Understanding the Customer Lifetime Value (CLV)

During a town hall with our CEO when I was a junior business analyst, I asked what type of company we were. The answer was that we were a “customer” company. It is only through the following years that I fully started to understand what it meant to be a customer company. All companies have customers (whether consumers or other businesses), yet very few understand the strategical aspect of being a customer-centric company. When you place the customer first, you force the company to create products and offer services for the customers and not for yourself.

In the journey of being a customer company, it is also important to understand the customers’ behaviors to provide them a tailored experience. In my experience, Google has become a key leader in providing the ultimate customer experience with its software. It has done this through diligent analysis and experimenting on the user experience.

Being a customer company also means to be a company, and the financial aspect of operation and customer actions are critical. One wants to create more actions that drive value to the customer and from the customer. As such, it is critical for all companies to have a strong understanding of the total value of each customer: a Customer Lifetime Value or CLV. To reach this objective, you will need to start a CLV project, and the note below highlights the key steps in this project.

If you want, you can directly jump to the end of this post to see the checklist on how to get started immediately.

1. What Is A “Customer”?

The first step in the CLV journey is to define a customer. Although this seems basic, it is often not as trivial as it seems. Is it an account? Is it a person who might have multiple accounts? Is it a household, which can include multiple persons? The usual short-ended solution is to pick an account, yet this defies the objective of the CLV project, which is to better understand the impact that a customer can have on your business.

For example, let’s assume you define a customer as an account. You might identify an account with low value and provide that account with a lower tier of customer service under the assumption that, if the account leaves, the impact to the business will be minimal. However, that would be fundamentally wrong if the person behind this account has other accounts, which could be highly valuable to your company. If you upset that person, he might well terminate all his accounts, and your company will be impacted to a higher degree than you initially thought.

The bottom line is to define a customer at the level where decisions are made: to start or stop using your products and services. This is usually done at the person level although you might want to consider a household.

Another argument to use on the person level is to be able to reuse the insights and segmentation you derive from a customer reopening a new account after he has left your company. In traditional CLV project, an account defines the customer. For example, if you leave your mobile provider company and decide to come back later, you will be considered as a new account, and, hence, a new customer. Although this approach might simplify the legal and privacy aspect, it strongly diminishes the ability to understand the customers and provide them a tailored experience.

The ability to track activity at the user level is critical for any CLV project. This can be done by assuring that a sound tracking system is in place on your websites, mobile apps, and offline stores. It is critical that you can identify the customer for all events that have a financial impact (revenues or costs). This requires, for example, adding a user identification tag to your mobile and web tracking.

One of the primary advantages of starting a CLV project is that it will enable you to assess the quality of your data. At the same time, you also need to identify legal boundaries in your tracking activities and respect the user privacy laws in the market in which you operate.

Beyond tracking, you will also need to review your backend data flow and unify all your data sources into a single repository. You could also augment your data with 3rd-party data such as Acxiom data to enable better customer segmentations. In the end, you need to have a unified place where all your data reside. As this include web traffic and Internet of Things events, you are most likely looking at a Big Data infrastructure like Google BigQuery, Teradata, or a Hadoop system.

2. How Does The Customer Provide Revenue?

The second step is to define the revenue streams that can be associated with a customer’s action or decision. Some are simple and straightforward, but some others are more complex: events triggered by multiple customers, the allocation of subscription revenues, etc.

For example, on eBay, when an item is sold, the transaction fee revenue from that sale can be attributed to the seller as well as to the buyer. On other marketplaces, such at Bottle Waiting or Open Table, when a table is booked in a nightclub or a restaurant, the customer (host) offering the venue and the consumer booking the table are co-responsible for the revenue.

One way to solve this is to admit lack of knowledge and split the apple in two, providing one-half of the revenue to each side. Another way is to identify whether you are in a demand-constrained market or a supply-constrained market. In a pure demand-constrained market, the offer from the “sellers” is close to unlimited, and only the buyer is responsible for the revenue. In a pure supply-constrained market, the demand from the “buyers” is close to unlimited, and only the seller is responsible for the revenue. The truth is somewhere in the middle. With advanced analytics, you could find a better solution than 50/50.

3. How Does The Customer Generate Cost?

The third step is to identify cost driven by the customers. Some are straightforward, such as incentives, direct marketing, or customer service. Other P&L costs are more subtle: usage of the website, operation cost, and overhead cost from various business functions.

When looking at marketing cost, you will need to identify whether you want to allocate a cost per customer or a cost per prospect.

For example, you send a letter to 100 prospects (potential future customers), and the cost is $1 per letter, so there is a $100 total cost for this campaign. If only two prospects start using your company, do you want to consider only their own letter cost ($1 cost to acquire each one of the two new customers) or the full cost of this campaign ($50 cost to acquire each one of the two new customers). There is value in both methods. It is really up to your company to decide which way to go.

When looking at operation costs, you need to identify how you can allocate the cost at the customer level. As not all customers are equal, it is really important to identify an allocation scheme that it follows the real activity as closely as possible. For example, when considering the website cost, you could identify a metric that closely scales with the customer activity. You could use the number of web page views or the number of web searches. Working with your IT organization will be critical to understanding the real driver of cost and how scalable or fixed the cost is.

I often hear that some costs are fixed and wouldn’t change if you added or removed a customer. Although it is true that such costs are fixed, it is also true that they can be seen as flexible costs at the macro level. Indeed, should your company serve two times more customers or two times fewer, would the cost be the same? Probably not. Hence, although they can be seen as fixed costs at the micro level, they can be easily approximated as a scalable cost for this CLV project.

A good way to check on the progress of the CLV project is to run a report comparing the monthly P&L report with the total aggregated revenues and costs by area from the CLV project. This means that you should compare the total monthly marketing cost from the P&L with the sum of the marketing cost from all your customers for that month. The closer you get to covering 100% of your P&L, the better.

Concerning employee-related costs in which teams are not directly linked to customer activity (administration, finance, human resources, etc.), we could either ignore them in the model or peanut butter the total cost to all customers. The advantage of the latter is to reflect better the full P&L, yet from a customer ranking perspective, it will not help as you are just lowering the total customer value by the same amount for every customer. Again, it is up to your company to decide how to move forward.

4. Indirect Value Driven By A Customer

Some activities from customers do not trigger direct revenue and cost but can trigger a halo of positive (or negative) value. This would be the case, for example, when a customer recommends your company to his friends and some of them join. Part of their CLV should be attributed back to the referral customer. In a similar way, a customer who is providing a unique offering could be awarded extra credit for his score to reflect the importance of his contribution to the company. On many websites, feedback and reviews are very important to helping a potential buyer purchase what he wants. As such, it is key to attribute the value of feedback and reviews back to the customer who left them.

At the same time, some bad actions from customers can create harm to your company, and you might want to attribute some of this lack of value back on this customer. Examples include a marketplace seller providing very bad customer experiences to the buyers or a customer who systematically breaks your policies.

5. Combining All This Into A Customer Value

Now that we have a flow of direct and indirect revenues and costs, we need to combine them into a unique total that best reflects the current value of the customer. Based on your business, you might want to include a shorter or longer range of time. I recommend using a 12-month window as this accommodates for seasonality as well as provides a short history of activity. One could use weighting factors so that recent months are worth more, yet this would create a fluctuation across the year due to seasonality.

With this Customer Value, you can directly provide a ranking for all your customers and implement this into your operation. For example, when a customer who is part of your top 5% most valuable customers calls your customer service, you might want to reroute that call to a team of senior agents. You also want to show the customer tier visually on the customer service portal during calls. If an agent spent a full hour (worth $25 of fully allocated costs for example) with a customer who has provided a $5 profit in the last 12 months, you need to hope for a significant turnaround in the customer’s behavior to compensate for that cost.

6. Predicting Customer Lifetime Value

The next step of this CLV process is to combine all this information from the customer into a predicting algorithm for the full lifetime of the customer. You know the total value that the customer has brought to your company since his acquisition. Now, you need to develop a predicting model to define the revenue and cost for each following month that you will weigh with the likelihood that the customer will still be “alive,” i.e. still be an active customer.

In a traditional CLV project where you consider accounts (instead of people) as customers and where your business model is based on subscriptions, this predicting model is rather trivial.

However, in a non-subscription-based business model where it is free for the customer to join and to leave, this creates an issue. In my experience, the measurement of future value and the likelihood of being active reach a plateau very quickly. As such, there is a point where each future month will increase the CLV by the same amount until infinity (well, actually, until the physical death of the customer). In mathematical terms, this is a non-converging series, and it yields to an infinite value. As this is not useful for the business execution, I recommend that we define a time horizon that makes sense. For example, we can define the CLV as the current value since the acquisition and the future value for the next five years. I usually recommend a much shorter window of 12 months. This will enable us to create a sense of urgency and work on very impactful short-term initiatives as well as remove any seasonal effect. However, if you execute costly initiatives with payback time ranges in multiple years, you might want to increase this window.

When trying to predict future value, you can also compare to the most basic predictor, which is to say that the next 12 months will be the same as the last 12 months. Although this seems very trivial, my experience of working with half a dozen different statisticians showed me that it was the best model across a wide range of algorithms. Although many of those algorithms were very good at a cohort level, they significantly failed at the customer level. As such, I often recommend using the last 12-month customer value as the CLV score.

7. Showcase CLV To The Company Using A Customer-Level Dashboard

CLV can be rightfully perceived as very complex. With complex attribution schemes and statistical models, CLV is a very unfriendly project to showcase. However, you can significantly improve your communication of the CLV project by developing a CLV dashboard at a customer level and then enabling your employees and colleagues to see their own CLV reports. This report can include the CLV value itself as well as revenue and cost charts, key behaviors, and recommendations on how to increase CLV (see point 8 below).

We had tremendous success when we showed those reports at an extended senior leadership team event. All the participants received a closed envelope with their own personal CLV reports. Senior leaders being quite competitive, it was a greatly enthusiastic scene with all the participants sharing and comparing their values using their own products/services.

This created a strongly positive halo to extend the CLV metrics, reports, and predictions across the company.

8. Build A Recommendation Engine

Every company is offering more than just one product or service. There is a variety of features, options, or upgrades that a customer might choose from. This can also include good tips and tricks on how to use the products. There are also bad events or actions that would result in a decrease in customer value.

In the objective of increasing customer value, it is essential that you understand which action will impact each customer the most. You can use statistical models to understand the change (increase or decrease) of the customer value by actions. Then, you can set up a recommendation engine that will calculate and rank the impact from all the potential actions of all customers. Finally, you can inject this tactical information into your business operation.

The first step is to organize small workshops with key domain matter experts to collect all the actions that could change the value of a customer. For example, if the customer starts using various features of your product or if the customers receive some specific outreach or marketing material, the workshop should start to prioritize the impact on the CLV value using simple t-shirt sizes: small impact, medium impact, and large impact. You might also want to address the ease of conversion. In the end, you would like to identify the low-hanging fruits: large-impact actions which are easy to execute.

Then, you will need to measure the change in CLV due to each action using statistical methodologies. You can also use the simple methodology to compare the average (or median) CLV for the cohort of customers who have gone through the action versus the cohort that has not. You need to be extremely careful of self-selection bias.

For example, if you are a car manufacturer, you might find that customers who purchase premium cars are much more valuable than customers who purchase budget cars. Hence, the straightforward conclusion is that you want to push customers to purchase premium cars instead of budget cars. This might be inaccurate due to the self-selection bias. Here, “rich” people purchase premium cars, and “poor” people purchase budget cars. As you cannot convert a “poor” customer into a “rich” customer, prioritizing this upgrade to premium cars might not be the most interesting action to do. However, if you normalize your cohort of customers and keep them similar, then you could find that “poor” people who purchase budget cars can be easily converted into “poor” people who purchase budget cars with a financing offer.

Then, you need to define the optimal time when to recommend an action based on the life cycle of the customer. The simplest method is to define a triggering metric, and when that metric reach a specific threshold, the recommendation will kick in.

For example, as an eBay seller, having a store is very beneficial if you have a large inventory. However, small sellers will be overwhelmed with the complexity of opening and maintaining a store. Here, the triggering metric could be the total number of items ever sold on eBay by the seller. The threshold could be for example 10 items, i.e. after his 10th sales, a seller who has not yet opened a store, would qualify for the recommendation: “open a store.”

Once you have an engine that scores all potential actions for all customers, you can rank the best actions for each customer and inject this information into your operation. Propose those actions at the point of sales, during customer care calls, through marketing campaigns, etc.

9. Inject CLV And Recommendation Across The Company

The final step of the CLV projects is to hook up all the operations of your company with the CLV score and the customers’ recommendations.

For example, you can implement the CLV score in your customer care call centers. Indeed, it is critical to understand the historical value and the potential future value when routing a customer call to the relevant agent. Ideally, all customers should receive the best support from the most expert agents. However, in reality, the number of expert agents is limited; hence, you need to define who should get access to that top-tier support. In the same way, for very low-value customers (or even negative-value customers), you might want to think twice before sending them to an agent that might cost $10 per call. Ideally, you would like such customers to find the answers to their problem by themselves using your company website or through automated and smart answering machines.

Let’s Get Started

In summary, it is critical for every company to understand the value of its customers whether they are consumers or other businesses. To fully access this customer value, you can follow the following steps:

1. Define what a customer is.
2. Define which revenue elements are triggered by the customers’ actions.
3. Define which expense elements are triggered by the customers’ actions.
4. Define any indirect values triggered by the customer (promoters, etc.).
5. Combine all data elements into a unique repository and provide the current value for each customer.
6. Predict the future value and calculate the total lifetime value for each customer.
7. Having now a CLV score for each customer, celebrate and communicate the process to the rest of the company before implementing CLV into your operation.
8. Set up a statistical model to measure the impact on CLV for large collections of customer actions and sort the top recommendations for each customer.
9. Inject CLV scores and customer recommendations across your company (marketing, customer care, sales, etc.).