# Case study: How to optimize the cost of A/B testing in Facebook

Lets start simple. Looks like the algorithm is run targeted advertising in General. Let us give a rough example. And so the campaign begins to slowly unwind.

But at some point, you find that when the intended target CPC or CPA is quite expensive, and you are not willing to spend their money at this price. What happened. Often advertisers, especially beginners, are faced with this problem related to the lack of A/B testing. If you take a neutral picture and show it to the whole of Central Asia, it is likely that the results will leave much to be desired.

The most obvious hypothesis is that come to mind that I want to test A/B test is split by gender. Here is an example. If women in the target audience the online store to show in advertising dresses and skirts, and for men sneakers, shorts and t-shirts, the advertising results will improve.

We highlight the key basic principles of A/B testing (which are suitable for any similar promotion channels). Each audience has its own sub-campaign with the same creatives — ideally the campaign should be divided by gender, age, devices, and geography. In every class is to calculate a separate budget* maximum rate** (in English maximal bid), as it has its own set of indicators, often homogeneous, respectively, and the result of the campaign will be different.

* Budget on the audience depends on its scope, CTR and CR, and the expected cost per unit of result. For example, if the audience consists of 10 thousand people, the expected (predicted on the experience of other campaigns and other insights) CTR = 2%, CR = 20%, and the expected price per unit of the result is not more than \$5, the budget should be about 10 000 * 2% * 20% * \$5 = \$200. ** The maximum rate is the maximum value of the money that you are willing to pay for the advertising unit or the unit of the result on the auction site.

It depends on the type of campaign and its objectives. For example, to a wider audience is to set the maximum rate for the price per click equal to the value of your KPI relatively cliques; while the narrow audience is usually, experts recommend to take it higher than your KPI (despite the risk of penetrate increases your chances to show your ads among all competitive announcements for every person in this audience, and hence the probability of increasing the number of impressions that need bringing clicks and conversions). Each audience to show the same set of different creatives. Tens of thousands of publicly available case studies show that different audiences have different perceptions of the information and react to it.

This way you will be able to figure out which set and which combination will provide the best result. Conclusion. At the end of each test iteration can be complicated hypotheses and test them. But for the auction platforms like Facebook, there are limitations in scope and budget, about which we will describe below.

Although the basic principles of optimization is quite simple, they require a large amount of time from the Manager. The actions consist of the following items. You can use the automation steps 1 and 2 by implementing the partitioning of targeting in the form of podcampnyc.

How it works. As we wrote earlier, there is a limit on coverage. A minimum of ~1000 people*.

Each classroom should have a sufficient budget** to participate in the test, otherwise the corresponding adset not are created (the tool will generate an error). For Facebook, this minimum amount is approximately \$1, provided that the rate will be less than half of the budget. For example, if you have only \$100 to the campaign, not be divided at the same time on all 3 parameters (as will total more than 100 podcampnyc), try to divide into two parts on the floor.

* Generally, this minimum is limited to 20 people by Facebook and depends on rates and activity of the audience; in practice, the frequency in a small sample is increased for a short time, which negatively affects the effectiveness of the campaign. And in General if you have a low CTR results near zero. ** It is recommended to choose the amount for the budget of more than \$10 for the period up to 3-4 days.

Example — disable inefficient adsyou. We were approached by a very large portal of health services (several thousand items), more than 1,500 clinics in Moscow, own blog and online consultation. Task. 200 calls a day from Facebook at a cost of 500 rubles. When treatment was 11 calls a day for 1,100 rubles.

We recommend to do the automatic partitioning by sex and age, desktop and mobile. Each break had coverage of about 50,000 people. Find groups that have generated a value of up to 500 rubles per call.

In the end, only left those adsite that were within 500. The number of calls per day has increased to 50.