Lifetime Value (LTV)
Lifetime Value (LTV) is one of the most important and popular terms in marketing. It is defined as the total value or profit that a user will generate over his or her lifetime, and therefore, is also interpreted as the upper limit on money that marketing will spend to acquire that user.
The main advantage of the concept of LTV is that it allows marketers to shift the focus from minimizing cost to maximizing value of the acquired user. LTV as a principal metric makes sense because the marketing team can focus on one metric that is function of many other important KPIs such as retention, monetization, and virality.
Drivers of LTV
Despite its increasing popularity, there is no standard formula to calculate LTV – every business calculates LTV differently. However, the main drivers for LTV fall under one of three categories: retention, monetization, and sometimes, virality. Quite often, developers choose their own definition of LTV as long as it makes sense depending on the app’s business.
For example, an app that monetizes through video ads may choose the total playtime of qualified ad views as a measure of LTV because the users who watch more ads are more valuable. Similarly, an app that is designed only for sharing content may choose number of shares as a measure of user LTV because the user who shares more is worth more for the app.
A simple google search for “calculating lifetime value” yields more than 500K results; from simple averages to complex predictive models, from using spreadsheets to programmatic calculations. LTV can be modelled by numerous methods making it quite confusing for developers to choose what’s appropriate for their product.
The LTV model you choose should be a function for what you’re trying to solve at a given point in a product’s lifecycle.
For instance, when first launching a product, it might be enough to have directional indicators that can inspire reasonable confidence as to how initial marketing spend should be allocated. This initial model will be simple, but fast to build.
Over time, as your product grows to have millions of users, your needs will change from basic understanding of positive or negative return on spend, to highly accurate optimization.
Below are common examples of LTV models deployed at different stages of a product’s lifecycle.
The simplest method to estimate LTV is to multiply Average Daily ARPU and Average number of days a user spends in app before leaving (or within the selected duration)
LTV = Average ARPDAU x Average number of days a user is active in a certain period.
Note that the LTV calculated by this approach is not very precise and is only used to get rough estimates. There are two main drawbacks of this approach: it assumes all users are the same and it does not account for seasonality, product updates, or promotions.
Moving Average and Cohorts
We can build on the Simple Average model by introducing daily cohorts and moving averages, which will measure LTV for each daily cohort.
Because moving averages take seasonal patterns into account, this method of LTV calculation is inherently more accurate. However, this introduces a challenge: what is the appropriate time for your moving average? Too long of a time duration and you overlook seasonalities and product updates, but too short of a time duration and results might be too volatile to interpret and take action.
Regressions and Curve-fitting
To achieve better accuracy, curve fitting is also a popular method. During the 2013 Slush conference, Eric Seufert, VP of User Acquisition at Rovio, described two approaches that he commonly uses to calculate LTV:
- The first, Retention Approach, uses the power curve to predict lifetime from real retention data, and then multiplies the lifetime by trailing x-days ARPDAU to estimate LTV.
- The second, Monetization Approach, predicts the LTV of users from historical data using logarithmic method.
Regression analysis, in general, is a very popular approach to fit the curve. Moreover, Polynomial regression is especially useful if the data is nonlinear and can therefore, yield better predictions.
Typically only deployed in large organizations, the most accurate methods for predicting LTV will be driven by Probabilistic Modeling, which will not coincidentally, require significant expertise and resources.
While there are many modeling techniques, the two most common are BG/NBD and Pareto/NBD methods. These methods require deep statistical knowledge and will be covered in separate post.
Professor Bruce Hardie, Professor of Marketing at London Business School, has explained these techniques in details in his various lectures and papers. For more information on using BG/NBD and Pareto/NBD, please see his Tutorial at the AMA’s Advanced Research Techniques Forum, San Diego, CA, June 2015
Need for a flexible solution: LTV and Omniata
Driven by product lifecycles and the relative need at each stage, the methods and models used to predict LTV need to be flexible and able to evolve.
At Omniata, we have designed our system to support multiple LTV calculations. Omniata allows developers to define, edit, or update formula for any user properties in real time. Using our flexible data modeling support, you can define multiple formulas for LTV as user properties and then use it to slice/dice data to get more insights, or set up automated campaigns to target users based on their predicted LTV.
Show Comments (0)