What should I do if I don’t know how to perform data analysis for information flow advertising optimization? The answer is here...

When doing data analysis for Internet operations , the first thing to do is to learn how to “decompose” . Break down the data, break down the problem. All data can be broken down layer by layer to find more "sub-data". By mining and optimizing the sub-data, we can often break them down one by one, find the direction, and improve the final "key indicators" . Many times we cannot find the direction to work towards, often because our ability to decompose is not enough. We only focus on the final large transaction volume indicator and do not explore the related factors behind this indicator. These factors are the so-called details, and if they are done well, they become "extreme".

Another benefit of the decomposition approach is that it can help operations better divide the work and optimize the organizational structure. Make employees more professional and focus on a certain business, thereby cultivating experts in a specific function. When each specific function has an expert, it will be reflected in the perfection of operational details.

Operational problems are discovered through tracking, not just identification in one go. All data can only be discovered through accumulation and sedimentation. A single number has no meaning and can only be called a "numerical value." For example, a store’s traffic today is 2,000, the conversion rate is 1.5% , and the transaction volume is 3,000 . I don’t know whether it is good or bad, whether it is an improvement or a regression. Only by studying the data from the past week, month, or even year in a linear trend can we find the "problem" and the data at this time is meaningful.

Therefore, no matter which stage the store is in, it is necessary to develop a habit of accumulating data every day. We know that Taobao’s backend will have mature data products that will show you a trend and historical data, but this is far from enough. You need to extract all the data into your own database, combine different data dimensions to comprehensively analyze the problems, and establish a tracking mechanism, which is the "combination of ideas" mentioned below. Medium-sized and larger e-commerce companies will have their own data management model to monitor key indicators to ensure that problems are identified in a timely manner and corresponding decisions are made.

Tracking only one data point will be biased and the conclusion drawn may even be wrong. Because all the core data of e-commerce are accidental and correlated over a period of time.

Randomness means: one day, the conversion rate may suddenly drop to be much lower than usual. This is very possible. As a result, everyone panicked and looked for factors related to conversion rate, looked at the design of the product details page, the price of the product, and the customer service chat records. They "optimized" the design of the details page for a whole day, making the product price lower, and the pre-sales customer service was rectified. In the end, I found that everything was still the same, and I wasted a day doing a lot of useless work.

Correlation means: most indicators are correlated, positively or negatively correlated. The conversion rate suddenly dropped, and we finally found that the traffic suddenly surged yesterday. Looking at the source of traffic, most of it came from promotional traffic, which is not accurate, but there are many people.

Therefore, tracking data must be viewed from multiple dimensions. Generally speaking, conversion rate and traffic are negatively correlated. If traffic surges, conversion rate will drop; if conversion rate rises, average order value will drop . (Except for major promotions)

However, even if we track the data and combine multiple dimensions to analyze the data , the conclusions may still be inaccurate. The reason is that these two ideas are comparing with "ourselves", and we also need to compare with "others". This is the "contrast idea" introduced below.

Comparison means comparing yourself with other people. The other person must be selected as "appropriate". It can be the store data with similar brand positioning to your own, or it can be the store data that is performing relatively well in the same industry. The most comparable ones are those with the same level of stores as your own. Only through comparison can you discover where your gaps lie and find the right direction for optimization.

An actual case: When we were making microwave oven products, our sales volume was always lower than that of our competitor, Galanz. Then we analyzed the data and found that the traffic volume was much different. As a result, they increased their investment in display (diamond booth, CPM) and bidding (direct train, CPC) advertising, but found that the results were minimal and even sacrificed most of the profits. Finally, we took a product of the same type and did an in-depth comparative analysis. We found that there was a big difference in the natural search traffic sources, and we realized that it was a problem of brand awareness. This has prompted brand owners to focus on building product brands.

The node idea is to mark major marketing events as nodes separately, and remove the data for separate analysis. In daily operations, marketing activities have a huge impact on data, especially if you suddenly participate in Taobao's official activities, such as Juhuasuan, which will cause the traffic, conversion rate, and transaction volume to soar in a few days. At this time, if we insert this data into the daily operation data analysis, it will cause "distortion" and affect the judgment of the optimization direction of the store's daily operations.

In information flow advertising, some people think that data is not important, while others think that data is very important. I think data analysis is an indispensable link in advertising: exposure, click-through rate , click volume , bid, conversion, etc. are all key indicators; every piece of data is important. The above ideas are the basis of data analysis. With such analysis ideas, no matter what content you are working on, you will quickly find the core problem and then find a solution to the problem. These analytical ideas are applicable to Internet operations, and many of them can be applied.