Lifetime Value
On the Growth team at Coinbase, I spend a lot of time on programs focused on acquiring new customers or activating users on new products.
But like any business, every project is evaluated based on ROI in terms of money. On one end is the cost of attracting new users, which is already represented by marketing budgets. On the other hand is the value of those new users that sign up.
This raises an important question: How do we assess the monetary value of attracting new users? This is where lifetime value (LTV) comes in.
What is it?
Lifetime value is the average amount that a customer is expected to spend over the course of their time using a company’s product/service. It literally puts a dollar value on the worth of a customer.
There’s a lot of ways to calculate LTV, and they vary in pros/cons, as well as complexity.
For instance, for a subscription service, a quick estimate for it is the monthly subscription fee divided by the churn rate (the rate at which customers leave the service).
LTV can also be addressed as a regression problem in machine learning. In particular, the target variable would be the revenue generated by an individual user, with the input data being the metadata we’re able to collect on that user and their activity in the first few weeks. This does beg the question: How much data are can be collected on new users for use in an LTV model?
This also raises questions of “bake time”; how much time does a customer need to spend in order to get a good estimate? Of course, the more data we have, the better the prediction will be. But the longer we wait for data, the less useful that prediction becomes. This creates a balancing act in terms of accuracy vs usefulness. Ideally, we could get a precise prediction the day a user puts their credit card information in, but that’s not realistic. So in order to have a prediction that’s useful, we will have to sacrifice some accuracy.
In fact, that plays into the first rule of statistical modeling:
“All models are wrong, but some are useful” - George Box
Ultimately, every user is different, and no model can perfectly predict how much a user will spend.
Each prediction from an LTV model comes with error bars. Those error bars can (and should) be used to create a best and worst-case scenario for each campaign. In my experience, one of the hardest things to communicate from statistical models is the uncertainty that comes with it. When predictions are presented as point estimates (i.e.: a single number), they can be mistakenly interpreted as fact.
Why is it valuable?
1: LTV represents an estimate for marketing and customer incentive budgets.
If LTV is the monetary value of a new customer, then it’s unprofitable to spend any more than that number to acquire new users. For instance, Coinbase offers $5 to every new user that signs up. So it follows that the LTV is higher than $5.
For instance, If a company spends $100,000 on social media ads, and 2000 users sign up, then the LTV makes the difference between whether or not that social media spending was successful.
2: It’s a measure of customer engagement and retention.
As a company whose primary revenue source is transaction fees, customer engagement is crucial. The more customers interact with our software, the more revenue they generate, and consequently the higher the LTV.
Unfortunately, we’re currently facing a crypto winter, where users are either liquidating their crypto assets or HODL-ing them. Either way, people trade less, and that means a lower overall average LTV per user.
3: It helps target and influence high-value customers
If we can estimate lifetime value across different customer types, and even have a model to predict it at the individual level, then that sheds light on factors that are associated with high value customers. Some customers spend $10, some spend $100, and some spend $1000. With that knowledge, a business can be better equipped to either acquire more high-value customers or convert low-value customers into high-value ones.
Parting Thoughts
In a lot of cases, LTV represents predictions on future cash flow, which is crucial for long-term business planning and success. So there’s a lot of value not just in attracting new customers, but also for increasing LTV. What that often entails is upselling add-on products, incentive programs to encourage longer term commitments, and ultimately keeping customers happy.
