Maybe someday I'll get around to expressing this idea in a more refined blog post, but here is the rough thinking for now...
Irrespective of whether one approaches UX using intuition/experience, or statistical methods, both tend to incur a flawed assumption: There is a single 'best' version of a product/app/website, in terms of the overall UX decision.
Consider AB testing, where x measures some objective result:
f(A) = x
f(B) = 3x
So the results show f(B) > f(A). As an example, in context this could mean site version B, retains visitors longer and results in higher conversions. Therefore, UX version B 'wins' right?
What's the problem here? Perhaps it should rather look like this:
f(A,B,C) = [Ax,By,Cz]
Forgive me if the notation is a bit unusual, but the idea is you end up with 3 different UX's. 3 'best' versions of the website, that are always live, simultaneously. Or if you had a bimodal preference group, you would end up designing 2 UX trees. The number of clusters or 'groups' does not matter, only the fact that clusters exist to begin with.
We need to move away from the idea there is a single 'best' version of an interface. More often than not preferences when batched together (in the form of a product, app or whatever), do not form a neat normal distribution, even if approximating that is convenient.
A/B testing so highly regarded because it works. As effective as it is, (if used right), I believe we can do better.
Malcolm Gladwell eloquently expands on this concept, using an example from the spaghetti sauce industry, with further detail. I strongly encourage you to check it out, if you haven't yet, it's awesome![1]
This same idea of clusters applies for taste in movies,books,foods, which faces we find beautiful and almost anything really when looking at a population. Taste in website UX preferences is no different.
So the optimal webapp of 2020, is one that automatically knows which clustering a visitor falls into and presents a version of the site that makes them most comfortable, based upon UX decisions that align well to his/her group.
The current approach in software tries to shoot for a middle ground compromise, and in addition using a settings or preferences panel that the user needs to tweak. This could be considered suboptimal as more effort is required from users.
Example: There is a reason that UI animation annoys me (and others), to the point I'm willing spend hours researching how to hack the O/S, in order to turn it off. Clearly however, there exists other groups for whom it looks good, and there is no issue.
Why these clusters/groupings exist at all could be a combination of differences in our neurocognitive functions, and the summation of experiences over our lifespans.
Excellent article Matt! I had lots of conversations around this at SMX earlier this year!
The challenge is finding a way to clearly segment your user groups that doesn't have the same flaw of grouping based on majority otherwise the benefits will just be averaged out.
For example, if I can figure out that the chance of someone belonging to user group A when arriving from source X is 70% then I'm still serving a potentially unoptimized UX to 30% of the users.
Obviously this isn't a problem for an app that can clearly segment its users!
Interesting. Now that I think about it a tldr could be "Use segmentation for UX" ;)
(segmentation = groups = clustering)
I was thinking a bit further on this last night, and thought why not use evolutionary methods as well?
In other words we don't design the UX, we let it emerge and cluster around segments naturally. Kind of like if you took the concepts from genetic programming and mixed it with UX.
Irrespective of whether one approaches UX using intuition/experience, or statistical methods, both tend to incur a flawed assumption: There is a single 'best' version of a product/app/website, in terms of the overall UX decision.
Consider AB testing, where x measures some objective result:
f(A) = x
f(B) = 3x
So the results show f(B) > f(A). As an example, in context this could mean site version B, retains visitors longer and results in higher conversions. Therefore, UX version B 'wins' right?
What's the problem here? Perhaps it should rather look like this:
f(A,B,C) = [Ax,By,Cz]
Forgive me if the notation is a bit unusual, but the idea is you end up with 3 different UX's. 3 'best' versions of the website, that are always live, simultaneously. Or if you had a bimodal preference group, you would end up designing 2 UX trees. The number of clusters or 'groups' does not matter, only the fact that clusters exist to begin with.
We need to move away from the idea there is a single 'best' version of an interface. More often than not preferences when batched together (in the form of a product, app or whatever), do not form a neat normal distribution, even if approximating that is convenient.
A/B testing so highly regarded because it works. As effective as it is, (if used right), I believe we can do better.
Malcolm Gladwell eloquently expands on this concept, using an example from the spaghetti sauce industry, with further detail. I strongly encourage you to check it out, if you haven't yet, it's awesome![1]
This same idea of clusters applies for taste in movies,books,foods, which faces we find beautiful and almost anything really when looking at a population. Taste in website UX preferences is no different.
So the optimal webapp of 2020, is one that automatically knows which clustering a visitor falls into and presents a version of the site that makes them most comfortable, based upon UX decisions that align well to his/her group.
The current approach in software tries to shoot for a middle ground compromise, and in addition using a settings or preferences panel that the user needs to tweak. This could be considered suboptimal as more effort is required from users.
Example: There is a reason that UI animation annoys me (and others), to the point I'm willing spend hours researching how to hack the O/S, in order to turn it off. Clearly however, there exists other groups for whom it looks good, and there is no issue.
Why these clusters/groupings exist at all could be a combination of differences in our neurocognitive functions, and the summation of experiences over our lifespans.
[1] http://www.ted.com/talks/malcolm_gladwell_on_spaghetti_sauce...