Evaluation of the effectiveness of reach campaigns.
Attribution of unattributed.
The text was brazenly stolen from Aleksander Klimov!
Marketing budgets keep leaking to the internet because of granular targeting and measurability. And if the effectiveness of performance channels (CPC, target, context) has already learned to measure through using the web and end-to-end analytics, then it is not so obvious with tracking the effectiveness of reach campaigns (YouTube, media, TV).
This article uses buzzwords. Using words like these can add up to 20% more value to your resume if you're lucky. If you're unlucky, then no one will understand you.
We will speak the same language if you read the introduction from the article on Unit Economy. In a nutshell, all marketing communications can be divided into performance and brand communications. Performance mechanics are advertising campaigns aimed directly at increasing sales. Usually, such communications contain an offer (discount, special conditions, an offer to buy a specific product). Brand communications help build brand awareness and image, increase conversion performance marketing and lead to sales through direct visits. However, it is difficult to accurately assess the contribution of such communications to sales. The optimal proportion of the budget depends on the size of the company and the length of the selection and purchase cycle.
Life situation: you spend a fraction of your budget on promoting YouTube commercials and then decide to give them up because the CAC is too high. In the following periods, sales fall not only through YouTube, but also through direct visits and organic. And also the conversion on all other channels decreases. This has disappeared the incremental effect of YouTube ads.
Visualization of the incremental contribution of advertising to sales
In this article, all methods for evaluating incremental are divided into three groups: quantitative analysis of sales; user-base methods and sociological approach (polls) .
A primitive approach
It can be expressed with the phrase: "let's count the sales before and during the campaign"
Suitable for situations when a large, but not long-term campaign is launched at a certain period. It doesn't work exactly, because brand recognition today will lead to a purchase in the future. But this is the simplest method. You can imagine it like this:
Illustration of a primitive approach to assessing the incremental contribution of advertising
Comparison of expected and actual sales curves
It can be expressed by the phrase: "let's calculate the sales before and during the campaign, taking into account the seasonality"
But in real life, the fluctuation in the revenue / sales curve is much stronger. It is determined by external factors. First of all, seasonality. You can try to account for the seasonal component to refine the effect of your ad by comparing the expected and actual sales curves.
Remove seasonal component Predict trend. Draw the sales expectation curve in the X period. Draw the actual sales in the X period. Explain their differences by the incremental contribution to advertising. Disadvantages: influence of outside factors: product improvement, good weather.
The difference between green (actual revenue) and dashed blue (expected) during an AC is its contribution to revenue
It can be expressed by the phrase: "let's get the formula"
A logical continuation of the previous method. Only in addition to seasonality, it also takes into account other external factors affecting sales: the weather, exchange rates, the number of films with Nicolas Cage, whatever ..
A fictional example of a regression model
If there is a relationship between the advertising budget and the level of sales, then the linear model will notice it and return the coefficient. This ratio will reflect the cost of acquiring a customer through the reach campaign. Care must be taken when interpreting such models. First, you need to make sure that the p-value for this ratio is below 0.05. This will mean that the advertisement has a statistically significant effect on sales.
Secondly, you need to calculate R², which will show how much of the variation in revenue/sales can be explained by the factors that you included in the model. If the indicator is low (there are no exact boundaries, but let's say less than 0.3), then there are other unaccounted-for factors that largely explain the sales.
It can be expressed by the phrase: “let's launch advertising in Moscow and compare sales in Moscow with sales in St. Petersburg, where advertising was not launched”
The approach is based on the classical mechanics of the AB test: simply compare the treated group with the control group where there was no impact. This is what Avito and Yulmart do, who can afford to buy advertising on a TV set in regions 1,3,5 and not buy in 2,4,6, and then compare business metrics between these groups of regions. For the more modest guys, there are also options: you can target the YouTube campaign only to women and compare the sales curves by gender during the advertising campaign. If you have a large customer base and you run retention / return campaigns, then 80% of the base can be allocated to the remarketing audience, and 20% of customers can not show ads, and then compare the segments that were affected by advertising and organic return.
Surveys on the ad systems side
Brand lifts are carried out on the side of advertising systems. Such tests are based on the same idea of dividing the test sample (who was shown the advertisement) and the control sample (who was not shown anything). After that, users are asked questions about brand knowledge or intention to buy. Answers can be analyzed in subgroups and find out that your ad was remembered by young people and brought purchases by the older group. Or vice versa. Or that the metrics haven't changed at all.
AB tests on actual sales
If you have a simple product that you can quickly select and buy right on the site, then Conversion-lift is ideal for analyzing the incremental contribution of advertising in a specific channel.
For example, Facebook: