done right.
• E-Commerce businesses require a specially tailored, predictive CLV that forms a key insight in further knowing your Customer Equity at the organization level.
• Predictive CLV is one of the key predictors (factors/variables) for the end prediction likelihood of purchase/repeat purchase as part of predictive Marketing.

I will expand in detail on each of the topics discussed so far, in the subsequent blog series. Currently, these are discussed at a very high level for easy consumption at a conceptual level. 

Thank you for reading the post. If you liked it, please share it with your network or leave a comment or question for me.

References: http://www.seas.upenn.edu/~cse400/CSE400_2013_2014/reports/13_report.pdf