As you might know, Return Path acquired OtherInbox some months ago. One result out of this merger is… a new dataset for us, 🙂 which contains some really interesting international email performance measures. It has been published yesterday. Return Path said, they
“analyzed over 3 million email campaigns sent to users of OtherInbox and calculated an engagement score for the most popular TLDs”.
I just had some quick fun playing around with it. Let’s have a look.
The domain’s engagement scores (table) are mainly built upon
- the “read rate”, which is directly measured in the user interface and therefore more accurate than traditional open rates, and
- the email “deleted without read rate” (dito),
- the “ISP spam rate”, and
- the user “user marked as spam rate”.
You can approximately recreate all the scores by taking the “read rate” times 8.228, then subtracting the “deleted without read rate” with a weight of 0.438, subtracting the ISP spam rate times 0.984, and finally adding +2.117. For instance, to calculate the engagement benchmark for .ca and .edu (22% read rate, 10% deletion rate, 23% spam rate), you would do the math like this: 8.228*0.22-0.438*0.1-0.984*0.23+2.117.
If you were looking for a score, which discriminates the top-level domains better (using a different weighting scheme), have e.g. a look at the following table. I stretched the index a little bit more:
tld scoring .ca 3.53 .edu 3.23 .tv 2.45 .org 2.28 .uk 2.12 .us 2.10 .de 1.85 .es 1.43 .biz 1.40 .net 1.34 .com 1.15 .br 0.81 .fr 0.41
Canada rocks, mailers sending from .ca addresses enjoy highest engagement rates. France… well… But there’s more than meets the eye.
Clustering the data
Some more interesting findings can be spotted, if we cluster the dataset. So let’s just slice out two segments. This might present us with the following picture:
tld read del ISP User cluster_0 .ca 22,00% 10,00% 23,00% 0,19% cluster_0 .edu 22,00% 10,00% 23,00% 0,38% cluster_0 .us 15,00% 9,00% 38,00% 0,30% cluster_0 .org 15,00% 9,00% 38,00% 0,18% cluster_0 .tv 15,00% 9,00% 27,00% 0,15% cluster_0 .de 13,00% 9,00% 46,00% 0,20% cluster_1 .uk 14,00% 6,00% 45,00% 0,14% cluster_1 .es 11,00% 7,00% 49,00% 0,25% cluster_1 .net 10,00% 6,00% 56,00% 0,16% cluster_1 .biz 9,00% 9,00% 40,00% 0,13% cluster_1 .com 8,00% 6,00% 52,00% 0,11% cluster_1 .fr 6,00% 5,00% 67,00% 0,28% cluster_1 .br 6,00% 7,00% 40,00% 0,10%
The centroid plot in the figure above visualizes each cluster’s characteristics. Blue represents cluster 0, red is the other cluster 1. The x-axis contains the four attributes (read rate, deletion rate, spam rate, complaint rate). The vertical y-axis holds the (normalized) observations from Return Path/OtherInbox. Finally, the table below the plot lists the data appended by the corresponding cluster for each row.
What can we conclude?
- The blue cluster_0 holds top-level domains with a.) rather high positive and negative engagement (reads, deletions, user complaints), and at the same time b.) very little ISP spam rates.
- The red cluster_1 is – no wonder – exactly the opposite: a.) low engagement and a high ISP spam rate.
Blue versus red, active subscribers versus rather inactive ones, high inbox placement rates versus spamfilter bugged top-level domains etc. Note that it’s not necessarily good versus bad, as the colors perhaps might suggest. Because otherwise you would probably want to exchange the colors of .uk and .de for example. In addition, those are only tendencies; the clusters aren’t that homogenous, if you look closely. But anyway, quite an interesting view, isn’t it?