A slightly skewed look at segmentation.
Most times stats are shared they are 'whole' by which I mean that, for instance, when I elsewhere shared referring traffic it looked like:
* Organic Search (40% of traffic ... CR: 3.4%)
...Organic SMM (20% of traffic ... CR: 13%)
...APP (14% of traffic ... CR: 20%)
* Google: 55% of search . CR: 2.7%.
...Bing: 30% of search .... CR: 4.8%.
...Yahoo: 10% of search .. CR: 3.8%.
All the comparisons are of the average of each 'whole'; whole organic search, whole Google, etc. In part because it is standard usage; in part because that is the data generation level of the 'tool' used; and in very small part to share while obfuscating.
Recently, I made a decision for one last language translation for my sites. Software translation being NOT an option (insufficient for desired quality requirements) this is an intense 5-year commitment. An undertaking that on the face of 'normal' niche numbers might look foolish:
* English language niche market annual sales: 100 billion USD.
* New language niche market annual sales: 20 billion USD.
* English language market breakout: 80% commodity, 20% luxury sales.
* New language market: 20% commodity, 80% luxury sales.
Or high value, high margin, high ad/aff market cap:
* English language: 20 billion USD.
* New language: 16 billion USD.
With this data 'twist' in mind back to traffic referrers, conversion stats et al.
Oh, one critical point that needs be made about averages: in 2018 the average American household consisted of 2.53 persons. An average is a statistical artifact, it is useful but NOT 'real'. The larger (quantity) the segment the less 'truth' (quality) in the information and so the greater the probability of error in it's use.
Note: the current range is from 1-person to 19 (Duggar, Bates families) persons; quite an upside to 2.53.
Some analytics tools allow one to, for instance, further split each segment into three, i.e. 'average' plus both 'above average' and 'below average'. In too many cases the tool is simply playing with numbers. If it can not differentiate from where the differences arise... If it can, it may show certain high value Google traffic does convert on par with, in above example, with Bing average and provide insight into acquiring more.
As traffic increases it can be worthwhile splitting into quintiles (fifths), i.e. poor, fair, average/good, better, best. However, one must be able to support some idea of what brought each, where they went and what they did on site, whether they return, etc.
Note: few webdevs track and analyse click tracks. Etc.
Note: click tracks, as hurricane tracks, tell actionable stories. Etc.
Another use of segmentation is in selling direct ad space. Many/most sites sell ad space at one rate or perhaps one rate for a particular size ad or at most varied by site category, i.e. current events, lifestyle.
I work quite differently :) iamlost after all!
I compare my site content against the advertiser and the brand(s) and service(s) they want to highlight; then split the site into 3-main value page segments (A, B, C) and each of those into three (A1, A2, A3; B1, B2, B3; C1, C2, C3) such that I create a site value map of available (not spoken for by other advertisers) that they can mix and match as desired.
Note: I have a wait queue for all 'A', many 'B', some 'C' pages.
Note: what is 'A' for one advertiser may be 'B' or even 'C' for another. And vice versa. An intriguing game of matchup.
Note: of course one has to be able to make the case for the different valuations...
Don't just work to 'whole' averages if your traffic and/or ad/aff demand is decent as it misses too much information, too many growth possibilities, and far too much revenue.