How to conduct refined data operation analysis?

The degree of refinement of efficient operations may gradually become a moat between future products. To implement refined operations, we must first do a good job of product data analysis. This article is a summary of the author's insights from his own data career.

It has been five years since I last wrote an article. Looking back, the happiest thing at that time was competing with my peers on Weibo. Big data was very popular at that time, and I was young and energetic, full of aspirations, and I never got tired of it. However, the development of any concept goes through a cycle from rise, to hype, and then to decline. Whether it can rise again depends on the test of "financial report performance" - whether the value of the data is overestimated or underestimated.

On the road of practicing the value of data, every data person is actively preparing for the exam in his own way , so I feel it is necessary to write something more and open my exam paper to my colleagues, not for him, but only for the dream and passion for data.

Looking back at the above, my career in data started from a consulting company, where there were slightly more academic indicator systems and methodologies. Then apply it in data service companies, where there are more mature data collection implementation technologies. Later, he applied and commercialized the previous knowledge and skills in Weibo Data Startup Company. Looking back, the views at that time have temporarily stood the test of these five years. At the same time, precisely because I had no experience in data analysis for the client company, I did not have a deep understanding of the importance of the business at that time, and I did not mention it much above. This article will focus on:

1. If you don’t understand the business, analysis is just a matter of numbers

To borrow a joke from big data, the phrase "data analysis requires business understanding" is like "Teenage Sex" - everyone is talking about it, but no one knows what it means to understand the business. Everyone thinks everyone else understands it, so everyone thinks they understand the business . Please allow me to repeat what it means to understand the business from a personal perspective.

Let’s take other people’s products as an example——

For content products like Toutiao, its business model is nothing more than (content) production, distribution and monetization, thus realizing a closed business loop from investment to profit and then to profit growth. In order to make this model work, it must have content (producers), users (consumers), platforms (consumer platforms), and advertising (to incentivize producers and platforms).

(1) If you understand the business, you would not simply suggest increasing advertising when the daily active users show a downward trend, which is a suggestion that everyone knows; you would not simply report the same-month and year-on-year growth when producers show signs of leaving due to the increase in incentive fees for competing products; you would not crudely suggest that the business side reduce the frequency of advertising globally when the amount of feedback about advertising increases abnormally...

(2) If you understand the business, you will know how to advance data work. The first stage should be the construction of basic data to ensure the standardization, panorama and expansion of data collection, and to ensure the efficiency and stability of the data production process from tracking → collection → cleaning → statistics → storage.

The second stage focuses on the data performance of seed users, which product functions users have difficulty using, which categories of content they prefer, and which channels have higher quality users. The above conclusions are simultaneously output to the business side, and the PDCA cycle is continued until the life-time calculated by the retention rate is sufficient to support the outbreak period.

The focus of analysis during the outbreak period is to continuously improve operational efficiency, such as organizing production in a targeted manner according to user preference characteristics, and then expanding distribution scenarios, from distribution within the APP to distribution outside the APP, and continuously improving the distribution efficiency of a single piece of content; optimizing the product's column layout, function buttons and other traffic flow designs to meet the usage preferences of different groups of people and improve "sales per square meter and sales per person"; from buying users to waiting for users to increasing users, who is the current core user group of the product, has it been thoroughly penetrated into the social population, if not, through what channels can they be "captured" and through the design of sharing/forwarding strategies to achieve user self-growth. The various goals in the second phase are to continuously expand the scale effect of users and content and prepare for commercialization .

The third phase of analysis focuses on the performance on the commercial side. Whether the content is self-produced or UGC, it has costs. The costs are converted into traffic, and the traffic is monetized through commercialization. Therefore, it is necessary to optimize the current advertising format and strategy through data to help sponsors find the most suitable users and let users find the most needed ads, so as to maximize ROI.

The fourth stage should focus on innovative development, what are the current development models of domestic peers and their respective differentiated competitive points, whether there are similar industries abroad and what the current situation is, what extended needs of users are not met, what are the future development trends of the content industry, and policy risks such as laws and regulations that may be encountered.

(2) If you understand the business, you will know what your boss is concerned about at the corresponding stage, and you will be able to design reports that are more in line with the business perspective. Through corresponding thematic analysis, you can answer the "needs" that your boss has not yet expressed.

(3) If you understand the business, you will first think of understanding the KPIs of each business role. When it comes to team collaboration, the most powerful method is to drive them by benefits rather than by reason. When business personnel know that you are a community of interests, good collaboration is guaranteed.

…

Having said so much, the question is, how do you test whether you understand the business? I have a little experience, which is to see where you spend most of your time and where your output is? If you understand the business, your main output will definitely not be to increase the numbers, because your boss and the business department know that asking you to increase the numbers would be a waste of the company's manpower and harm their own interests. Otherwise, the analysis is merely a matter of numbers.

2. Return to the essence: data can empower business

Quoting Baidu Encyclopedia’s explanation:

Data is numerical value, which is the result we get through observation, experiment or calculation. There are many kinds of data, the simplest of which is numbers.

The essence of data is numerical value, which is just a result. If you want to change the result, you can only find the cause and make changes to the cause to cause numerical changes .

This passage may not be easy to understand. Here is a universal formula for water flow that everyone knows:

Turnover = daily active purchase rate x average purchase amount per person. This formula can be broken down further and the factors after decomposition can be assigned to different business groups, which is called KPI. Experienced people all know that the greatest significance of this formula is tracking and monitoring, and it cannot be used as an execution target. It may be okay in the early stage, but after a stable period, once the daily active users increase significantly, the purchase rate and average purchase amount will decrease instead; the business has done a lot of optimization, and the purchase rate has been improved with great difficulty, but the purchase amount has decreased instead; in order to meet the average purchase amount target, the operation recommended a lot of high-priced products, but the purchase rate dropped again. . .

Why? Because cash flow is just a result, which is generated by user decisions. The correct factors that determine cash flow should be the user's demand intensity, purchasing power, and the degree of match between users with corresponding purchasing power and products in the corresponding price range . If we do not consider the solution from the perspective of cause and effect, but only pursue various pseudo-factors under the current stock purchasing power, we will end up suppressing one problem but causing another to pop up. Especially in the client's business environment, each team is closely focused on the basic route of the core KPI. If the data side falls into the KPI analysis needs of each business team and is not corrected in time, the consequences will be disastrous.

Let me give you another example. The following figure shows a commonly used data report view in the business. With the iteration and refinement of the business, various reports are piled up. Even if there are tens of thousands of such reports and even if only minute-level abnormal monitoring is performed, it may not help improve performance. The stock price will still fall.

Let's change the view.

The above table headers are just for reference and are not expanded in detail. The main idea is to transform the result-based report into a process-based report, and divide the entire report into four units from the user's perspective: basic attributes, interest preferences, usage characteristics, and business contributions .

1. The basic attributes are mainly user basic descriptions represented by the date of addition, channel, model, gender, age, etc.
2. Interest preferences are characteristics that users show after using a product, such as liking card games, RPG games, and other categories of games.
3. Usage characteristics are the data behaviors left by users when using the product, such as the number of views/clicks/searches.
4. Commercial contribution measures the user's contribution to commercialization, such as the number of purchases and the purchase amount. Commercial contribution combined with basic attributes is actually the entire monitoring of user LTV.

With this view, we have independent variables and dependent variables, and we can return to the multivariate analysis methods we are familiar with, such as factors, regression, and discriminant. As for the output of RFM, CRM, channel evaluation/anti-fraud and other solutions, it goes without saying.

It is based on this procedural data structure that we have done a lot of interesting project research, such as how to increase game downloads, how to improve user activity, how to reduce uninstall rates, how to improve PUSH conversion efficiency, how to double current revenue, how to recall lost users, etc. , and innovatively combined with text data such as user feedback, we have smoothly reproduced quantitative + qualitative research methods that can only be implemented in traditional market research companies in Internet business models. As for the project results, I'm sorry that I can't provide too much, but I want to say that this circle is actually not that big, and it's not difficult to find out.

I have always believed that there is no end to the business of data analysis, because the subjects who generate data - people are always changing. All experiences and methods may be your weapons today, but may hurt you tomorrow. The advantage of being familiar with the business is that you can have the same dialogue context and position, but the disadvantage is that you often forget the essence of the data because you are too close and too fast. A good analyst needs to build an analysis system of his own, among which a very important link is the self-correction mechanism, which I am also exploring.

3. Data first, growth can be more stable, accurate and ruthless

With the demographic dividend disappearing and the Internet traffic growth approaching its ceiling, the concept of growth hacker is becoming more and more popular. Here, I would like to say two things.

1. WeChat fission, community operation, user subsidies, and group buying are all growth methods

Means have validity periods and environments. Their effectiveness is often based on overdrawing the industry's average success rate . After all, the replication of latecomers will accelerate the construction of the population's epidemic prevention capabilities , which will not only gradually become ineffective, but may also cause harm to themselves . In this track of imitation means, there is probably only first , no second .

There is a doggerel in the e-commerce industry that says:

"One sentence to promote user activation: push information and send coupons,

Send a text message whenever you want: You need to log in to get the gift, the quality product is on sale now, if you don’t come now, it will be gone~

If the customer doesn’t buy it, just hit them with the coupon.”

The consequence of this kind of mindless copying and following trends to achieve growth is that the costs are getting higher and higher, and the results are getting worse and worse. The users’ purchasing decision-making system is disrupted and the merchants’ pricing power is also questioned. “The prices are too high. I’ll buy when there’s a discount. There’s no rush anyway.” With more and more users trying to get the freebies, the platform has fallen into the embarrassing situation of drinking poison to quench thirst.

2. The correct approach for growth hackers should be data first

The advantage of data is that it can be objective and global. It can restore user scenarios and motivations through a set of indicators, and then summarize and deduce -> find differences -> seize growth points . To put it in a broader sense, data growth should also include a series of links in the entire chain, such as user positioning, product design, and pricing strategy. We will talk about this later when we have the chance.

At the same time, the effectiveness of growth projects also depends on two premises:

1. Data growth is an efficient organizational form that is independent of products, operations, technology, and brands. It breaks the conventional division of labor and business inertia and requires cross-departmental/cross-role linkage. The more efficient this linkage is, the better.
2. Precisely because it is independent of and coexists with the original division of labor system, collisions and integrations are inevitable. Therefore, the higher the management level that directly authorizes and is responsible for the growth team, the better.

The following figure is a growth case based on the PR draft of Taobao’s family account and a set of fake data.

As mentioned above, turnover = daily active purchase rate x purchase amount. Under the existing purchasing power, simply increasing a certain factor will not help to increase the total turnover, but what can be done is to restore the user's demand scenario through data , and then recreate the scenario to achieve stable and healthy growth of performance goals.

IV. Conclusion

Data analyst is a lonely circle. The loneliness lies in the inability to communicate and exchange ideas with peers. There is no value in general talk. In the end, it’s just a few points - trends/segmentation/comparison/multivariables. What is valuable is the attempts at various methods and the pitfalls encountered behind them. It is inevitable that business details are involved, so it is impossible to describe them in detail. I have to use a lot of other people’s cases. If there is anything unclear, please leave a message and comment for specific communication.

As for the prospects of data analysis, there is no need to do much publicity. I will just mention one thing. When traditional industries such as power banks, bicycles, coffee, and even cars and hypermarkets are gradually becoming Internet-based, it means that the Internet is moving from a light-asset era to a heavy-asset era. Do you think companies will still not pay attention to refined operations? If a product has a bug, we can roll it back in time, but the production and manufacturing of smart hardware has costs. If we produce too much and cannot sell it, it will become an inventory backlog. If we produce too little, users will not be able to buy it and the experience will be bad. As for the essence of new retail, it is to improve the turnover efficiency of people, goods and places. These are all data issues. The degree of refinement of efficient operations may gradually become a moat between future products.

Good gunmen are fed with bullets, and good analysts are educated through a large number of project practices .

The above picture is a summary and refinement based on the data and my own understanding - the three-layer value model of data application & the data personnel capacity development system. The author has been in this industry for eight years. Regardless of whether it is unilateral self-evaluation or evaluation from others obtained from the outside, the application of data on the Internet is still in the underestimated stage. Only by enduring loneliness can we maintain prosperity. On the road of data analysis, let us dream and move forward together.

Source: Laoqi