Something Better A / B Testing
Hello! My name is Vasiliy. My main occupation is to increase the conversion of sites, the key metric in any IT business.
According to statistics, the average conversion of sites in RuNet does not exceed 1 percent, and this means that a huge number of Internet entrepreneurs lose a lot of money every day. These entrepreneurs can be mentally divided into two main types: the first are those who just sit and look at their un-optimized resource and do nothing. Unfortunately, they are doomed to be crushed by time by innovations and competitors. The latter are those who are trying to do something about it. As a rule, for this they draw landings, experiment with advertising channels and conduct A / B testing. A / B testing is an extremely necessary tool for modern marketing, it allows you to determine which version of the selling page is the most effective for most users. To understand how effective this method is, it’s enough to remember the sensational story about Obama and $ 60 million . However, this method has a downside, it lies in the fact that all people are different, and if for example statistics show that women mostly love cats , it’s not at all necessary that your girlfriend does not have allergies to them. That is, if our A / B test showed that a certain landing version is ideal for 60% of users and we choose it for our advertising company, then we automatically lose the remaining 40%, which it does not catch. So, A / B testing is not perfect and you need to urgently do something about it. And right now, because we are losing 40% of leads to all the hell and this is not cool at all!
In order not to lose customers, we need a system that does not just conduct an A / B test and post factum shows you the results, but is able to predict which version of the landing page your visitor will like and show this option to him. It is a pity that this is not possible, because no one until the end knows what your potential customers have in mind, right? Or not?
')
Although, of course, it can be assumed that a certain number of your visitors do not watch the presentation video during office hours, because they are fired by the authorities, and if your grandmother came to your site, then the selling text that was written for your son Immediately in the head climbs a huge amount of matches. But what if you find these matches automatically, if you create an algorithm that finds similarity in the behavior of different users on the site by different parameters, for example, by time, geolocation, browser, operating system, UTM tag from advertising companies? What do you think, how many matches would he find?
I will say right away that we did this thing and got an interesting result, it turned out that with a probability of about 80%, the algorithm can predict which landing variant will be preferred by a particular user. Yes, we do not know what your user ate for breakfast, also, we do not know what his dog or wife's name is. But even the available data was already enough to automatically look for matches. In order for our algorithm to work better, we attached to it the so-called weighting system. If the landing variant doesn’t like the users and they don’t perform targeted actions on it, it slowly goes down in rotation and as a result isn’t shown to users at all, or is shown only to the group that likes it only when they like it more. Again, there is a small problem, it lies in the fact that it takes some time and proper attendance to identify correspondences and the normal operation of the system. And the more options of landings in the rotation, the more visitors need. For each option, at least 500 is desirable. Unfortunately, if you have 200 people per day and 5 landing pages, you’ll rather lose your mind than wait for our algorithm to segment your users. However, if you have an attendance of 2000+ under equal conditions, he will learn in one and a half days. And if you have this traffic from an ad on which utm tags are set, then the algorithm will create magic for you. However, do not forget that even the magic gin does not fulfill the desire, if you do not rub its lamp, to put it simply, if you made 5 variants of creepy non-usable pages that differ from each other only in the number of unicorns in the footer, do not expect the algorithm to help you.
In general, the thing turned out to be interesting and it seems to me that this is similar to the future of technology for marketing.
I apologize in advance if I missed any errors in the text of the article, if you find it in a personal email.
PS I'm the only marketer with dysgraphia in this country.
According to statistics, the average conversion of sites in RuNet does not exceed 1 percent, and this means that a huge number of Internet entrepreneurs lose a lot of money every day. These entrepreneurs can be mentally divided into two main types: the first are those who just sit and look at their un-optimized resource and do nothing. Unfortunately, they are doomed to be crushed by time by innovations and competitors. The latter are those who are trying to do something about it. As a rule, for this they draw landings, experiment with advertising channels and conduct A / B testing. A / B testing is an extremely necessary tool for modern marketing, it allows you to determine which version of the selling page is the most effective for most users. To understand how effective this method is, it’s enough to remember the sensational story about Obama and $ 60 million . However, this method has a downside, it lies in the fact that all people are different, and if for example statistics show that women mostly love cats , it’s not at all necessary that your girlfriend does not have allergies to them. That is, if our A / B test showed that a certain landing version is ideal for 60% of users and we choose it for our advertising company, then we automatically lose the remaining 40%, which it does not catch. So, A / B testing is not perfect and you need to urgently do something about it. And right now, because we are losing 40% of leads to all the hell and this is not cool at all!
In order not to lose customers, we need a system that does not just conduct an A / B test and post factum shows you the results, but is able to predict which version of the landing page your visitor will like and show this option to him. It is a pity that this is not possible, because no one until the end knows what your potential customers have in mind, right? Or not?
')
Although, of course, it can be assumed that a certain number of your visitors do not watch the presentation video during office hours, because they are fired by the authorities, and if your grandmother came to your site, then the selling text that was written for your son Immediately in the head climbs a huge amount of matches. But what if you find these matches automatically, if you create an algorithm that finds similarity in the behavior of different users on the site by different parameters, for example, by time, geolocation, browser, operating system, UTM tag from advertising companies? What do you think, how many matches would he find?
I will say right away that we did this thing and got an interesting result, it turned out that with a probability of about 80%, the algorithm can predict which landing variant will be preferred by a particular user. Yes, we do not know what your user ate for breakfast, also, we do not know what his dog or wife's name is. But even the available data was already enough to automatically look for matches. In order for our algorithm to work better, we attached to it the so-called weighting system. If the landing variant doesn’t like the users and they don’t perform targeted actions on it, it slowly goes down in rotation and as a result isn’t shown to users at all, or is shown only to the group that likes it only when they like it more. Again, there is a small problem, it lies in the fact that it takes some time and proper attendance to identify correspondences and the normal operation of the system. And the more options of landings in the rotation, the more visitors need. For each option, at least 500 is desirable. Unfortunately, if you have 200 people per day and 5 landing pages, you’ll rather lose your mind than wait for our algorithm to segment your users. However, if you have an attendance of 2000+ under equal conditions, he will learn in one and a half days. And if you have this traffic from an ad on which utm tags are set, then the algorithm will create magic for you. However, do not forget that even the magic gin does not fulfill the desire, if you do not rub its lamp, to put it simply, if you made 5 variants of creepy non-usable pages that differ from each other only in the number of unicorns in the footer, do not expect the algorithm to help you.
In general, the thing turned out to be interesting and it seems to me that this is similar to the future of technology for marketing.
I apologize in advance if I missed any errors in the text of the article, if you find it in a personal email.
PS I'm the only marketer with dysgraphia in this country.
Source: https://habr.com/ru/post/241461/
All Articles