How to conduct data analysis for information flow advertising?

Because information flow advertising perfectly balances the interests of the media, advertisers and users, and information flow advertising can achieve "one thousand faces for one thousand people" push through algorithms, information flow advertising has become an important part of the commercialization of media advertising. Common information flow ads: WeChat Moments, Toutiao, Douyin, etc.

From the perspective of the information flow advertising ecosystem, it currently covers advertisers, media, third-party creative platforms, data platforms, and monitoring platforms.

Third-party creative platform: provides multi-dimensional screening and viewing of delivery materials by industry, media, advertising style, material type, equipment, time, etc.;

Third-party data platform: The services provided generally include user insights, i.e. consumer portraits, user/crowd package management, and delivery conversion analysis;

Third-party advertising monitoring platform: provides statistical monitoring services for delivery and effect data. Advertisers are buyers of traffic, media or distribution channels are sellers of traffic, and monitoring tasks are generally performed by third-party agencies.

Currently, information flow ads are mainly sold through RTB (open bidding). The media will sell ad space to advertisers who can maximize their profits. eCPM (estimated cost per thousand impressions) is usually used to measure the revenue that an ad can bring to the media. Among them, eCPM = CPC bid * estimated CTR. As for why eCPM is used to measure the revenue that advertising brings to the media, if the CPC bid of a position is very high and the possibility of click is also very high, then the media will maximize its revenue.

After successfully bidding for an ad exposure opportunity, the actual charge for the ad is not calculated based on the bid; instead, it is calculated based on the second highest price mechanism, which is calculated based on the eCPM of the second highest bid and the estimated CTR of the ad itself. The specific calculation formula is:

Based on the logic of advertising bidding and the final billing logic, it can be deduced that the estimated CTR is an important factor affecting the success of advertising bidding and an important factor in improving advertising ROI. The estimated CTR is strongly correlated with audience targeting, delivery time, delivery context, and material type. The optimal estimated CTR under the delivery combination needs to be accumulated from multiple advertisements.

When analyzing the effectiveness of advertising, we must first clarify the measurement indicators of advertising. Different business scenarios have different effectiveness measurement standards, but generally speaking they are derived from user conversion rates and revenue generated. Taking the information flow advertising in the e-commerce industry as an example, the user conversion path after advertising is as follows:

Therefore, the indicators we focus on on a daily basis include the following: Generally, ROI is the key measurement indicator of channel value because it represents the true relationship between traffic cost and conversion revenue. ROI is calculated based on a certain time period, such as 24-hour ROI, 7-day ROI, etc. This can be selected according to actual needs.

After determining the ROI measurement indicator, we use it to determine whether the conversion rate of the advertising delivery group meets the standard. For delivery combinations that do not meet the standard, we can use the formula decomposition method to determine whether the conversion rate is too low, the customer unit price is not up to standard, or the cost consumption is too high. Then we can optimize the problem in a targeted manner.

• If the conversion rate is low : you can optimize the user conversion path from a product perspective, and optimize the advertising materials and audience targeting from an operational perspective;
• Low average order value : From a product perspective, we can optimize activities for new customers to increase user stickiness, and from an operational perspective, we can refine operations and fully tap the potential of different users;
• Cost consumption is too high : From an operational perspective, you can increase the estimated CTR of the ad and reduce the bid.

Determine performance metrics: click-through rate and conversion rate;

As shown in the example below, we set up two AB test groups at the same time to test whether there are significant differences in conversion rate and click-through rate under the same creative for different genders and system versions.

2. Advertising and data collection

The control group and the experimental group are launched at the same time to collect ad exposure, click and conversion data. The general amount of data collection needs to meet the following requirements:

For the two test groups mentioned above, the data we collected are:

3. Significance test and conclusion

This is an ab test tool for the overall proportion. You can input the conversion value to output the P value and whether it is significant. It is very convenient for daily analysis: https://vwo.com/tools/ab-test-significance-calculator/

Naive Bayes algorithm to optimize advertising targeting

The Naive Bayes algorithm is an algorithm for classification based on the Naive Bayes formula, which can calculate the probability of belonging to a certain category; it is called naive because it assumes that the features are independent of each other. But in real life, this assumption is basically not true. Even when the assumptions are not true, it still performs well, especially with small samples.

Bayes' formula is:

Where P(A|B) represents the probability of A occurring after B occurs. The Bayesian formula shows that to calculate P(A|B), only the last three items need to be calculated. The following is an introduction based on an actual information flow delivery case:

1. Obtain audience profile and conversion data

Assume that the user conversion data of the ad targeting is known as follows:

2. Calculate probability based on Naive Bayes

This can be calculated based on the above audience portrait data and Bayesian formula;

Suppose I want to know the probability of conversion for users with ad targeting X=(gender="male", age="35-39 years old", operating system="iOS"),

It can be calculated that P(conversion = "1" | X) = 0.9275,

3. Guidance on targeted advertising

Through the Naive Bayes algorithm and the user portrait distribution of historical conversion data, the conversion probability under each advertising targeting combination can be calculated. This way, you can give priority to advertising targeting combinations with high conversion rates, or give higher bids to targeting combinations with high conversion rates and lower bids to targeting combinations with low conversion probabilities, thereby achieving overall optimization of advertising conversion effects.

3. End

The above is the work that data analysts often need to do during the advertising process. To summarize, before the advertisement is officially launched, the optimal delivery combination will be found through AB testing; the optimal advertising targeting will be predicted through the Bayesian algorithm; after the advertisement is launched, the advertising effect will be analyzed based on indicators such as ROI, the channel value will be measured, and the product and operation will be assisted in optimizing advertising conversion to increase ROI; in addition, after the advertisement is officially launched, it is necessary to analyze from a data perspective to discover abnormal activations and abnormal orders, and then measure the channel value more reasonably based on the corrected total GMV (after removing the abnormal GMV).

I hope the advertising data analysis method introduced above can help you in your daily work or study~