In the past, companies were not particularly focused on considering their customers as human beings with emotions and the free will to make choices. Instead they were just viewed as a series of transactions: discrete-time cash flow events that either generated a profit or a loss for the company. So the focus of a business was just to decrease the cost to serve these customers and increase the profit margins of their products. The ultimate goal was to maximise the overall profitability of the company.

This worked for a long time but was flip-turned upside down very rapidly. The internet was invented; people became more computer literate and discovered that it’s pretty easy to steal other peoples ideas and implementations. The strategies behind the product-centric view described above were easy to copy. This is a small aspect of the idealogical shift that sparked the realisation in marketing departments that their suite of products are not necessarily their most important assest.

If managed expertly, your customer base becomes your companies most important asset. This is an asset that cannot be copied.

So modern marketing has made a clear shift towards viewing customer relationships with the utmost importance. The underlying notion is that companies should endeavour to provide superior customer value and satisfaction to continually develop and nurture long-term relationships with customers.

This sounds great in theory, but there are risks - it takes careful understanding to execute these strategies effectively. Relationship marketing is expensive. The risk for a company engaging in this model is that they can become over-enthusiastic about developing relationships and lose sight of their overall goal. It is a waste of resources and mental anguish trying to acquire and retain every customer in the world, and then exactly service every one of their needs. Companies should only be aiming to acquire and retain customers that they know will be profitable over the course of their entire relationship with the company. Rather than expending energy on increasing product revenues, strategies should now focus on creating longevity, breadth, depth and diversity of customer spending.

The expectations of customers have responded rapidly to these changes. Technology has vastly increased the number of channels through which ads can be delivered, and the general population has become much more savvy and selective about the marketing they pay attention to. Customers now expect (read: demand) targeted and personalised marketing, and have no time for relationships with companies who waste their time. If companies are unable to meet these expectations, they will struggle to remain competetive.

With this ever-changing marketing landscape, customer relationship modelling has become crucial for strategic success. Modelling these customer relationships provides a way to quantify and evaluate a companies relationships with it’s customers.

It’s important though to remember that data scientists aren’t the prophesised saviours; companies shouldn’t lose sight of the fact that it takes co-ordinated collaboration to effectively leverage data science in their business. The best practices for marketing combine both creative and analytical skills. Marketers are able to design experiments to see which of their channels are communicating with customers most effectively, or design tests to find out which new candidate marketing campaign will be the most effective. In this customer-centric world, marketing analysts should also be able to effectively model these customer relationships.

The purpose of customer relationship models is to inform decision making.

These models aim to understand the uncertainty that is inherent in any relationship with a customer, allowing the company to effectively manage their immense numbers of these relationships. There are basic models that capture the relationships of groups of similar customers, and there are new models constantly being developed that allow companies to shift from targeting segments to targeting individual customers.

This next section will give details of a general framework for customer relationship modelling solutions. The purpose of this is to give an indication of the size and complexity of such systems.

This is the first introductory post in a series of detailed tutorials of common models and methodologies. These will involve considerable amounts of math, and plenty of R code implementing the ideas that are presented. It will be continually updated as I research more and more models and develop more and more code (and might become a sneaky R package if there’s enough interest).

CLV: A Sea of Misunderstanding

What Is CLV?

Note: CLV is underpinned by the concept of the ‘Future Value of Money’ - which is presented on day one of any Intro to Finance course. This just means that you’re meant to accept that you don’t properly understand it and move on. I’ll try to get a finance professional to write a post about what this actually means and why it’s so universally accepted. Just roll with it for now.

“CLV is the discounted sum of future cash flows attributed the the relationship with a customer” (Pfeifer et al, 2005).

One of the main goals of a profitable company should be to maximise Customer Lifetime Value. CLV should also be used as a rationale for making decisions about resource allocation - a company should only invest in activities that will increase a customers incremental CLV above relevant costs.

Just measuring and knowing CLV is fairly useless. What companies need to consider is the effects that their decisions have on CLV and more importantly on the underlying customer relationships.

What Isn’t CLV?

This following list is a few definitions of terms that are absolutely relevant in the discussions and modelling of customer relationships, but are not CLV:

  • Customer Profitability - The difference between revenues earned from, and costs associated with, a customer relationship during a specific period. The key difference here is that profitability is not probabilistic, and does not account for future values. It just summarises past value.

  • Frequency - The number of previous purchases made by a customer.

  • Monetary Value - The sum of revenues earned from a customer during a period in the past. Monetary value can be directly observed from data, while CLV cannot; it occurs probabilistically in the future so must be estimated with a statistical model.

  • Customer Equity - The sum of CLV over individuals within some chosen group of customers. If you choose the group to be all of the customers, then you have a nice little estimate of the value of your customer relationships. This can also be used to quantifiably forecast the value of a companies customers.

  • Long-Term Value - This one is pretty subtle. CLV is calculated over the entire lifetime of a customers relationship with a business. Mathematically, this necessitates summing to infinity over time values, and then taking advantage of convergence properties for numerical calculations. Long-term Value however shortens this to a more reasonable time period - like a year or 10 years or something more manageable. CLV and LTV are different concepts mathematically, but people use the terms interchangeably in conversation.

The Value of Customer Relationships

So CLV is just one aspect of a customer relationship and decision making framework. And it will be a huge task, so there better be a good reason to do it. The purpose of such a system is to manage the interactions between an organisation and it’s precious customers. And at the end of the day we want to increase the profitability of the business by increasing CLV.

For the following section we will need the definition of a contact point. A contact point represents an interaction with a customer. Contact points include all customer encounters with a brand, not just a purchase. Some examples:

  • Contact point initiated by the business (e.g. ads, direct marketing)
  • Contact points initiated by customers (e.g. calling a support line, or simply using the product)
  • More passive contact points (e.g. product reviews, internet personality endorsements)

Then a general process for increasing CLV should consider four main steps:

Step 1: Understand, Segment, Value

Main Goal: The main goal of this step is to understand the needs of customers, and then segment them based on these needs in order to reduce the size, complexity and error present within the problems being solved.

Ways in which CLV can be increased wil vary across customers because of the inherent heterogeneity of wants and needs across different humans. This means that contact points should be tailored to different customers, and should always be financially justified.

It is not a simple task to understand the wants and needs of a customer. Better understanding of customers motivations will come from expertly designed studies designed collaboratively by data science and marketing teams. There also needs to be co-ordinated work from the product teams to deliver a product that people actually want.

This goes a little beyond the scope of the tutorials that will be in this series. I’ll be presenting models that focus on customers future behaviours; estimating their future value and identifying opportunities for marketing intervention. I will get to post-hoc experimentation one day I promise.

Step 2: Strategies for Increasing CLV

Main Goal: To set customer objectives and company spending levels.

So once you’ve perfectly executed Step 1, you will have a number of customer segments. You can then specify objectives for each of the segments, informing the creation and maintenance of contact points (detailed in the next couple of steps).

The objective should always be to increase CLV. At the very least it should be maintained.

This is why CLV becomes the central concept that people talk about when discussing customer relationship modelling. And why it’s definition gets so easily confused. It is important to measure CLV because you always want to make decisions to increase it. But you need a massive suite of modelling techniques to measure it’s influencing factors and to measure the effects of changing all the variables that a company has at their disposal.

There are four basic ways to increase CLV:

  1. Retain them longer
  2. Increase Customer Revenues
  3. Decrease the cost of service
  4. Decrease the cost of marketing to customers

So if the customer objective is based around the first case here (i.e. retention) then the company should be trying to predict when customers will leave, and then developing contact points to avoid these relationships being terminated. However the cost of creating and deploying the contact point should not exceed the incremental value of CLV. (An example in later posts will show how to model the effects of varying retention rates).

If the objective is based around the second case (i.e. customer revenues) then this is generally done in 4 possible ways:

  • Increasing share-of-wallet - getting customers to stop paying other companies for the services that you offer

  • Cross-Selling - using marketing techniques to sell customers other products that you offer

  • Up-Selling - getting customers to purchase higher margin products from you (e.g. a premium service or cooler sunnies)

  • Increasing frequency - getting customers to buy things from you more often

If the objective is based on the third case (costs of service), then this becomes another case where studies need to be designed to inform decisions like whether to charge for using offline services, determining lengths of contracts, increasing cancellation fees etc.

Costs of service and marketing costs are worth giving separate headings to because there are some subtle differences. Mainly around the fact that it is not an easy task to quantify marketing value. Measuring less tractable concepts like customer engagement value make this a bit more of a random process than designing a product contract plan - since a contract won’t tell it’s friends that it hates you.

The logic above has been applied in the context of retaining existing customers. But it generalises without much added complexity to increasing the CLV of prospective and former customers:

  • Prospective: The objective becomes to get an initial purchase from a potential customer. This may be broken down into shorter-term subgoals (and can then be handled by the marketers heh). E.g. set objective to get a click on a web ad, then a product trial, and then eventually a purchase.

  • Former: Objective becomes to understand why they left and to develop contact points to bring them back.

Another aspect of the (honestly pretty massive) second step of the framework is to set the spending levels of marketing campaigns. The main point of these decisons is really main so it goes in bold:

“The incremental gain in CLV informs how much money can be spent on a marketing decision, NOT the absolute level of CLV”.

This will become more clear in later examples, but the basic intuition is that “investing more resources in your best customers” is not a sensible decision if throwing more money at them won’t make you any more money (because maybe they’re already on the most expensive plan and love it and will never leave).

Step 3: Create and Monitor Contact Points

So once you have models to help you see what customers are likely to do in the future, someone has to actually go and do something about it. This part of the process leans more heavily on marketers. They ultimately control the communication pipelines with customers, and have a much deeper understanding of previous context around brand consistency and appropriateness of contact methods.

The main take-away from this step is that the decisions around contact points should be informed by the impact they have on CLV - not just on arbitrary metrics like “oh yep we’ll put 10% more into marketing than last year” or “just chuck 20% of profits at it”.

Marketing is still definitely a creative enterprise though, so while models should inform decisions, the other part of the genius that goes with a successful marketing campaign is in the creative inspiration that it takes to get people excited for a new product, or keeping them happy with your current products.

Step 4: Measuring Outcomes

Main Goal: To measure what happens as a result of the contact points the company decides to implement.

This again becomes an exercise in determining the effects of any marketing actions that get implemented. Modelling work is done to inform this decision making process, and studies must be designed to understand the response of your customer base.

Knowing the outcomes of such experiments will then inform future decisions. You have to follow a customer as their relationship with the company changes and should always be making the best decisions to increase their incremental CLV. Steps 1, 2 and 3 above are not ‘set and forget’ types of decisions, companies should be repeating this process every time a customer interacts with them.

Let’s Get Mathematical

The above sections give a conceptual overview of customer relationships, and a brief modelling framework. Now it’s time to get into the math behind these models. The following tutorials will be split over a number of blog posts, so relevant links will be posted below as they’re uploaded.