+3 votes

If you've been at more than one company or worked with more than one app... How do you approach picking your core retention metrics for your newest product?

Do you stay pretty standard or do you ALSO dive deep to try and find unusual moments that are less obvious? (Possibly with help of your data science teams) 

If you mainly rely on standard retention metrics, what are those?

If you largely rely on unique / unusual moments, how long do you go before trying to re-evaluate those as important North Stars?

by (260 points)
edited by

3 Answers

+3 votes
I typically find that this isnt an either-or question - and that you typically want to take both approaches, because these provide complementary insights.

With standard retention numbers(d1, d7 retention etc.), I typically try and find some comparables to understand if our retention metrics are healthy(or not) - and if not, what they should be.

With non-standard metrics, I prefer to use the phrasing: 'unusual moments that are unique to the app.' I prefer this because these moments arent non-obvious - they're pretty clear and intuitive to me when I've worked on the products. They just happen to be very product specific -> I've often found and used metrics such as moves made, puzzles solved or hangouts started.

Your standard metrics(Dn retention) generally convey the overall health of the product in terms of its propensity to retain users - and your non-standard metrics convey the activation rate of the product(how many users reach an a-ha moment).

In general, I'd recommend using both the above.
by (1.8k points)
Agree with this. I would however like to note there are different ways of calculating "standard" retention. I've seen daily, 24hr, and a few others. There have been some academic studies that show different types of retention calculations are more correlated to different LTVs.
0 votes

I agree with Shamanth -- these aren't mutually exclusive. Even though most savvy marketing / growth teams are optimizing around product specific "moments" and events in their products in the contemporary, algorithm-driven marketing paradigm, I doubt we'll ever see the end of the traditional retention profile, and I think DX retention metrics will perpetually be something that PMs and marketing teams track.

There's also a downside to focusing exclusively on "northstar" metrics like Facebook's (in)famous 7 friends threshold: it's easy to be so singularly obsessed with a metric that you end up optimizing to a local maximum and influence user behavior in a way that ultimately hurts the product. Like the article linked points out, correlation <> causation, and these northstar metrics can be interpreted that way.

Here's an example: let's say I have a social video messaging app with some existing users, and through my analytics I determine that the best users -- and I define "best" as those who within 7 days shared 100 or more videos -- all shared three or more videos on their first session in the app. I could interpret this in two ways:

1) Users who understand how to share a video are more likely to become engaged. If this is my thesis, then I'd focus my attention on the FTUE and in holding users' hands through the video creation process. The thesis might be that all users have the same enthusiasm for sharing videos but some users grasp the mechanics of it better than others, so if I can just be more explicit in instructing users how to share videos, I'll push those users onto the "highly engaged" track.

2) Users who have the most enthusiasm for sharing videos do so quickly. If this is my thesis, then I might focus on making some improvements to the funnel / tutorial, but what I'd really focus on is going after users who look like the ones in my app that are highly engaged, because user intent is more of a determinant of ultimate engagement than ability to understand the mechanics of sharing a video. My priority in this case would be to do better / more targeted marketing -- to find more users that match the deterministic profile of those that are highly engaged in my app.

With the first interpretation, I'd be dedicating my attention to the northstar metric -- user intent / attributes are irrelevant. With the second interpretation, I'd be dedicating my attention to targeted marketing -- user intent / attributed are the sole determinants of engagement and the best users will figure out how to use the app (put another way: if people need to be educated on the mechanics, they're not good users, and they won't engage like the ones that didn't need their hands held).

Obviously both of these interpretations is extreme, and an effective approach to optimizing engagement sits somewhere between them: acquiring "relevant" users but also ensuring that all users are onboarded properly. To execute against that, you'd want to track the range of engagement metrics.

by (9.8k points)
0 votes

I agree with both Eric and Shamanth

Using only one method is absolutely wrong. Trying to articulate your question, I think the real question here is whether retention can be measured detached from engagement metrics (and not UA). The answer is an undefinable 'NO'. Retention is not a metric that unveils the quality of UA but the quality of your product, first and foremost.

D1, D3, D7, D14, and D30 are the basics. The non-standard is all about the pitfalls - those moments when users who engage with the app, decide to stop engaging. It can be an expensive IAP, bug, or a boring level. Finding those pitfalls is crucial. 

 

by (660 points)
edited by