←back to thread

Vijaye Raji to become CTO of Applications with acquisition of Statsig

(openai.com)
176 points tosh | 3 comments | 02 Sep 25 18:18 UTC | HN request time: 0.621s | source
Show context
iamleppert ◴[03 Sep 25 14:14 UTC] No.45116064[source]▶
The problem with A/B testing, and anyone who has ever done it at scale can tell you, beyond the banal basic stuff that people are already aware of like making things accessible and discoverable, once you get to a certain point, people just have no opinion. The default opinion is no opinion.

It's why every mass consumer product devolves into a feed or a list of content delivered by an algorithm. Once you reach a certain point, you come full circle and even that doesn't matter anymore: users will happily consume whatever you give them, within reason.

A/B testing platforms are mostly used by an odd collection of marketers and "data driven" people who love to run experiments and drag out every little change in the name of optimization. In the end, it all completely doesn't matter and doesn't tell you anything more than just talking to an average user will.

But, boy, are they sure a great way to look busy and dress up an underperforming product!

replies(1): >>45116608 #
1. beeon ◴[03 Sep 25 14:59 UTC] No.45116608[source]▶
Maybe the industry you work in is relevant here. In e-commerce, A/B testing the position and color of "add to cart" button can yield legitimate revenue multipliers. People's opinions are irrelevant in that kind of A/B test, all that matters is the likelihood they continue down the funnel.
replies(1): >>45119099 #
2. iamleppert ◴[03 Sep 25 18:41 UTC] No.45119099[source]▶
This is always the straw man A/B testing people reach for. When in reality, this only works on the worst possible designs or bizarre layouts, that no one really uses. It's a myth that you can somehow ring out extra few % by making tiny changes to fonts and colors -- in every case, they simply stop the experiment when the results tip in their favor.

The ONLY time I've ever seen it used successfully was to make changes to the layout of ads, making them look like more organic content or likely to be accidentally clicked. You can see this at work if you've ever clicked on an ad by mistake, or maybe you were trying to close the ad and noticed someone has "optimized" the close button placement using one of these A/B tools.

There it is, that's the market for these tools. That represents the vast majority of the these companies use-cases, revenue and usage. Anyone else who implies an innocent intent is either ignorant or inexperienced.

replies(1): >>45124157 #
3. tibbar ◴[04 Sep 25 06:20 UTC] No.45124157[source]▶
Experimentation is a toolkit for optimizing things. That's it. For example, if you are Uber, you might offer a bonus to people to work more. How much money do you offer? Is there a relationship between how much you pay and how well people perform? To determine this, you need empirical data, you need to be able to split people into groups, and you need to be able to measure things about the groups. And you need some guardrails to keep yourself from p-hacking. As the number of people and the amount of money increases, the number of interesting questions rises rapidly. These are all things you get from an experimentation platform.

Experimentation works best when a) it's easy to try different parameters for a thing, and b) there is a lot of money involved.

Your comment is weird to me because you seem to be implying that experimentation is a cult because there are no things worth optimizing, and/or optimizing things has a negative societal effect. Those things are obviously true some of the time, but it's not very interesting.