Analysis of practical cases of information flow in 2019!

Information flow advertising is called performance advertising, which means that the effect can be directly seen through data in the short term .

Optimizers deal with data every day: they have to prepare hourly, daily, weekly reports... Any adjustment of accounts must be based on data analysis. Today we will look at the basic ideas of data analysis and introduce a simple formula and case.

I am familiar with Toutiao and am used to it. At the account level, I still use the term "account-group-plan-creativity".

1. The main problems of data analysis

There is too much data.

The first is that there is too much data on advertising. The media provides data in nearly 100 dimensions, including display, click, download, activation, conversion, video play rate, number of likes, number of shares... plus the backend data, how should I analyze it? I can’t even watch it all!

What should I do then?

2. Have a key indicator

You have to have a key indicator, and the fewer the better, preferably only one.

This will have two advantages. One is that it will be very convenient for you to judge what is a good plan. The other is that if you only look at this one indicator, you can achieve the effect of "it doesn't matter whether the cat is black or white, as long as it can catch mice, it is a good cat."

Let’s look at two cases:

One day I discussed this with classmate Ji.

She said, "What do you call a good plan?"

I said "the cost is adequate and the quantity is high".

"So what is the judgment on cost?" She works on loan products and assesses credit (credit means that the credit limit has been approved for the user and the user can withdraw the money at any time).

"Subject to the cost of your final assessment"

“What if there is a plan where the registration cost is very low but the credit cost is very high?”

"Then it is not a good plan, it will only work if the cost of the final assessment is low."

Xiaoxi said that he found that the CPM of headlines connected to deep conversions was incredibly high. I asked, "How high? 40 or 50 CPM?" She showed me the average CPM of several main materials, the highest of which was almost 300... I was shocked.

Xiaoxi mainly invests in loan products, which are connected to effective customer acquisition. Since the conversion rate is very high (a conversion rate of more than 20% means that there is 1 conversion for every 4 or 5 clicks), even with such a high CPM customer acquisition cost, it is OK.

By only looking at key indicators, you can be more lenient with or even ignore the process costs. No matter whether the registration cost is high or low, or how ridiculously high the CPM is, you don't have to care too much. If the cost of the final step you want is OK, it is a good plan; if it is not OK, it is not a good plan.

“There are thousands of data indicators, and if one is good, you will make money!” The core of advertising is cost and volume. If you can identify one indicator, you can just work on cost and volume based on this one indicator.

With the key indicators, we can clearly see what to look at in data analysis. Next, let’s take a look at the specific analysis process.

3. Prerequisite for data analysis: the data sample must be large enough

A data sample is the portion of the data that you want to analyze. If the sample size is not large enough, the analysis will be meaningless.

For example: A reporter asked three girls: "What is your favorite lipstick color?" Girl 1 answered "green", Girl 2 answered "I think it's green too", Girl 3 answered "still green", so the reporter concluded "Amazing! This year, there is a new trend in lipstick colors, and the most popular lipstick color is actually green!"

Don’t you think it’s ridiculous? How can we draw conclusions based on just these few people?

Regarding the amount of data, the professional term is called "data confidence" . We just understand that when the data sample is very small, it is unreliable and no conclusion can be drawn.

Such situations often occur in advertising, such as the following plan: when the number of clicks was only 8, the number of conversions was 4, and the conversion rate reached 50%! Isn't this a genius plan? Should we immediately increase its budget to unlimited?

Look at the consumption, friends, it’s only 2.2 yuan.

Can the conversion rate of this plan be stabilized at 50%? No way. Even if the data seems surprising now, you can't increase the budget too much all at once, because it is very likely that the conversion rate will drop if you spend too much, and you will be stupid then. Similarly, if the cost is very high, for example, the required cost is 5 yuan, and you have spent 15 yuan but there is no conversion, you cannot just turn it off because the data may also get better.

So, remember the first point: don’t make any judgments when the data sample is very small. When the data sample is large enough, we can analyze the advertising effect.

So what does big mean? To give a rough reference value, at least one of the following conditions must be met: more than 100 clicks/more than 2,000 impressions/more than 50 consumption. If your conversion cost is high, then these reference values ​​will also need to increase.

After meeting the prerequisites for data analysis, let’s look at the ideas for data analysis. I have summarized a mnemonic that is easy to remember.

4. Data analysis tips

Doing data analysis is like investigating a case. You have to follow the clues and look for the problem bit by bit.

Large pumpkin on the vine

The melon grows on the vine, just follow the vine and feel it little by little. How do you touch it? You can follow this mnemonic : “from big to small, from many to less, from one day to several days”.

Let’s take a look at this with a case.

"From big to small" means "first look at the account as a whole, and then look at the account as a whole down to the group, plan, and creativity separately."

First, look at the overall cost and volume of the account and get a total. Judge the overall situation of the account by the total number and have an idea of ​​it in mind. After looking at the totals, look at the breakdown. An account is composed of many advertising plans. So when the account cost is high, it must be that some advertising plans have high costs. So we need to find out which plans have become more expensive. Because operations are mainly at the planning level, you have to find a plan to solve the problem. You can't see the problem just by looking at the overall picture.

Next, we need to use the second sentence "from more to less".

What does “from more to less” mean? For example, if you have 100 plans in your account and it would take you a whole morning to look at them one by one, which one should you look at first? Of course, we should catch the leader first, and look at the one who spends more money first! Sort by consumption in descending order and look for those with the highest consumption, as they have the greatest impact on account costs. Those small fry who can only spend a few dollars can never achieve anything big and have no impact on anything.

“From more to less” means “spending money from more to less.” After sorting the data of the entire account in descending order by consumption, use KPI as a yardstick to find the plans with obviously high or obviously low costs. This is the plan we need to focus on. "Oh, I found out. The cost of the plan that consumed the most yesterday doubled. It caused the overall cost of the account to increase! Why did it increase?"

You can’t draw conclusions by looking at the data for just one day, because data will normally fluctuate, so you have to “go from one day to several days” and look at the data from the previous few days. It's like when you have locked on a suspect, you need to go over his whereabouts in recent days - pull out the data of the suspect's plan in recent days and then make a judgment. At this point, you can roughly determine whether the costs have risen suddenly or have always been high, whether it is a new plan, whether it is in the compensation period, etc.

I found out that there was a problem with the data on a certain day, but I still don’t know what caused it. What should I do?

At this time, you have to be more meticulous. You can't just look at the key indicators. You have to work backward from the key indicators. Any indicator has two influencing factors (the previous step and the cost and the conversion rate to this step). Work backward step by step.

If there is a sudden drop in volume or a sudden increase in costs, you need to "check the time-sharing data in the corresponding operation log" to see if there is a significant change in the data after the operation.

I have to praise the updated report of Toutiao. Although I won’t use it for a while after the update, this function is absolutely good: draw a line in the data trend, click "Details" from the planning level, open the data to see when the operation was performed, and you can intuitively see the trend of the data after the operation.

5. Example analysis

Let’s take optimizer Forrest Gump’s analysis of an account as an example to see how the above ideas can be applied.

The first step is “from big to small”, looking at the overall situation of the account first.

The key metric is the last registration cost. The total consumption yesterday was 150,000 yuan, with a cost of 27 yuan, which was lower than the customer's requirement of 28 yuan. There is no problem with the overall cost, but the scale needs to be improved.

Pull out the data of an account to see the sub-plan situation.

We look at it from "most to least", in descending order of consumption, and first look for those with relatively large consumption, which looks better. Then use KPI as a baseline and find those that are lower than KPI and those that are higher than KPI.

Analysis:

• The first four plans are relatively expensive, but they can be invested freely.

• Looking further down, we found some items with very low costs, such as item 5, where the KPI is 28, but it only reached 22 or 23. This is a real surprise. We marked this part of the plan in red and will take a closer look to see if there is any way to spend more money.
• There are two other plans with very high costs, reaching 35 and 39 respectively, far exceeding the KPI. These are problematic plans, and we have marked them with black frames and will take a look at them later.

We have already identified very low-cost and very high-cost plans, and these are the ones to focus on. The idea is "from one day to several days" , so let's start with Plan 14. Pull out the data of recent days of Plan 14.

I found that this plan had just started, and the consumption was very low in the first few days, and the cost has been very high. There is no problem at this point of view, but we need to look more closely.

• If the cost is high, you must first look at the price. The account backend conversion goal is activation, from assessment to registration, and there is a conversion rate step in between. And the registration cost can only be seen the next day, so the adjustment of the account requires some speculation.

The conversion bid is 11, and the cost yesterday was 10.5, which is basically close to the bid. Then, based on the registration cost, the registration rate is nearly 30%, which is neither high nor low in the overall account. The activation cost is 10.5, which is relatively high.

• What about the consumption speed? I spent more than 5,000 yuan yesterday. Looking at the real-time data today, I have spent more than 3,000 yuan so far. I am spending money faster than yesterday, so my spending power is not bad.

So, for the time being, we have to control the budget today and not spend too much if the cost is high. How much you can spend depends on the overall cost of the account. If the cost of other plans is lower than the KPI, this plan can make it cost about 10,000. In the long run, the registration rate of this plan is unlikely to increase, so lower the price. Lower its price to around 9.5 and see if the registration cost will go down.

This is another very expensive plan, with data for Plan 15. How would you adjust it?

Overall, it has been invested for 4 days and cost less than 2,000 yuan. The data sample size is acceptable;

Looking at the cost, its activation cost is normal, but the registration rate is only 20%, which is far below the average level (average 32%). According to the current registration rate, if the registration cost is to meet the requirements, the activation cost must be 5.5. The bid is 10 yuan, and it is easy to bid too low, and it is basically impossible to reach the target. You can close it now, or you can invest again to see if the registration rate can go up.

I usually give the plan more space, and the account cost was relatively low at the time, so I didn’t turn it off. Later, the volume increased and the registration rate did rise, but the cost was still high. It was shut down on April 3rd.

Through such systematic analysis, we can clearly judge the situation of each plan and make targeted adjustments based on the performance of the plan.

6. Today’s optimization experience

To summarize, the basic idea of ​​data analysis is:

1. There are thousands of data indicators, and if one is good, you will make money!

There are many dimensions of data, you only need to focus on the most important one. This will help you get more volume.

Just like a boy looking for a girlfriend, you want to find one who is gentle, virtuous, and good-looking... then the only one is Chang'e, where is Chang'e? You can only look up to the sky; if you say directly that you want to find someone good-looking, there will be some drama. Don't ask for too much, my friend.

2. The premise of data analysis is that the data sample is large enough.

You can’t jump to conclusions just because three people say something is good. That’s not wise. You can’t judge the quality of an advertising plan before you spend 20 yuan. You need to observe it for a while before you pause or increase the budget.

3. The key to data analysis is “from big to small, from many to few, from one day to several days”

You firmly believe that if you patiently "follow the vines to find the melon" and search bit by bit, you will definitely find the "melon".

When looking at data, first look at the overall picture, then look at the sub-plans. This is also the order when reporting data to customers. Account

If there is a problem with the household as a whole, check the sub-plans; when looking at the plan, find the "boss" first, and look at the consumption from most to least, using KPI as a yardstick to find the costs that are obviously high and obviously low; if an unusual "suspect" is found, check his whereabouts in recent days; if it still doesn't work, you have to check the "surveillance video" and analyze the problem based on your experience. If you have done all these and still don’t know what’s going on, then let’s talk about “media traffic fluctuations”.

Author: Aunt Ning

Source: Sanlitun Information Flow