Modeling advertising campaign parameters to improve cost-effectiveness and reach with a fixed budget

When using Yandex Direct advertising, you must always set a budget for each individual campaign, as well as an average daily budget. Failure to do so can result in uncontrolled spending and large amounts of money being written off..

The second fundamental problem with advertising is the frequent occurrence of situations where advertising campaigns are enabled, funds are available, and all parameters are set correctly, but ads are not showing well. This means the advertiser is constantly balancing between maintaining their budget and not seeing any impressions..

The algorithm used to train advertising campaigns is completely hidden from the user and does not allow for a proper understanding of all the patterns without the use of modeling and the study of statistics obtained as a result of long-term advertising use..

Below, we demonstrate how linear regression analysis can be used to identify factors that increase and decrease click-through rates with a fixed budget, as well as how to increase impressions and identify factors that determine this metric, again with a fixed budget..

To explain the impact of company quantitative parameters on the resulting advertising audience reach and, consequently, the number of clicks, the same linear regression model was used as in the previous section, the coefficients of which are presented in Figure 1.

A quantitative variable corresponding to the number of clicks was used as the dependent variable. First, a correlation analysis was conducted between all quantitative independent variables to eliminate correlated regressors from the equation and double-count the same factors.

As a result, the following quantitative parameters were retained for the regression analysis: average cost per click, average ad impression position, impressions, cost, and bounce rate.

The R² level of explained variance for the resulting model is 0.69, which is high enough to be considered adequate. Two parameters are particularly important for identifying patterns: average cost per click and cost per click.

The resulting model shows that increasing spending by 1 ruble increases the number of clicks by 28. Decreasing the average cost per click by 1 ruble increases the number of clicks and, consequently, ad reach by 49, all other parameters held constant. This confirms the validity of the algorithm used by the company to display ads to as many users as possible at the lowest possible rate within a fixed budget.

If an advertiser receives recommendations to increase their bid per click to improve their ad's position relative to competitors, which should logically lead to an increase in clicks, then the following should be considered. Since the budget is fixed, the model allows us to understand that a bid per click of 50 rubles and a budget of 1,000 rubles will result in fewer clicks than a bid of 10 rubles and the same budget of 1,000 rubles. (See Figs. 8, 9)

Figure 8: Linear regression. Number of clicks with an average cost per click of 10 rubles and a fixed budget

Figure 9: Linear regression. Number of clicks with an average cost per click of 50 rubles and a fixed budget

Also, changing any other parameters that turned out to be significant and were included in the model, except for consumption, does not lead to a significant increase in the number of clicks..

Therefore, for a company operating throughout Russia, where there are many regions with attractive rates, it is recommended to adhere to a strategy focused on lowering rates for all types of advertising campaigns..

But as mentioned above, lowering bids and fixing the budget can lead to the system refusing to show ads. To understand which factors influence the number of impressions, we'll build a separate regression model..

Table 3: Linear Regression. Number of Impressions

The resulting model clearly demonstrates (Table 3) that, with a fixed budget, increasing the bid per click will lead to a decrease in the number of impressions. But more importantly, the specific pages shown in the ads have a much greater impact on the number of impressions than simply increasing advertising costs. Therefore, by choosing the right pages, you can increase the number of impressions without increasing your budget or increasing your bid per click. Without using modeling, the ad user typically tries to solve this problem by again increasing the bid per click, without changing the budget or changing their approach to selecting ad pages.