The traffic pattern of information flow video ads!

After working in advertising for so many years, when I asked many optimizers around me what they thought was the most important ability of an optimizer, everyone generally mentioned two points. One is the ability to create or sensitively perceive high-volume materials, and the other is data analysis ability.

But if you ask specifically, how to do data analysis well and what data need to be analyzed, almost everyone's answers are similar, click-through rate, conversion rate, CPM, cost... How to optimize? If the click-through rate is low, optimize the attractiveness of the material. If the conversion rate is low, optimize the landing page... A small number of optimizers will also mention that indicators such as 3-second play rate and completion rate are important, but if we go a step further, how much impact does the 3-second play rate have on the running volume? How does it affect? Many people may be confused.

01. Conventional information flow data analysis

Here is a brief description of what actions conventional information flow data analysis may take

① Look at the click-through rate, conversion rate, play rate, CPC, etc. to roughly analyze the quality of a material;

② Look at the conversion rate, bounce rate, reading progress, etc. to see the attractiveness or appeal of a landing page;

③ Look at the time trend and time period data, and analyze it in combination with the magnitude and cost trends;

④ Look at the population report, age, gender, regional distribution, etc., to determine the high-quality conversion population;

⑤……

All of the above are correct, and the specific elaboration is also a very systematic data analysis.

But if the video material does not reach a certain volume, or if the material does reach a certain volume, what will be the performance from a data perspective? What other breakthroughs can be found in data analysis?

When we export the video analysis table in the huge background report, we can see so much content

I have been wondering, with so much data, how can we make use of it besides the conventional display, click, consumption and conversion? Indeed, after studying the conventional data for so long, we often find that they are distorted. High click-through rate and high conversion rate do not necessarily mean high volume, and high thousand impressions do not necessarily mean high volume. As for the high volume, the various indicators do not seem to have any special advantages... No wonder some people say, "delivery is metaphysics."

02. Is it possible to have a comprehensive indicator of running volume?

Later, after a series of research, I found that the following indicators may have an impact on running volume and can be taken into consideration when analyzing data:

1. Bidding indicators: click-through rate, conversion rate, bid, ecpm... Everyone knows this;

2. Completion indicators: 3-second playback rate, effective playback rate, 99% playback completion rate, etc.

3. Interaction indicators: number of comments, number of shares, number of likes

4. Negative indicators: Reporting, lack of interest. Many people may have noticed these.

But is there a formula that takes all of these into consideration and, through simple calculation of the above four indicators, derives a "comprehensive indicator"? The higher the comprehensive indicator, the more likely it is to achieve volume growth? Theoretically, there isn't, because from the system's point of view, there are more influencing indicators, and there will be account weight, crowd attributes, competitive environment, human intervention, low-quality material flow restrictions and other factors that we cannot see (just the point of "what kind of material will be judged as low-quality" is a black box, this is a process of rough and fine sorting in the media").

Here I would like to share with you that some optimizers may not know that there are 10 processes before an ad is displayed:

1. The number of ad requests initiated by users at this traffic level

2. The number of ad requests remaining after filtering out user requests that are not within the ad targeting range

3. The number of remaining ad requests after filtering out requests from users who have viewed the same type of ad too many times

4. The number of remaining ad requests after being filtered out due to lack of interest in the ad or inappropriate material style, etc.

5. The number of remaining ad requests after filtering ads without sufficient available budget and balance

6. Estimate the probability of being successfully placed after entering the rough ranking

7. Estimate the probability of being successfully placed after entering the refined ranking

8. Participate in ECPM ranking

9. ECPM bid wins

10. After mixing with user content, ECPM bidding wins

Here, both coarse sorting and fine sorting are difficult to directly control and adjust through optimization. Relatively speaking, the four indicators I mentioned above are visible to optimizers and relatively controllable.

Another point that needs to be emphasized is that no matter what indicators we analyze in the end, it does not represent the "answer" to whether the material is running in volume or not. I very much agree with what some people have said that at the algorithm level, the impact of the data analysis we do on the volume is limited, but optimization is still a "do your best and leave the rest to fate" job, and we need to do our best to optimize within the scope of what we can see.

03. How to explore the running volume indicators of information flow video advertising ?

So assuming that other factors (black box factors) are similar, is it possible for us to find a similar running volume indicator? My idea is to use the results of running volume to work backwards. First, I export the delivery data of all video materials, and then compare the various indicators of each material. If the final calculated comprehensive indicators of each material are also distributed in descending order (there is no way to be completely consistent, we can only show this trend), then we have found the "possibly correct" formula.

Here is an example of a formula I used during my own exploration:

Assume that the bidding index = click rate * conversion rate * 2

(Because there is no obvious difference in the bids in the accounts I tested, and no automatic bidding was launched, so I did not consider the bid factor when comparing)

Assume that the interaction index = (comments + shares + likes) / effective playback number * 1000 / 4

Assume that the comprehensive index = bidding + completion + interaction - negative * 2

Where did these formulas come from? Why do we need to multiply by 2 and divide by 4? In fact, it is just a bold assumption + verification. This process requires repeated adjustment of parameters and calculation methods, and finally making the results of the comprehensive indicators as consistent as possible with the running volume results.

And in the process of adjustment, you need to think about which factor has a greater impact on the running volume. For example, I think that the competition indicator is very important and needs a higher weight, so multiply it by 2. The interaction indicator may not have such a high weight, so divide it by 4 to lower the indicator.

Similarly, you can also test how to calculate the playback indicator. Is 3 seconds of playback more important? Or is 99% playback progress more important? Or is the effective play rate more important? In the interaction indicators, is it possible that comments, shares, and likes are not equally weighted, because maybe likes and shares have higher weights, and the number of comments is not important? If you are interested, you can try to calculate in these directions.

Finally we can get the data analysis table as shown in the figure:

(Note that you need to exclude materials with low consumption or display. If the data is too little, it may not be meaningful for reference.)

What is the use of these?

1. You will begin to pay more attention to the various indicators of the video materials. For example, the volume of your material suddenly decreases after only two days. It may be that the negative indicators are increasing, while the click-through rate and conversion rate have not changed. At the same time, you can also roughly know the impact of the completion data on the volume. Some materials may have poor playback data, but the bidding indicators are too high, so the volume is very high.

2. The test will produce some verification results. For example, I found that the impact of competition indicators is indeed the greatest. Increasing click-through rate and conversion rate is useful, but blindly improving a single indicator is useless. For example, if you think the click-through rate is low and you produce a material that increases induced clicks, the conversion rate will definitely drop, and the competitiveness of this material is still not strong. In comparison, it is more useful to optimize the landing page without changing the material;

3. Under the condition of less consumption, it is possible to preliminarily judge the quality or life cycle of a material. For example, although a material may have a high bidding index and can run volume, its interaction index and completion index have always been poor. In this case, its attenuation may be faster, and other materials need to be quickly reserved. The specific life cycle of indicators needs further testing.

4. Actually, to be frank, the result of this analysis is not important...or in other words, there is no need to overestimate its role. What matters more is our thinking during the analysis process and how we judge different materials based on the data, and even where the differences between accounts lie.

The above is the summary result after a certain period. We know that many materials have cycles. They may be good today and bad tomorrow, or suddenly increase in volume one day and suddenly decrease in volume another day.

So we can also review the data of some running materials combined with their consumption trends. For example:

The above consumption is the total consumption within a period of time, and the total index situation. It can be seen that the material with the first consumption has the fifth index, while the material with the first index has the sixth consumption.

Combining the consumption trends of these materials, the material with the highest consumption did have a wave of volume in the early stage, but then it decayed, pulling down the indicator. In fact, this indicator should be above 20 during the period of volume growth.

As for indicator No. 1 and material No. 6 in consumption, after about a week of exploration, the volume did pick up. So the trend is that the higher the indicator, the easier it is to grow the volume.

Furthermore, we can pull out the daily indicator reports of these materials for analysis, and study which indicator's change caused the material's attenuation process. This can also provide reference to a certain extent, but I won't go into details here. It’s enough as long as everyone can understand the logic and thinking direction here.

Then some friends may want to ask me what the formula I measured is. I think this cannot be applied. The reasons are: 1. First of all, this is my personal opinion, not the "answer" of the algorithm. It is only for reference in data analysis and cannot be blindly believed. 2. The accounts and industry data I analyzed in the derivation process of this formula are not much. It may not be applicable to other industries or accounts, but it may be misleading. 3. I suggest that you analyze the materials with this idea in mind based on your own accounts and materials. If you can summarize something, it will be a greater gain than getting the formula I measured.

Finally, I would like to add a few more words. If you have found a possibly accurate indicator formula in your account and industry, don’t just focus on the indicator. There are a large number of “black box indicators” that we cannot see, and they are more of a reference.

Moreover, this has limited effect on actual decision-making. Ultimately, we still have to return to the form, type, and copywriting of the materials that can be sold in large quantities. At the same time, we must avoid the homogenization of materials and return to the users, so that we can have stronger creative and optimization capabilities.

Author: Shen Zhanyong

Source: Sanlitun Information Flow