Interpreting the 5 major chaos in information flow advertising and data analysis!

In the era of data marketing, more and more people are beginning to accept the use of data to guide advertising. But many people don’t know that the premise for data to guide advertising delivery is that data analysts/advertising optimizers can correctly interpret the data.

The same data, when interpreted in different ways, can provide vastly different guidance for investment.

Is it to help advertisers better achieve KPI? Or will it lead marketing into another bottomless abyss? This article will discuss various data analysis chaos from the perspective of data analysis, aiming to remind you how to avoid playing rogue with data.

Five major chaos in data analysis

1. The data itself is fake

The most common and most basic form of rogue behavior is data falsification.

For example, in order to make advertisers satisfied with the advertising work, or to persuade advertisers to continue to place advertisements, the advertisers will process the data in the closing report, beautifying everything from exposure to clicks to click-through rates, so that advertisers feel that the money is well spent and are even willing to continue placing advertisements.

Previously, a domestic real estate tycoon discovered that a certain video platform had exposed data fraud, and in a fit of rage, he checked all the online platforms he cooperated with, causing panic among the people who worked with each platform. After all, they can invest tens of thousands of dollars in advertising in a year. If they are put on the credit blacklist because of data fraud, it will be basically impossible to cooperate with them in the future.

Precisely because data falsification is becoming more and more common, more advertisers have begun to monitor their advertising.

So how can advertisers who do not monitor ad traffic identify data falsification?

There are two situations here:

One is that the advertiser himself does not understand data analysis. In this case, he can understand and analyze the industry's market conditions to see if his data performance is too good or "beyond imagination" compared with the data of his peers.

If the data you get is very good, it is best to ask the publisher to give corresponding targeting, creativity and other reasons that are sufficient to support such good data performance, and then decide whether to continue the advertising.

The other option is that the advertiser understands data analysis. By understanding the context of the campaign and the coupling relationship between indicators, they can identify whether there are problems with the data.

The so-called coupling relationship, in the field of data analysis, refers to the relationship between various data indicators. It is not isolated. For example, if the dwell time on the landing page is very short, then your landing conversion rate will usually not be very good, because the user has not read the whole message of your landing page. Users who click on the landing page and have conversion actions will basically not appear in information flow ads.

In addition to paying attention to coupling relationships, you can also refer to other analysis methods. For example, generally speaking, the distribution of values ​​of certain data indicators follows a certain pattern. For example, the peak visitor times for some advertising categories may be around ten in the morning and three in the afternoon, and the peak visit times at night are around nine or ten before going to bed. If the traffic volume inexplicably increases at three in the middle of the night, you should pay attention and analyze the cause of this abnormal phenomenon.

2. Different definitions of indicators

Indicators are concepts that describe the quantitative characteristics of a population. Many companies have their own KPI indicator system. Simply put, it uses several key indicators to measure the success of the company's advertising. For example, click-through rate, exposure, conversion rate, downloads, ROI, etc. are all evaluation indicators for information flow advertising.

Normally, indicators need to be summarized and calculated under certain prerequisites. Conditions such as time, location, and scope can all be used as prerequisites for indicator statistics, which are what we often call statistical caliber and scope.

For example, if we want to calculate a company's sales in 2017, then "sales" is the indicator, and "a company" and "2017" are both prerequisites for statistics. To obtain this indicator, we need to add together the company's sales flow for each month in 2017, and the final amount calculated is the indicator value.

In actual work, misjudgment of indicator values ​​often occurs due to unclear definitions of indicators.

For example, we often hear colleagues chatting in offline salons: "Our average customer acquisition cost is only 120 yuan."

When the friends who attended the salon heard this, they thought to themselves, you guys are so awesome, “It costs us 400 yuan to acquire a customer! What a huge gap! Can you give me some advice?” After some advice, the guy whose customer acquisition cost was 400 yuan suddenly realized that in fact, everyone spent 400 yuan, but the guy who spent 120 yuan was talking about the cost of sales leads, not the cost of actual orders.

The difference between the two indicator values ​​of 120 yuan and 400 yuan is actually a misunderstanding caused by the premise of the indicators, that is, the inconsistency of caliber and scope. If you make a comparison without understanding it clearly, the comparison itself is meaningless.

For example, e-commerce companies all have indicators of DAU (daily active users) and MAU (monthly active users), but each e-commerce company may have a different definition of these two indicators. Some companies consider customers active as long as they log in, while others may define them as active only if they make a purchase. More stringent requirements are that only users are considered active if they confirm receipt after making a purchase. So when we talk about indicators, we must make sure that everyone has the same definition of the indicators, otherwise such comparison and discussion will be meaningless.

3. Hide key information

This usually happens when someone is trying to convince you to accept a certain indicator.

For example, if the company’s management wants to set a KPI indicator for advertising, the method known as Bench marking (benchmark management) begins to be used.

Bench marking, also known as "benchmark management", is essentially about constantly looking for best practices and using them as a benchmark for continuous "measurement, analysis and continuous improvement". For example, optimizers often hear their leaders say that a certain peer’s ROI can reach 1:10 , and you see our company is not worse than theirs, and our product quality is better than theirs, so we can’t be lower than 1:10 - is this idea correct?

We say that this is a typical phenomenon of blindly following the trend, where one only knows one thing but not the other.

Why? Because we have no idea how our peers came up with the 1:10 investment, and how to follow up? How to surpass it?

When we conduct benchmark management, we must at least know why others achieved 1:10. Was it because of a large-scale promotion? Or are they using a combination of media to seize the market? As mentioned above, without knowing the overall marketing background, it is not advisable to blindly follow the benchmark just by looking at the delivery results of one channel. This will also bring great difficulties to the optimizer's work.

This problem is also prone to occur when setting indicators within the project. For example, when a fresh food e-commerce APP company issues large coupons to customers, the download rate, registration rate and order conversion rate obtained after placing information flow ads will naturally be better than when no coupons are issued. However, if the company uses the delivery indicators when issuing coupons to evaluate the daily delivery work of optimizers, it is playing tricks with the data.

Why? Because coupon issuance is mostly short-term marketing stimulus, after the stimulus fades, the probability of user response will naturally drop significantly. If certain delivery indicators are not obtained through natural growth, and the data report does not mention the key factors affecting these indicators, then the indicator requirements put forward in the context of hiding the key background are typical rogue behavior.

The natural growth mentioned here refers to the original operation of the business without the stimulation of marketing factors. The so-called key factors affecting indicators refer to factors that have a greater impact on the results of a certain indicator. For example, products, brands, prices, promotions, etc. are all key indicators under certain background conditions.

We believe that the benchmark for setting reasonable indicators is natural growth , which means that it is reasonable and feasible to evaluate indicators after excluding promotional factors.

For example, during the just-passed Double 11, a Tmall store boasted about how amazing its performance was, with sales of six or seven million in one day. But what it didn’t tell you was that the subsequent return rate might be as high as 25%.

The return rate here is the key indicator, because while the sales volume is important, what is more important is that the buyer must confirm that they accept the goods you delivered in order for it to be considered a real transaction - this behavior of ignoring the key background and only telling you the sales volume is also a kind of hooliganism.

Based on my own experience in operating Tmall for many years, for some categories, such as accessories, a return rate of more than 20% is normal. But if you don’t know, you won’t be able to accurately interpret other people’s awesomeness, and you might even blindly follow them and say things like “You’re so awesome, I might as well open a Tmall store!” or “You’re so awesome, can you help me with the operations!”

Therefore, when you see an advertisement with particularly good or poor indicators, you should clarify the relevant key background information, otherwise you may make misjudgments and cause more losses.

4. Messing with causality

Let’s take a look at a survey:

Big data shows that hospitals are the third leading cause of death among humans after heart disease and cerebral thrombosis.

Let me ask you, after reading this survey, do you still dare to go to the hospital?

In fact, we all know that the reason for death in the hospital is that these people were already sick and happened to die in the hospital. It was not the hospital that caused their death. The hospital and death established a correlation relationship, not a causal relationship.

Therefore, the conclusion that "hospitals are the third leading cause of death among humans after heart disease and cerebral thrombosis" is ridiculous. It confuses causality and correlation.

There are also classic funny cases of confusion:

The shorter the sleep time, the higher the income (so we should stop sleeping in the future)

The more people drown while swimming, the better ice cream sales will be (actually it’s because of the hot weather)

…

Why does this misunderstanding of the relationship between these two occur?

Because when we do attribution, we only see the changing relationship in the data and ignore the essential connection between things. Especially when doing data analysis, we often only see the numbers and not the process, which leads us to believe that the occurrence of a certain problem is caused by one of the problems, that is, cause and effect.

Once we confuse causality with correlation, it will directly affect our judgment and lead to wrong advertising decisions. For example, the causal relationship problem of price leading to transaction is common in information flow advertising .

We have interviewed several optimizers who run wedding photography ads and asked them why they all use the creative idea of ​​"How much does a wedding dress cost?" Their answers are surprisingly consistent: "Because customers care about the price. They will ask about the price. If the price is right, they will make the deal! "

In other words, in the eyes of optimizers and front-line sales, price is the only incentive that affects transactions, so they choose to use price as a creative direction for their marketing campaigns.

So what is the truth?

Starting from common sense, friends who have a little consumption experience will know that asking the price is a necessary step before the transaction, and it is true that the next consumption decision will be made only when the price is right. But, do you go into the store because of the price? Do you only care about price when you go into a store? Did you end up placing an order just because the price was right?

Similarly, when target customers who are interested in wedding photography enter the store or answer the consultation call, do they only ask about the price? No other questions?

Not really. They also care about shooting style, shooting scenes, clothing, makeup, etc. However, these concerns of users are not necessarily expressed in the form of questions. Not asking questions does not mean that they don’t care.

Therefore, it is problematic to simply think that the final purchase decision is directly caused by asking the price. After all, price is only one part of customer satisfaction, and there are more reasons that lead to the final transaction. We need to understand the various relevant factors that affect the target customers' decision-making and create targeted ideas.

5. The whole from the part

We say that there are actually many factors that affect the dependent variable, but analysts often only see one or two of them and hastily assume that the change in the dependent variable is caused by the change of one or several variables. Generally speaking, the conclusions drawn from this approach are one-sided.

For example, in the delivery of information flow advertising, friends often complain that the effect of XX channel is poor and not as good as XX platform, and there are inquiries as soon as it goes online. Leaving aside some differences in the platforms themselves, if we look at this issue from the perspective of advertising delivery rules, you will find that the evaluation of effectiveness requires consideration of multiple factors.

When I was working on an e-commerce project, my boss said that the ROI of celebrity store ads was stronger than that of Diamond Ads and other through-train ads, so he said, "Remove all other ads and put all the money into celebrity stores!"

So should we obediently spend all our advertising dollars on celebrity stores?

Let's use an example to illustrate: Suppose I want to buy a mobile phone:

During lunch break, I opened the JD.com website on my company computer and looked at the introductions of various brands of mobile phones;

On the way home, I saw an ad for VIVO's full-screen phone while waiting for the bus at the bus stop. I then realized that it was endorsed by the celebrity Lu Han.

While I was on the bus, I was browsing news on Toutiao and saw a review of a VIVO phone, so I clicked on it to take a look.

Not long after I finished reading the review, I saw an advertisement for VIVO on the headlines, but due to time constraints, I didn’t click on it to see it.

Before Double 11, I searched for VIVO full-screen phones on Taobao. After entering the landing page from the advertising entrance of a celebrity store, I found that full-screen phones were on sale, so I clicked in, placed an order, paid, and waited for delivery.

These 5 steps are a typical purchasing journey. From generating interest to searching for information to evaluating alternative brands and finally to purchasing decisions, I was influenced by several advertisements at the same time, but I did not pay for them until I clicked on the advertisements of celebrity stores and made a purchase. So, are other advertisements unnecessary? Can BBK Electronics remove all advertisements except for those for celebrity stores?

Of course not. My final purchase was the result of the combined influence of multiple previous advertisements.

This is also what we often call multi-channel attribution (Attribution Modeling), that is, consumers' purchasing decisions are not only affected by the last media advertisement, but there is an interactive relationship between media.

Most of the friends who have done e-commerce advertising have had this experience: after removing those channels that seem to have low ROI, they thought they could save a lot of advertising costs, but found that the channels that originally had high ROI were inexplicably not doing well. This is actually because each channel plays a different role in triggering user conversions:

Some advertisements are used to build brand awareness.

Some advertisements are used to strengthen brand awareness.

Some ads are designed to promote sales leads.

…

Isolated evaluation of which channel is effective or not does not conform to the behavioral patterns of consumer purchases.

Similarly, when you are evaluating an information flow and find that one or some platforms are not performing well but other channels are performing well, you can try to consider it comprehensively from the perspective of multi-channel attribution to see if the platform with poor performance is the straw that triggers the final purchase decision. If, after analysis, it is determined that the platform with poor performance cannot bring any help to your performance, then consider whether to cancel the delivery. Do not make hasty decisions based solely on poor conversion rates.

Three recommendations for data analysis

1. Don't assume a position

The most basic requirement for data interpretation is not to have a preconceived position.

Let me give you a very common example. For example, when a young girl in our office first came to apply for a job, she saw that I used a white mobile phone and drove a white car, so she asked me, "Do boys like white? I see many boys use white mobile phones and drive white cars. "

We say that this girl's conclusion is a typical example of availability bias. That is , once you have a preset position, you can't help but pay attention to things, objects, and people that are related to your ideas. For example, this girl, after having a preset idea, will unconsciously pay attention to people who use white mobile phones and drive white cars to see if they are men. If they are men, it fits her imagination, and she will even say self-satisfiedly, "As expected."

In fact, when you think this way, your data analysis has already shown a directional deviation.

What’s more troublesome is that in most cases, we may not even realize that we only see the things, objects, and people that we want to see.

Similarly, when doing data analysis, if you have a preset position, you may only count what you want to count and then use it to prove your existing point of view.

However, the essence of data analysis is not to use data to prove your existing views, but to discover insights from the data and find directions and breakthroughs for KPI growth.

As marketers, advertising optimizers, and data analysts who use data, you should have a rigorous and responsible attitude, maintain a neutral stance, and objectively evaluate problems in data analysis work.

2. Make good use of the God’s perspective

We say that data analysis is part of business. When we look at the role of data analysis from a marketing perspective, we can understand what the data is all about.

For example, many sales-oriented information flow ads do not convert well. What data analysts need to do is not to delve directly into the click-through rate, exposure and other delivery data, but to first identify what the prerequisites for the effectiveness of the current advertising behavior are?

That is what we often say, while you are busy working, don't forget to look up at the sky.

Let’s take the common second-tier e-commerce marketing behavior of placing advertisements for sales as an example. For example, a leather shoe OEM factory owner invested tens of thousands of yuan in information flow advertising for the purpose of sales. He optimized it many times during the period, but it was of no use. The final sales data was very bleak. Why is this?

Judging from the launch itself, he seemed to have tested all possible situations, but to no avail.

But if we analyze it from a marketing perspective, this question is easy to answer.

We say that most sales-oriented advertisements only do one step, which is to put the goods in front of consumers. In an era of backward productivity, such as when the planned economy was supply-based, putting the goods in front of consumers might be enough to trigger purchases.

But now many industries, such as the leather shoe industry we mentioned above, have an overall overcapacity, and the battlefield of competition among merchants has shifted from shelf to the minds of consumers. So at this time, it is far from enough for us to simply get the goods in front of consumers. Because your competitors have already placed their products in the minds of consumers before you appear. In other words, long before you arrive, your competitors have already completed pre-sales in the minds of consumers - consumers have already bought them, so why should they buy you?

Note: Du Jiang's distribution theory

Product distribution to the brain: Build a brand and seize the mind

Goods in hand: build channels to facilitate purchases

Goods delivered to the mouth: Optimize experience and promote word of mouth

Therefore, once we understand the distribution theory, we can explain why the conversion data of leather shoe advertisements is not good, because the key to conversion is not in the current marketing behavior. What the boss needs to do is to conduct research and analysis to see what opportunities for differentiation there are in the leather shoe category, and seize market opportunities by creating and leading a new category.

3. Be sensitive

As mentioned at the beginning of this article, the premise that data can guide advertising delivery is that data analysts/advertising optimizers can correctly interpret the data.

Given a certain set of data, the main factor that limits the effectiveness of the data is the analyst's ability to interpret the data. For example, the situation mentioned in the previous article where causality and correlation are confused, the misunderstanding of interpreting the overall pattern from a local perspective, and so on.

So I often say that if you want to do data analysis, you'd better be a sensitive person: you should be sensitive to data changes and even more sensitive to the reasons behind data changes.

For example, we often see major Internet apps offering crazy subsidies in the early stages of promotion. Some of them have gained a large number of users through subsidies, and the company's development has become smoother as the number of users grows, such as Didi Taxi. However, some disappeared shortly after the subsidies were given. What is the reason behind this?

If you only analyze the conversion funnel diagram in the early stage of promotion, you will be more confused and will not gain much, because from the funnel diagram, the situation of APPs with good subsequent development and APPs with poor subsequent development in the early stage of acquiring users is very similar:

But if you dig deeper into users’ behavioral responses after they use the app for the first time, you’ll find more valuable insights.

For example, for an APP that develops well in the future, many of its users will spontaneously refer others to the APP, bringing more users to the APP. However, for an APP that develops poorly in the future, its referral and follow-up usage are very poor:

If you discover this problem before your peers, you can analyze the data before them to see what causes the difference and what causes the difference. Then, you can use these insights to guide subsequent marketing activities to achieve more efficient market penetration.

So, more often than not, data analysts are more like Sherlock Holmes, maintaining a sensitive curiosity and asking more questions about everything:

Why is this the result?

Why isn't that the result?

What is the cause of this result?

Why is the result not as expected?

…

In this way, you can find the breakthrough point faster and more accurately than others, and then follow the clues to find the business truth behind the data.

Finally, to borrow Steve Jobs’ famous words, “stay hungry, stay foolish” , I hope that more data analysts like me will have smoother progress in their pursuit of the truth and discovery of business insights.

Source: Precision delivery of information flow advertising (ID: feed-advertise)