3 concepts for practical data analysis and decision-making!

The new graduates have a good understanding of the theory, but cannot apply it independently. What is missing?

In my past work experience, the deepest impression given by foreign companies is that they are "data-oriented". Reason and data are the bridges to cross cultural barriers and communicate on an equal footing. In domestic companies, "experience-oriented" is the mainstream, and the role of data is mainly to "assist in proving the ideas gained from experience."

This article does not discuss the pros and cons of both, but one thing is very obvious: when it comes to training new teams, rational and data-based methods are easier to use. Empiricism, on the other hand, is the so-called "learning from nature outside and finding inspiration from the heart inside", with vague guidance and self-understanding. It is okay to train artists in this "free-range" way, but it is obviously unreliable to use it to train Internet product operation teams with data awareness.

Some people might read this and say:

The new graduates have a good understanding of the theory, but cannot apply it independently. What is missing?

The knowledge of operations research and decision making in school is based on one assumption: the information is relatively sufficient and correct. What if it’s not quite enough, not quite right? Can I just complain?

This article believes that how to make decisions based on data when information is sufficient should be an "extreme exception" and that "how to make decisions based on data when information is insufficient" is the norm.

We know that the basic assumption of economics is that resources are "scarce" and economics is about how to "do more with less" when resources are insufficient. Data analysis decisions that can truly be used in actual combat should also be based on acknowledging and facing up to the fact that "information is not sufficient."

First idea: Try to quantify “insufficiency”

Common problems include incomplete data due to point-of-sale problems. To what extent is the data incomplete? Can it be quantified? Is there other data that can support the data? Based on these, a lot can be done using the existing data.

The second concept: Try to "accommodate misleading"

Neighboring departments/partners are unwilling to share sufficient information, resulting in a large difference between the forecast and actual data results? This is also a very common situation. There is no point in complaining. There are also a large number of natural disasters that affect the data. Compared with these, it is actually easier to predict how much deviation there is in the data provided by the neighboring department. After all, it has a "purpose", while natural disasters have no "purpose".

Most of the data with "purpose bias", such as event cheating and discount brushing, are objectively located in the middle of the two levels of "purpose" (for example, news from both Russia and Turkey are closer to the truth if they are summarized in the middle).

Black swans are accommodated when making predictions, and abnormal factors are taken into consideration. Data prediction, like programs, has the so-called "robustness" that can accommodate unexpected situations to a certain extent. The degree of "wiggle room" can be continuously approached to a reasonable value through comparative testing.

The third concept: Don’t make excuses for not thinking.

Why are comprehensive data analysis plans repeatedly rejected by superiors? A common inspirational quote is that it is because superiors see more information, so "the party with more information makes more scientific decisions than the party with less information." This is a typical "excuse-making" statement, which is not only quite negative, but also easily leads people to become lazy in thinking, gradually away from rationality, and devote themselves to the pursuit of power.

We mentioned before that there is no such thing as sufficient information in real life. It is more likely that the executive level has more information (which explains why "decentralized" Internet companies have a greater chance of success). Moreover, even the top level cannot predict new competitors and major changes in industry policies. If the data analysis has a unique approach, complete logic, and robust fault tolerance, it is entirely possible to "see the world through a grain of sand", which is also the most mentally enjoyable part of playing with data.

summary

This article was written in one breath without much polish. My original intention was that if I want to teach children to use data analysis to explain the world in the future, I don’t want to say something like “if the actual situation does not meet the hypothesis, then just adapt to the circumstances.” I feel that this would be the most unsuccessful father in the world. . . . . . .