Visualizing the importance of retention in Freemium


A freemium product’s retention curve is generally accepted to assume the shape of a negative exponential function, as below:


A retention curve might take this form if retention in the earliest stages of product use was determined solely by quality, but it is not. Rather, early retention — especially Day 1 retention — is affected by a number of other factors, mostly related to marketing.

  • Marketing channel mix can impact early stage retention substantially. When a product is being marketed across channels that are not optimally descriptive or targeted (such as incentivized mobile installs), Day 1 retention may be reduced.
  • Demographic targeting, when not effectively implemented, can drag Day One retention down by exposing the product to users for whom it is not at all appropriate. Generally, broad, high-volume marketing campaigns are associated with attendant drops in Day 1 retention.

In other words, any freemium product that is being actively marketed — and all freemium products should aspire to be in a position where active, ongoing marketing can be conducted profitably — will exhibit lower Day 1 retention than would be predicted by an exponential function.

Because of this, when a freemium product’s retention metrics are calculated on a cohort basis (as described here), its Day 1 retention value will reflect a large swath of the original cohort churning out of the product given a mismatch in functionality and need. With that portion of the cohort gone, retention metrics from Day 2 and beyond assume the negative exponential function, producing a curve similar to below:


The large reduction in cohort size in the earliest portion of the retention curve has important implications on total, aggregate revenue. Users can only contribute money to a product while they’re still users: as the graph above depicts, fewer users exist from a given cohort on each successive day since adoption. Thus the greatest potential for monetization exists in the earliest portion of a cohort’s tenure, when the most users are active.

But monetizing a user early in their tenure with a product runs counter to the purpose of the freemium model, which is to allow for higher levels of engagement (and thus monetization) than would be possible with paid product access.

Users generally don’t become highly engaged with a product on the first day they interact with it; they need time to realize the relevance of the product’s use case to their needs, become acquainted with the product, and carve out a space for that product in their lives.

Products that aggressively attempt to monetize users in their first day of use run the risk of alienating the users that would have become the most engaged (and, as a result, contributed the most money over their lifetimes). A balance must be struck: more users from a cohort exist in a product early in that cohort’s tenure than later, but a product that attempts to extract money from users before proving its use case may cause premature churn.

While the competing effects of shifting the monetization schedule within a product forward are difficult to gauge without rigorous testing, shifting the retention curve up doesn’t present any potential for accelerated churn or user disaffection: when retention increases, so does monetization.

Retention metrics are eminently more important to optimize for in product iterations and updates than monetization. For one, the effects of changes to monetization mechanics are very difficult to isolate and very frequently persist over an extended period of time, which can muddy their interpretation.

Secondly, product iterations focused on increasing retention, by definition, attempt to increase the number of users present in a product in the future. Ceteris paribus, a user base that has been retained to a greater degree through product improvements focused on retention has higher monetization potential than one which experiences unabated churn. Users that have been retained can always be exposed to improved monetization mechanics later.

Acknowledging the realities of the retention curve discussed previously, there is no practical difference in net revenue on equivalent improvements to retention or monetization. Consider a user segment of a product with Day 1 retention of 50%, ARPDAU of $0.05, and a retention profile matching that illustrated above: an initial cohort of 10,000 users will produce total aggregate revenue of $24,731, as shown below (the spreadsheet used to create this graph can be downloaded here):


Both an increase in ARPDAU of 20% to $0.06 or an increase in Day 1 retention to 60% result in an approximate increase of 20% to total cumulative revenue (the increase to retention actually has a slightly lower effect in this model because ARPDAU is applied to the first day of the cohort).

But a 20% increase in ARPDAU would be much more difficult to achieve — and prone to downside risk — than a 20% increase in retention. The model linked above assumes that retention stays constant despite changes in ARPDAU, which isn’t realistic: more likely, retention would decrease as the product more aggressively or more stringently compelled users to monetize.

The same can’t be said of retention improvements, which often come in the form of a clearer on-boarding process, a more intuitive user interface, better graphics, lower loading times, and any number of fundamental improvements that increase the enjoyability of a product. These are truly objective improvements; changes that make the product more entertaining, efficient, or effective.

The knock-on effects inherent in changes to monetization aren’t present in retention considerations; retention improvements provide an opportunity to singularly and explicitly improve the product experience without worrying about latent, detrimental consequences that will only surface late in a cohort’s tenure.