Retention Primer

There is a golden standard for what developers want their app or game to reach. It is for people to not be able to put down their product. Over the years there have been many such games and apps; some of them stood the test of time and are still thriving while others have filled their life cycle and are slowly forgotten. I will go over a key concept that measures and aims to understand this phenomenon: engagement.

Wait, the title of the post is Retention Primer, then why is engagement the key concept? This is because retention is engagement over a period of time. Engagement is getting someone to play, while retention is getting someone to play again and again. This is a key connection that a lot of people miss. The product and the core loop should be the main driver; repeated visits can only come after that.

Now that the engagement-retention connection is established, we can think about different types of retention. Retention can be checked in two different areas: short vs. long-term and on-day vs. period-over-period retention. They form a diagram like the following:

The short vs. long-term axis is easier to grasp; how long is the retention period we are looking at? If we are looking at daily or weekly retention, we are mostly interested in the short term; for example for a limited-time event that will last for a month. On the other hand, we may seek to keep people playing for months and years on end, and would look at long-term retention in this case.

The trickier section is the on-day vs. period-over-period retention. Giving the definitions one by one is the first step:

Period-over-period retention: this is the percentage of people that are active in one period, and then active again in the following period. This can be days, weeks, months or even years. If 100 players were playing in December 2022 and 40 of them play as well in January 2022, then our Month-over-month retention is 40%.

On-day retention: this is the percentage of people that are engaging a certain number of days after they sign-up. For example, if I sign up on Thursday, 22nd of December, 2022, and then log in to play on the 29th of December (instead of socializing with friends and family) then I am retained on Day 7. If 100 people signed up on the day that I did, and 30 of them were active on day 7, our Day-7 retention is 30%.

These two are calculated differently and are showing two different types of retention. The best way to consolidate and compare the two is visual.

Notice that the month-over-month retention of the different groups (new, existing, returning users) are different from the total, but that is a more advanced topic for another time).

Now we focus on new users only. The people who signed up or started using our app in December. Even more specifically, the people that signed up on the 1st of December.

From the example above, our On-day retention numbers are:

• Day-7: 30%

• Day-14: 18%

• Day-30: 10%

But isn’t a month equal to 30 days on average? How can Month-over-month retention be so high but Day-30 be so low?

The key is the population we base the calculation on. When checking period-over-period, we do not care when the exact date of activity or signup is. I can sign up on December 27th, and be active on January 5th and I would be retained month-over-month, although it has only been 9 days! This fuzziness is completely ok to have as long as we understand that period-over-period retention is closer to an approximation than an exact number. In fact, it is balanced out by the other end of the spectrum; people who have more than 30 days between their activity but are considered retained over only a single month.

On the other hand, on-day retention is very precise, and usually based on single days, and averaged where needed. If I signed up today, I am interested in activity exactly 7, 14, 30 etc. days later, not an arbitrary time window. To aggregate this number for an entire audience, we average it across days:

We ended up with two values for D30 retention and month-over-month retention. Putting these values into sentences as one final comparison:

• Out of all the active users this month, 40% of them will still be active next month.
• When a new user joins, there is a 10% chance that they will still be playing on their 30th day.

There are caveats to these sentences of course, but I hope this write-up gave a good primer on the topic of retention and how it is measured. In the next post, I will go over how to think about and analyse retention, and how to increase and optimize it. Retention is an important concept that likely will come up across many posts, so the fundamental understanding is going to be very useful.