A easy and highly effective method to segmenting your product options into Core, Energy, and Informal.
![Towards Data Science](https://miro.medium.com/v2/resize:fill:48:48/1*CJe3891yB1A1mzMdqemkdg.jpeg)
Within the earlier submit, I confirmed you a simple manner the way to measure product options retention.
After making use of the evaluation we acquired the desk with retention per characteristic like this (sorted by reducing [Average % returned users]):
On one hand, we acquired useful details about which product options affect product retention probably the most.However, we don’t have at hand figures about what number of customers used these options so we will’t be assured that these figures are dependable.
Let’s add [# users] and take into account this desk yet another time.
Now we will simply spot a difficulty: for instance, the primary two product options (feature27, feature34) with the best [% returned users] have fairly a modest quantity when it comes to [# users].
Truly, this downside is commonest in a whole lot of analyses that I’ve seen. Usually analyst brings a fairly fascinating high quality measure however it isn’t backed up by amount measure. In consequence, a few of our choices could be at the very least suboptimal and at most simply fallacious.
So how can we repair this concern?
Let’s mix each metrics (qualitative and quantitative) into one chart. Essentially the most appropriate strategy to do it is a scatter plot:
let’s placed on the X axis the metric [% users], it’s our amount metric that measures the recognition of a product characteristic.let’s placed on the Y axis the metric [% returned users], it is our high quality metric that measures the worth of a product characteristic.
The ensuing chart might seem like this:
Plainly thus far it’s reasonably arduous to make any significant conclusions from the chart.
What can we do to enhance the chart readability?
Let’s apply the 50/80 percentile rule from the earlier submit.
Truly, after making use of 2 thresholds for [% users] and [% returned users] we are going to get 9 clusters.
Clustered product options scatter plot might seem like this:
By including percentile thresholds to the chart we will now distinguish such product characteristic clusters:
Core: [% users] > 80 pctl, [% returned users] > 80 pctlPower1: [% users] > 80 pctl, [% returned users] in [50, 80] pctlPower2: [% users] in [50, 80] pctl, [% returned users] > 80 pctlCasual1: [% users] in [50, 80] pctl, [% returned users] in [50, 80] pctlCasual2: [% users] in [50, 80] pctl, [% returned users] < 50 pctlCasual3: [% users] < 50 pctl, [% returned users] in [50, 80] pctlSet-up: [% users] > 80 pctl, [% returned users] < 50 pctlNiche: [% users] < 50 pctl, [% returned users] > 80 pctlOthers: [% users] < 50 pctl, [% returned users] < 50 pctl
Let’s focus on a bit of bit every of clusters.
Core options are the true core of your product. These options are utilized by a whole lot of customers, and what’s extra vital customers return again to proceed utilizing these options. As a rule, there may very well be a really small variety of such options (2–3 options).
Energy options are workhorses of your product. These options mixed with core options ship about 80% of the common worth that your product creates. Among the energy options (Power1) are as common as core options however convey much less worth to customers. Different energy options (Power2) convey as a lot worth as core options however are much less common. As a rule, there may very well be 3–5 options in every energy cluster.
Informal options are options which are used every so often. In addition they convey some worth to customers however for probably the most half, they’re supporting options.
Set-up options are a novel subset of options which are designed to arrange a product for the next handy utilization. A whole lot of customers use them, however as regular, it occurs one time, on the onboarding section.
Area of interest options are a really particular subset of options that would convey an unlimited quantity of worth however this worth is perceived by a restricted variety of customers.
Now we’re prepared to match the outcomes of this balanced method to the outcomes from the earlier submit:
As we will see on the high of the record there are some Area of interest options.
For certain we will attempt to enhance their adoption and transfer them from Area of interest to Power2 and even Core cluster. For a few of them, it’s doable, for others — it’s not. However the primary level right here is to not merely assume that any characteristic with excessive retention is a core characteristic.
Additionally, please notice that some options can transfer from cluster to cluster over time. There may very well be completely different causes for this: new consumer acquisition efforts, UX adjustments in options, consumer base maturing, and many others.
Lastly, let’s group options into clusters and calculate cluster centroids:
There are a number of vital insights right here:
Core + Energy clusters account for less than ~20% of all product options.Others cluster accounts for 27% of all options and on the identical time, it serves solely 8.7% of customers.Area of interest options are utilized by simply 11.3% of customers and on the identical time have the best retention (even larger than for the Core cluster).
Within the subsequent submit, I’ll speak about one other perspective on the characteristic retention definition.
P.S. There’s a higher strategy to cluster product options based mostly on the MCC coefficient.