Customer Lifetime Value Modeling (Part 1): Overview and Business Applications
Who benefits from customer lifetime modeling?
The foundation of any business is the sale of goods and services to some type of customer, and the lifetime value (LTV or CLV) of such customers is a key metric for the reporting and decision making of all organizations. It is, in fact, one of the most important metrics for a business, as it summarizes revenue and cost over time all in one.
In my experience working with a dozen B2C marketing teams, I have found that many are not actively using LTV in their day-to-day operations, and some organizations have not even built out the capability to calculate it. Does this surprise you too? Do organizations hyper-focus on customer acquisition and product-market fit so much that they haven’t had a chance to look in the rear-view mirror to characterize the customers they have acquired so far? Are there too many technical/organizational hurdles to deploying LTV that some leaders have decided not to invest in it? Do they have it but keep it siloed from marketing teams? Or is there something else happening altogether?
To answer these, I collected some thoughts around a more practical question: Who benefits from LTV models?
In part 1 of this series on customer lifetime value, I review customer LTV – both the traditional and more recent formulations – and discuss the crucial role of LTV in today’s business environment, providing examples of business functions that benefit from having these calculations available. I’ll leave a more technical discussion of LTV modeling for a separate article.
Defining Customer Lifetime Value
Customer lifetime value is a forecast for future contribution margins calculated at the customer level.
LTV is not calculated using a simple formula. As we’ll see in this series, doing LTV correctly requires designing the right data infrastructure, finding alignment among various stakeholders, and building, tuning, and validating the actual models themselves.
Let’s break down the definition by discussing how those bolded words above impact the LTV framework setup.
LTV is a Forecast…
LTV is a future-looking metric. Its value for a customer depends both on historical purchase information and the methodology chosen by decision makers in the organization to do the modeling. There is no universal formula, as I’ll explain below, and all stakeholders should understand its assumptions and inputs in order to understand its limitations. You want ChatGPT to calculate your LTV? Tough luck. There is no global LTV model that automatically works out of the box, and every organization makes important decisions that influence how this number is calculated. We’ll tackle the inputs and assumptions in the following sections.
… for future Contribution Margins…
Formally, LTV is defined as the present value of expected future contribution margins (revenue minus costs for a customer). It involves knowing the cost of acquisition and ongoing servicing costs, such as marketing and sales efforts. This presents some technical complications, so I’ll mention up front that the related metric of Lifetime Revenue (LTR) is sometimes an acceptable alternative. LTV, however, requires merging data sources, as sales and cost data often live in separate entities. It also requires the business to forecast ongoing costs and decide on how (and whether) to amortize various costs to each individual customer.
… at the customer level.
LTV is calculated for each customer. That requires a cohesive identity resolution framework within the data infrastructure to be as accurate as possible. The granularity at the customer level is useful - it enables value-based decision making and unlocks a variety of interesting analyses like value-based customer segmentation.
Side note: If you want to calculate a more global metric, you can derive Customer Equity by summing all existing customers’ lifetime values. In essence, this is the revenue forecast. It indicates the health of the business at the current snapshot of time.
So, how do we actually model LTV?
LTV 1.0: Contractual products and services
The traditional formulation of LTV is based on a rather narrow use case: contractual products and services. It is equal to the net present value (NPV) of future margins for a customer that pays fixed amounts at a fixed cadence. This customer has the ability to cancel their contract and has to explicitly elect to do so. Not all products and services are contractual, and we’ll get to these in the next section.
Suppose you have a SaaS product that customers pay for annually. As you acquire new customers, they start contributing to your margin. At any point in time, there is a probability that each customer will churn, or in the typical formulation, there is a retention rate for the cohort this customer belongs to. Before the end of the contract (one year, in this example), some fraction of them will cancel future payments of the contract.
This is crucial for the traditional LTV calculation. These churning customers have actually raised their hand to say they will no longer be a customer. Once done, their future revenue is known unequivocally to be zero. There may be marketing efforts to recapture lost customers, but this traditional formulation of LTV would treat them as net-new customers (perhaps in a segment of their own with unique characteristics like retention rate, CAC, etc).
To calculate LTV for contractual products and services, we need the individual’s predicted present value, the retention rate of their segment, and a discount rate encoding the time value of money (i.e. future revenue is less valuable than revenue today).
Say we decide to calculate the LTV of customer A. We introduce the forecasted values by year, where years are indicated by the subscript n.
Let’s break down each piece:
Rn = customer revenue forecasted for year n
Cn = cost of marketing and servicing this customer forecasted for year n
r = retention rate forecasted for year n; probability that the customer does not cancel their service that year. This can be specific to the customer segment to capture differences in audience type, and is expected to increase over time as retained customers become easier to retain1.
d = discount rate encoding the time value of money. Standard practice is to use 0.1, but this needs alignment with stakeholders.
CAC = acquisition cost of this customer
This formula might differ slightly from one you’re familiar with; if so, this is a great illustration of how underlying modeling assumptions manifest themselves in the calculation. The formula will change depending on how long you are forecasting (one year? a few years? out to infinity?), whether you include the customer’s first purchase or only future purchases, and whether the billing happens at the beginning or end of the year.2
Even with a relatively simple formula like this, we can start to see the modeling challenges. We need to merge sales and cost data, which involves building a data infrastructure to house them with correct governance. Forecasting acquisition costs and ongoing costs is both art and science with a sprinkle of internal diplomacy. Discount rates are an assumption of the model, and retention rates must be estimated (the simplest approach is a global average across all historical customers, but machine learning techniques are available for an advanced treatment).
LTV 2.0: Non-contractual products and services
Most products and services are sold on a non-contractual basis. Think of an e-commerce store selling a variety of products and the cadence at which any one customer’s purchases come in. You simply don’t know a priori how a customer will behave. They may purchase a single time, regularly, or multiple time with no predictable pattern. The size of their purchases can be similar over time or variable. Most importantly for modeling non-contractuals, you don’t know when a customer has stopped being a customer because there is no reliable signal (such as a subscription cancellation) to indicate the customer has been churned. One customer might provide the same revenue as another but at a longer cadence, and they are equally important to capture in the model.
Peter Fader of Wharton and colleagues have been publishing research on LTV for these non-contractual products for the past decade or so. Their findings have resulted in a framework called Buy Til You Die using RFM (recency, frequency, monetary value) data. The concept is simple: a customer will continue to Buy at each interval with a probability of “Dying” given by a certain distribution, after which they have officially churned and all their future revenues are zero. One key difference from the contractual product formulation (LTV 1.0) is that retention rate for each time period and customer cohort is sampled from a distribution learned from the data.
Here’s the cool part. These models can work well with just those 3 RFM inputs: how recently the customer purchased, the number of purchases (frequency) in the time interval of interest, and average monetary value of their purchases. This is quite incredible given how many variables we intuitively think must be relevant in evaluating a customer. Notably, these are behavioral data collected naturally in the course of doing business. Demographic and other exogenous data features can improve the base model but are simply refinements.
Note that cost data needs to be incorporated into the model in order to reflect margin rather than revenue. This is a detail that many introductions and code samples on the topic of LTV gloss over or neglect.
Model Validation and Maintenance
Whether you are selling a contractual or non-contractual product, you can gain confidence in your LTV using a validation approach that shows how well the model would have worked in the past if you had been using it. In other words, learn the parameters of the model, then see how well a model with those parameters works on customers acquired after the time period of data used in building the model. As long as the business has enough historical data, you can validate any LTV model on a time split in which the earlier time period is used as a training set and the next time period is used for testing and tuning the model. The best model parameters are then used to build the production model to be used in the future.
To successfully transform a business, LTV must become a core metric in ongoing decision making. Maintaining the LTV framework involves a number of responsibilities. Data ingestion should be automated to minimize delays and human error in using the model. The model(s) should be deployed in an orchestrated environment so as to be up-to-date with recent purchase data and cost calculations. The models need to be retrained whenever their performance starts to degrade. These considerations should be part of the roadmap for integration.
So, who benefits from LTV modeling?
Part of my recent work was to built a lead scoring system for a sales team that incorporated, among other things, an LTV calculation for each existing customer. This allows the sales team to target their efforts on the small fraction of prospects likely to be high-value customers out of a large database of leads. This application of LTV to lead prioritization illustrates one of many ways to use this metric, and here are a few other examples3:
Customer Acquisition
Through marketing and sales decisions, businesses exert control over the types of customers they acquire. If an organization understands who their high-value customers are through LTV, they can prioritize their efforts toward those high-value customers and improve sales/marketing efficiency.
High-value customers are those who stay customers longer, buy more products/services, are easy to service, are loyal to your brand, and even provide referrals. In essence, they provide more revenue and/or better cost savings (in the form of lower CAC, ongoing marketing costs, or customer service burden) compared to the average customer. LTV, capturing both cost and revenue, incorporates these characteristics using real data in one metric.
But how do you identify a high-value customer before they enter your ecosystem? Teams will attempt to bring in a wide range of customers through marketing while narrowly aiming their direct sales tactics on more promising leads. On the direct sales side, this is a multi-step process: first, the business has to calculate LTV for the existing customer base, then they profile the existing customers to discover the characteristics of their high-value segment, then they can apply lead scoring to prospects and go after the leads most likely to be in the high-value category. For profiling, we’re not talking about creating demographics and personas; forcing the high-value segment to be an easy-to-remember persona, or an oversimplified demo, dilutes the data and creates confusion for anyone trying to use this method in practice. The better alternative is to identify a full set of characteristics that sets high-value customers apart. Then the task for marketing/sales teams becomes finding individuals in the world who fit that profile through any number of social networks, purchased lists, and referrals.
Customer Retention
LTV enables teams to balance their efforts in retaining customers. Obviously, high-value customers are important to retain, but you should also attempt to retain low-value customers, though not with as much effort. Retaining customers, through exceptional customer service and product quality, costs money. White-gloved customer service requires extensive resources, so it should be retained for high-value customers. You need to know who those customers are, and LTV provides the most rigorous quantification of their value.
Strategic account management is already the norm for many client facing teams. When I used to run media campaigns, I remember distinctly the differences in customer service my clients received from the large marketing platforms; brands who spent smaller amounts on media received sparse, low-quality service (“please reach out to customer service through this automated chatbot”), while large media spend clients were assigned personal reps who checked in monthly in an effort to be extra helpful.
Customer Development
Development means cross-selling and up-selling to existing customers. In either case, it means attempting to expand revenue with that customer. These activities often cost valuable resources, e.g. direct sales team hours. Even an automated cross-selling capability, like an engine recommending similar purchases to end users, needs significant data science investment to work well. Therefore, understanding the value of a customer is key to justifying the effort to develop them.
From loyalty programs to premium services and perks, there is an appropriate customer LTV-based value segment for each development initiative. For example, customers who buy product from you regularly might not be ideal for a loyalty program – the freebie you give them, they would have bought from you anyway! However, a customer who purchases infrequently but in bulk may choose your business over a competitor’s if given an extra incentive.
Media Planning
Marketing teams can use customer LTV to guide their budgeting and channel allocations. If the average LTV is $10, then spending over $10 to acquire the average customer through marketing is unprofitable. This could be acceptable if the business is investing in customer growth, but even so, I have found that many marketing teams benefit from having a benchmark for acquisition cost. In fact, many use the LTV-to-CAC ratio to guide their decisions (often a ratio of 3:1 is considered “good” but it clearly depends on the vertical and product).
Through pixel-based conversion tracking, many digital advertising platforms enable the estimation of Cost Per Conversion type metrics. Keeping in mind the limitations of digital attribution and that each channel plays a unique role in the marketing lifecycle, LTV can be a powerful reference point for channel efficiency. Analyzing LTV at the customer level allows for more nuance in evaluating the success of channel strategies (as opposed to comparing them on a platform-level metric like ROAS). Knowing the tier of the customers you can acquire from each marketing channel gives significant confidence to the budgeting exercise in media planning.
Corporate Valuation
Organizations can incorporate customer equity based on a rigorous LTV model in reporting the value of the business to stakeholders. It can guide data-driven conversations ranging from customer base growth to marketing and product development. Customer Equity complements Brand Equity, another metric recently taking hold in corporate valuation conversations.
Why are some businesses slow to adopt LTV?
Back to the original questions around adoption, we can start to see the challenges and propriety of building lifetime value modeling into a business. LTV frameworks are sophisticated and require a number of steps to implement. The main challenges we’ve discussed are:
Cost and sales data must be united to obtain the inputs to the models.
Retention (and ideally, discount) rates must be modeled. Academic literature makes it clear that it becomes easier to retain customers the longer they are purchasers.
Non-contractual products must be treated differently from subscriptions.
Business leaders and stakeholders must align on modeling methodology.
Ongoing maintenance of the LTV framework is costly.
Businesses who are using LTV for their sales, marketing, or corporate evaluation tend to be larger and more mature, which implies that they have the investment and staffing to build LTV into their processes.
More importantly, earlier-stage companies tend to focus on customer acquisition at all costs. They do not have the luxury of being picky. In fact, their success is measured in revenue and growth, rather than efficiency. At some point, this mentality shifts and cost efficiency starts to become a priority. This is when LTV should be incorporated into the business.
You should be calculating LTV. Here’s how.
Hopefully this lengthy exposition has illuminated the different flavors and applications of customer LTV. Investing in this metric enables you to keep on top of the health of your customer base and profitability, and that’s just the beginning. Using LTV to set marketing budgets, segment audiences for product development, and decide on future growth directions is where I think businesses really benefit.
The earlier a business can prepare the way for LTV, the sooner it can benefit from the efficiencies of value-based decision making. The broad steps to building out LTV capabilities are as follows:
Mapping out the types of products and services you sell (contractual v. non-contractual, pricing tiers, purchase methods, etc.)
Aligning with stakeholders on assumptions and inputs, everything from discount rates to modeling methodology and validation criteria
Uniting the data sources into a warehouse, with correct governance, to allow sales and cost data to be input into the models regularly
Building LTV models and validating them against holdout data
Deploying the models into a production setting for automated updates to LTV
Communicating LTV through reporting and dashboard tools
Retraining the LTV model(s) as needed for optimal performance
Janszki Consulting is prepared to help you become data-smart. If you’re interested in building a modern LTV framework for your organization, or you want to find out if your data is ready for LTV, please reach out.
See Fader & Toms, 2018 for further detail on these concepts.