And look at the big data and how the experts "bullying"

This is a home electric business industry very professional micro-letter group, I was fortunate in the group sometimes dozen soy sauce. It was a big data expert with soy sauce on the day, so the following conversation was made. The dialogue is a complete truth record, without making a word change. When it comes to real names, replace with "X".

My annotations are in parentheses in the dialog. Also note the delay of the micro-letter dialogue, so it is not strictly a sentence.

The story was uploaded from a group of micro-letters about how an American furniture dealer used large data to begin with.

When we open an E-commerce site, there are tens of thousands of products waiting for us to choose, that "the flowers are increasingly attractive eyes"? Of course, you can use the "Pastoral wind" or "children's bed" such as the label to screen. But in the eyes of the American furniture electric Trader Smart Furniture, this is not cool enough, the process should be one step ahead, and then simplify: Before users start browsing according to the user's characteristics and preferences, customized for the user to display the product may want to buy. It's like before you go into a furniture store, all the furniture that matches your taste is lined up for you.

and to achieve this effect, customers just go to the site to start shopping before a simple online test. Smart Furniture's Furniture experts have passed an algorithmic rule that matches the different test results with tens of thousands of product features, and what remains to be done is to select the products that impress them in the customized Web page. As a result, customers who use this feature have a 10 times-fold increase in turnover compared to unused customers.

The process of online testing is also very simple, and each simple question is accompanied by a picture that assists you in making your choice, which is more intuitive and convenient.

1 of course is to determine your gender.

2 Determine your preferred style of furniture, simple? European？ or modern?

3 can accommodate the size of the furniture space, spacious? Compact？ or general?

4 to determine the decoration style of the room.

5 to determine the trend of shopping, pay more attention to quality or pay more attention to the price or the balance of the two?

6 Choose personal leisure habits (often stay in the restaurant, bedroom or home office space).

7) huxing characteristics (spacious, comfortable compact or general).

8 The most attention to which point (price, quality or both?) ）。

9 like what kind of office chair (comfort, price, appearance).

This new feature, called "Smartprofile", is known as "standardized customization," and it matches each person's character, preferences, home space, budget and other features to show everyone "customized" merchandise. At the end of 2013, the company launched the beta, the average user rate during the test increased by 27%, so the company decided to provide all users with the use. and smart Furniture Company's goal is to continue to build this smartprofile for customers to create a more intuitive shopping experience.

Expert: Custom furniture is the same as the mobile application of the promotional treasure I made for XX lighting. (The experts appeared the first time.) ）

Me: This concept of personalized marketing has been attracting a lot of people. But the mysteries and pain points are rarely talked about. It looks like this smart furniture is another case. (I'm only talking about the case, not the experts, because I didn't quite understand what the expert said about the custom furniture and the case.) Custom furniture is a concept, personalized marketing is a different concept. The "customized" product in the story is a personalized product, not a custom piece of furniture we usually call it. ）

Me: Do not be infatuated with this so-called personalized marketing of large data applications, often do not have good results. (I still speak from the word.) ）

Me: The essence of the Internet is interaction. Personalization in this sense is actually anti-internet thinking. (Chatterbox!) An "interaction" that leads to an expert's discipline. ）

Expert: limited interaction.

Experts: They cannot interact with each other.

Me: This should be said in turn, not personalized. Interaction is essential. (From here I just talk to the experts.) It's a bit of an overreach. ）

Experts: Chatting is interactive, talking about business is the ability.

Me: Chatting is just a way of interacting.

Experts: Dating is the way to push down is the purpose. (experts usually talk more esoteric.) ）

I: To explain this to you, the Internet is to consumers to master the initiative. This is the spirit of the Internet. And the so-called personalization, still is the businessman wants to grasp the initiative. (This is the right attitude to talk to experts.) Self review. Oh。 ）

Me: This is Internet thinking. I also speak of the internet thinking, ashamed. ）

Experts: Customer experience is the only standard, others are only efficiency, so the offline execution is the core competitiveness. (Experts are experts!) ）

Me: No sermon. If you have a personalized system, can you tell me what the accuracy rate is for predicting user consumption requirements? I am a real person. In my opinion, the customer experience is to be explained by the specific figures. ）

Expert: That's what we're doing. Only through online data matching, we can achieve 40% accuracy, and then the store human intervention. (An expert indeed!) There are numbers! ）

Me: How many people can reach 40%? (I want to know what the sample space is for this predictive accuracy.) For example, 100 people on the Internet station, 10 people use this prediction system, 40% only 4 people. ）

Expert: Interesting? (from the context, this sentence should have been asked before my previous sentence.) But I was stupid enough to hear the contempt of the experts. ）

Me: You tell the case completely to know whether it is interesting or not. I am a real person. ）

Expert: noise and signal competition. (experts are the export of extraordinary.) Do you understand? I don't understand. ）

Me: To be honest. Or your users can understand, or the statisticians can understand. (I am anxious!) Experts and I speak certainly not on a level, I really do not understand. ）

Me: 100 Customers on your website, can you predict 40%? Is it too divine? (If you do not specify the sample space, it should be understood as a full sample.) That would be really magical, wouldn't it? ）

Experts: Effective communication can extract signals, unlimited communication to generate noise. To extract the signal from the noise, the accuracy is measured by the signal, not from the noise extraction signal. (Expert!!! ）

Expert: common sense. (Beginning to Blush!) This common sense how do not understand, what face to see Jiangdong elders?! ）

Expert: let you so? (experts, please don't be sarcastic.) I'm going to get mixed up in this group. Please! ）

Me: Can you say something we can read? The common sense you say may be really no one can understand. (I am a real person, still so winkle.) ）

Expert: "Noise and Signal", a book about large data. (Big data, you know?!) Alas, big data presses the dead. ）

Me: So, what statistical principles do you use to make predictions? (I am so real!) Why are you still asking about statistics? That's big data! ）

Experts: "Signal and noise" in the following interesting and worthy of study: The weather forecast said the probability of precipitation 60%, you go out with an umbrella? How much is the probability of being struck by lightning? Are we really unpredictable before earthquakes happen? Why would the CIA ignore the signal of the "9.11" terrorist attack? Why did the outbreak of avian flu suddenly disappear? Why are predictions of the big data age more likely to fail?

"Signal" is the fact that we want and need, such as the signal that can help us to detect the early shoe bomb case. "Noise" is another thing, usually irrelevant information that hinders or misleads us into searching for a signal. (&......%#@--() &%&%*￥#@&)

Expert: Recommended reading.

Me: It's too deep! (a heartfelt sigh.) ）

Me: What statistical methods do you use to make predictions? I can get the most stubborn award of the year. ）

(Long no answer.) ）

Me: You can make predictions without counting? (I am striving for International Stubborn award!) ）

Expert: You rest, you can see me as a fortune teller, sorry ah you. (The expert was angry indeed.) The experts are unhappy and the consequences are serious. ）

Experts: In the group of cases, only a few signatures cross matching can get the proposal, toss so complicated why? (not big data?) Not the noise and the signal? How did it suddenly become "several signature cross match"?! Is it not a statistical method? Are statistical methods more complex than noise and signal learning? I was completely confused. ）

Expert: (Provide a two-dimensional code) be interested in sweeping this off. (still very expert.) ）

(Fortunately at this time the group has friends out Yuanchang, the dialogue to end.) Think back to really scared, if I do not understand a little statistical analysis, in case I do not know a little bit of data, not only by the large data experts to be blindfolded, but also publicly humiliated. ）

This case, and the above dialogue, should be a typical big data flicker.

I'll just talk about the big data in the case.

It should be a typical case for personalized, large data for large data.

First, we consider why large data is needed and why it is personalized. The answer is to facilitate the user, in order to enhance the user experience. However, for this purpose, Smart furniture to trouble users to answer 9 questions first. At least nine clicks (you should actually add another two times: start and end) to provide the so-called personalized goods. This is at the expense of the user experience to pursue the so-called personalization.

If your Web page is optimized, will users need nine clicks to find the product they need? If the connection between your online products is done, when a user finds his or her favorite first item, you can provide the user with the other items that he or she might like through relevance.

This is the essence of large data applications. Roughly speaking, the smart furniture's predictive system is basically taking off the pants and pi.

This is a vulgar remark. Of course James is a metaphor. Have you ever heard of "Da Zhai Nong"? The people of Da Zhai and fighting with the sky, overcome the difficulties, obstinately in the Hutoushan on the layer of terraced fields. But how do people on the plains learn from the DA Zhai? Some people on the ground to build artificial terraces, and then carry water on the terrace irrigation, fully demonstrated the working people's hard-working spirit.

In this case, there are data to be large data, no data collection data also need to enlarge the spirit of the data and attitude, and flat terrace to learn the same way.

Second, the predictive system uses a relatively complex statistical prediction method, rather than what the expert above has just said, "several signature crossover matches." Using the nine variables selected by the prediction system to predict the user's preference for various commodities is a very complex model analysis, and the accuracy of prediction is very limited.

From the nine questions described in the story, it seems that the large data experts who operate this predictive model are not very scientific, because the nine questions they ask are not systematic, and the question of the five and eighth questions seems to have been repeated. Perhaps this is the mistake of the person who wrote the story.

I've always felt that people who preach or use these predictions generally don't know much about statistics, otherwise they won't be so bold as to dare to do such a predictive model. May be the so-called ignorant people fearless. Of course they're a little bit stronger than those who basically don't know how to use statistics to get big numbers and stuff all day.

Again, a key factor in measuring the effectiveness of the system is how much of the user is using this predictive system. Even if your prediction is 100% accurate, if only a few users actually use your predictive system, your predictive system may not be in vain.

Several data are provided in the article to illustrate the effectiveness of the system. Let's take a closer look at the data to see if they look so effective. "At the end of 2013, the company launched the beta version, the average user rate during the test increased by 27%." "Customers who use this feature have a 10 times-fold increase in turnover compared to unused customers." ”

It seems to work. But obviously, in order to understand what these two data really mean, we need to divide the site users into users who use the predictive system and those who do not. Again, the ratio is the key to measuring whether the system is really effective. But that number did not appear. According to past experience, the numbers that exist but are not provided are often deliberately not provided.

"The average number of users under the test has increased by 27% per cent," the significance of which is ambiguous. Is the average per-user rate up by 27%, or is the average percentage of users using the forecast system growing by 27% per cent? We are again doubtful about this ambiguous expression in this article, because the article has a sufficient motive to illustrate the validity of the system with the most accurate expression. If the average rate of growth for all users is lower, it may not be related to the predictive system. If the user is using the system under the single rate of growth of 27%, why not directly explain it?!

Also, during the test, more data may be needed to optimize the predictive algorithm. This growth of 27% may come from the optimization of the predictive system. This only means that the original prediction system is not good enough, but it does not fully explain the effectiveness of the system.

Let's take a look at why users are not afraid to bother using this predictive system because these users have a strong desire to shop on your website! In other words, the story of the so-called 10 times-fold increase in turnover may need to be understood in turn: it is because users have a strong desire to deal with the users will use your predictive system. Therefore, whether your system is really effective can not be explained by this 10 times-fold increase in turnover rate.

More importantly, when you focus your energy on this fancy big data prediction system, you are likely to overlook the rational optimization of the entire site, resulting in a decline in the efficiency of the entire site.

From this case we can see that we cannot blindly pursue large data and personalization. Big data or personalization is not our goal. Our goal is to enhance the user experience. If you need to sacrifice the user experience to achieve what large data or personalization, it is in bullying.

Since the development of the Internet, many people have seen the potential prospect of data application. Before this development is big data, appear such as "one-to-one marketing" pursue individual marketing trend of thought. The current large data application of personalized marketing, basically inherited more than 10 years ago, "one-to-one marketing" thinking. But the theoretical basis of this marketing thinking is wrong, and from the perspective of the method is also wrong. Ten years ago I was in a university in Shanghai and marketing Department of a group of Bo and doctoral students to discuss "one-on-one marketing" problem, I explained to them that the so-called "one-to-one marketing" is actually through the statistical method of market segmentation to achieve.

I've seen a lot of big data applications like this one, and a lot more fascinating than this case. But almost no one is successful at last. Because they are anti-scientific, but also the user experience.

Tang people, home electric business practitioners and explorers, focused on the traditional household enterprises in the development of electric business strategy and implementation. More articles, please Baidu "Chinatown home Electric Dealer."