Is big data accurate advertising reliable?
Source: Shangpin China |
Type: website encyclopedia |
Time: November 27, 2015
1、 Big data precise advertising connotation At present, big data has become one of the hottest words in the entire IT industry (including Internet Technology and Information Technology). It seems that any topic, as long as big data is mentioned, will become big. Overnight, big data has replaced subjective rational thinking and become a synonym for intelligent insight.
.jpg)
But when we walk right Beijing website production In the worship phase of big data, the veil of big data's practical application was unveiled, and instead, it gradually reflected on the big data filled with the discourse world. Because the great significance of big data in economic development does not mean that it can replace all rational thinking about social issues, and the logic of scientific development cannot be buried in the massive data. The famous economist Ludwig von Mises once warned: "As far as today is concerned, many people are so busy with the useless accumulation of data that they lose their understanding of the special economic significance of explaining and solving problems.
Taking the advertising application of big data as an example, accurate advertising should be the earliest application of big data and the one that is most likely to generate direct income. Nowadays, few advertising companies do not claim to be big data technology companies. What is the core connotation of big data precision advertising? In a word, it is procedural targeted delivery. Among them, orientation is the core and routinization is the means.
Take WeChat friends circle as an example. For non targeted regions, the open price at the beginning of the year is RMB 40 (per thousand exposures, the advertising price of friends circle is much higher than that of general media), RMB 140 for targeted core cities, and RMB 90 for targeted key cities. If targeted gender is superimposed, 10% will be added, and then H5 external chain (better traffic guidance effect), and 20% will be added. Just like imported cosmetics, tariffs are levied in a certain proportion first, then VAT is levied at the price including tax, and then consumption tax is levied at the price including tax.
For WeChat, although the customer's region and gender also need data analysis and interpretation, it is relatively easy to confirm. For other data companies, the region can still be obtained through IP or mobile terminal GPS, but gender is more likely to be a possible attribute of data analysis. Of course, big data does not only analyze such simple tags. For media alliances, there are many media selection projects, but also analyze customer media preference tags, as well as multiple directional combinations such as time periods, crowd attributes, device types, preference types, etc.
Well, the above gives a brief introduction to precision advertising. What kind of value can big data precision advertising bring? The following story is often mentioned by big data advertising companies.
If 10000 people visit the advertising space of a website every hour, the exposure per hour will be 10000, and the previous CPM will be 5 yuan, then a mobile advertiser will spend 50 yuan on advertising for one hour. This is the result of traditional advertising. Now there is a big data company to help the advertising media operate better. The company claims that it can accurately identify the attributes of browsing customers, and tells mobile advertisers that although 10000 people browse the ad space, only 6000 people are really suitable for launching mobile phones, and the remaining 4000 people are invalid exposure, because the remaining people are only interested in clothes.
Big data companies suggest advertisers to filter out 4000 people who are not suitable for launching mobile phones and pay only 6000 people who are suitable for launching mobile phones. If the unit price remains the same, the cost will be reduced to 30 yuan on the premise of ensuring the same effect. The remaining 4000 people's data companies sold them to clothing advertisers at a cost of 20 yuan. Thus, under the same effect, big data advertising significantly reduces the cost of advertisers. Of course, due to the existence of the RTB (real-time bidding) mechanism, when the price (the effect is the same) is low to a certain extent, different mobile advertisers compete with each other, making the real price generally higher than 30 yuan, but certainly between 30 yuan and the original expected cost of 50 yuan, thus forming an ideal situation of mutual benefits.
Such cases look perfect and impeccable. Because it solves the inefficiency problem of traditional advertising, such as looking useful, but not knowing exactly where it is useful, which is hated by CFOs of companies. Yes, through big data advertising, we can make all the money spent on advertising more reliable. We can online assess how many impressions an advertisement creates, or even how many clicks, how many downloads and uses it, and how many transactions it generates.
Any problems? That's all right. Any problems? What do you mean, do you want to doubt the truth?
A "House of Cards" has made the cultural industry around the world aware of the power of big data. Netflix, the producer and broadcast platform of House of Cards, added more than 3 million streaming media users in the first quarter. After the first quarter's financial report was released, the stock price soared 26% to $217 per share, more than three times the low price in August last year. All this stems from the birth of House of Cards, which summarizes the viewing habits from the data of 30 million paying users and creates based on accurate analysis of user preferences.
The database of House of Cards contains 30 million user ratings, 4 million comments, and 3 million topic searches. Ultimately, what to shoot, who will shoot, who will act, and how to broadcast are all determined by the objective preferences of tens of millions of viewers. From audience insight, audience positioning, audience contact to audience transformation, each step is guided by accurate, meticulous, efficient and economic data, so as to achieve the C2B created by the public, that is, production is determined by user needs.
Today, the development of the Internet and social media makes people leave more and more data on the network. Massive data, through multi-dimensional information restructuring, makes enterprises seek to fully connect the content, users, and advertising of various platforms, with a view to integrating user relationship chains and social restructuring of online media, Bring better precision social marketing effect to advertising users.
2、 Big data precision advertising is not as beautiful as it seems
Based on the principle of falsification, truth is only true when it is proved to be false (understand the limitations and conditions of its application). So we must first answer the question, what is advertising used for?
According to the previous consensus, advertising is regarded as a brand used to convey brand characteristics to consumers who cannot communicate face to face. Therefore, although advertising can promote sales, usually the content of advertising does not directly persuade consumers to buy, just like the award-winning advertising of China Mobile, "communication starts from the heart". In the book How Brands Grow published in 2010 (which means that the author has not read it and hopes to read it in the future), Professor Byron of the University of South Australia pointed out in the book that in order to achieve the best effect, advertising often does not need to be persuaded or indoctrinated, just to remind people of the brand name when buying. Gordon Brown, founder of market research organization Milward Brown, pointed out that the function of advertising is to make a brand on the shelf "interesting".
Well, back to the case of big data precision advertising, one of the most critical questions is how does big data analyze that these 6000 browsing users are suitable for mobile advertising? For this question, advertising companies have been prepared to give the following answers.
First, find customers who have used similar products from the history for matching. The commonly used algorithm is called "collaborative filtering", that is, finding potential suitable users based on the relevance of some experiences. For example, you have played a certain game, so you can think that you have the same demand for other games of this type. The author does not deny that the algorithm does work in some areas, for example, game paying users are basically heavy game users before.
However, despite these special areas, the algorithm's connotation idea of "brand growth depends on loyal consumers" is completely contradictory to Byron's theory. Through statistical analysis of sales data, Byron pointed out that among all successful brands, a large number of sales come from "light buyers": that is, customers who buy products relatively infrequently. Coca Cola's business does not depend on people who drink Coke every day, but millions of customers who drink it once or twice a year. This consumer model is applicable to all brands and commodity categories in countries and periods. Whether it's toothbrushes or computers, French cars or Australian banks, brands rely on large populations - in other words, Volkswagen - who occasionally buy them.
This theory has far-reaching significance. This means that you can never increase the brand's market share by precision marketing existing customers. Precision marketing for existing customers is what digital media is good at.
In the spirit of criticism, the author looks at the new and untested ideas, and hopes to quote the user change feature data recently released by Guangdong Mobile. Guangdong Mobile's terminal migration analysis of its users shows that 64% of users who use Apple's upgraded terminals continue to use Apple's mobile phones, with the highest loyalty. However, except for Apple, less than 30% of Huawei and Xiaomi mobile phones with the best loyalty continue to use the same brand after replacing 4G.
This shows that it is feasible for you to promote Apple 6 to Apple 4 or 5 users. The Apple Powder Effect overthrows Byron's theory and proves that it is feasible to grow and develop depending on loyal consumers in some areas. But in addition, it is inappropriate for you to promote mobile terminals of the same brand to any user of the current brand.
Therefore, it is hoped that it is invalid to infer the user's next need through the analysis of historical e-commerce data. For example, it may be more effective to promote clothing to users who have bought clothing than to promote a roll of paper or a barrel of oil.
On the contrary, successful brands need to find a way to reach groups outside the target market. Brand advertising must capture the interest of these people in some way - only in this way can the brand automatically appear in the minds of consumers when they are preparing to purchase.
Second, if the "collaborative filtering" is limited, the advertising company will tell you that there is a second algorithm, which is not based on the historical behavior records of customers, but the similarity of customers' own characteristics, to find the customer group most similar to the seed customers. It is called "Lookalike" for short. First, advertisers need to provide typical users who play a role in this advertisement. Take mobile phones as an example, there are about hundreds or thousands of users who are interested in clicking to browse or pre order a mobile phone affected by the advertisement. Big data companies use the Lookalike algorithm (the professional term is more likely to be sparse matrix) to find hundreds of thousands/millions of other customer groups that are highly similar to these hundreds/thousands of users.
This kind of algorithm really tests the computing power of big data platforms, because it is not empirical collaborative filtering, but regression calculation using hundreds or even thousands of variables. Finally, according to the probability of similarity, select the appropriate user group from high to low.
In fact, the connotation of this model is very simple, that is, the advertisement should be conveyed to the customers who should be conveyed. For example, the target users of milk powder advertising are the parents who raise children aged 0-3. If you know the specific identity of the user to be reached, all problems will be solved. However, for websites or APP applications, the only thing that is clear is the historical behavior data of customers. Moreover, due to the segmentation of the data itself, some focus on operators, some focus on APP alliance collection, some focus on e-commerce, and some focus on banks. Lookalike is an inevitable means to infer the customer's identity information from the segmented data.
The only problem is that if hundreds of new target users are inferred from hundreds of seed users, the accuracy may be as high as 90%, but if an advertising company claims that when launching mobile DSP for Tingyi Spicy Noodles, it will analyze and mine the historical data to form a sample database, and then enlarge the population through Lookalike technology, Find potential customers with the highest similarity to the target audience, expand the population by 13.67 million, and actually launch 20.89 million audience IDs. The advertising effect is maximized. What about the effect? Here, please allow me to make up a number. It is likely that the click rate will increase from 0.2% to 0.3%, and the accuracy will increase by 50%. Does it make sense? Maybe, but definitely not as obvious as expected.
Third, if you continue to doubt the effectiveness of our algorithm, we can talk about cooperation based on the effect. You can pay according to the number of clicks (CPC) or the number of activations (CPA). If the desired effect is not achieved, we will make up. This is the ultimate weapon of big data advertising.
As soon as the ultimate weapon is released, it means that the advertising is completely reduced to a click and activation channel, and the original intention of "communicating with consumers" of advertising has long been abandoned completely.
Generally, general consumption decisions follow the rules of S (Solution), I (Information), V (Value), and A (Access), which means that when a user generates a demand, he or she first meets the demand to form a solution. For example, 3G mobile phones are not easy to use, and the speed is very slow and the coverage is not good. If you need to replace a 4G terminal, it will become a solution. What are the 4G terminals, and which ones should be considered? Consumers still collect information, not from online search, but from past experience, brand effect and reputation of surrounding friends to automatically recall which brands and styles. The most important function of traditional advertising should be this stage, when the user needs it, it will automatically enter the user's view. Then compare and select from multiple dimensions to determine the preferred brand to purchase. Finally, where to buy, where to search for promotional activities, where the greatest incentives.
According to the SIVA model, the essence of the real effect oriented advertising is to solve the access problem, and the last step is to take the door. In this regard, search advertising is really effect oriented advertising. For example, there are more than 10000 merchants behind each product on Taobao. Users have to pay for advertising where they buy it. This is effect advertising. It has been reported that the click rate of search ads is up to more than 40%. Think about what Baidu and Alibaba depend on for a living. The advertising price at the doorstep is naturally too high. It is said that some hospitals purchase Baidu sexually transmitted diseases, people flow and other search ads, and the price of a single flow is as high as tens or hundreds of yuan.
Only a few monopoly access companies have business in search advertising, most of which are still display advertising. If the display ads also move towards the effect ads, the business law will put the cart before the horse.
The final result is that, on the one hand, the content of the advertisement is full of human greed (preferential/cheap) and lust (big breasted beauty), which has been transformed into neither fish nor fowl. After the last time, it has played a negative role in communicating with consumers. On the other hand, the advertising company has become a traffic and click company, which has no essential difference from the famous traffic streets of Wangjing and Zhongguancun in Beijing. Who actually clicked on these effective traffic? A big data company once analyzed the phased user group of a high-end financial software, which is highly similar to the very low-end users who want to use their computers and seek small and cheap things.
The divination and prediction function of big data technology means that if a social network platform with hundreds of millions of users can provide personalized and intelligent advertising push and service promotion services for enterprises through the deconstruction of big data, it means that enterprises can seize a larger business space.
With the rise of social media, consumers' dependence on advertising behavior has changed, and traditional advertising and marketing techniques are actually more difficult to work. "It's hard for those who are doing marketing in this era to understand consumers without understanding the concept of mobility. Fragmented consumption scenarios have changed physical stores," said Liu Shengyi, president of Tencent's online media business group and senior executive vice president of the group.
In this regard, Han Meirui, vice president of Starbucks China Marketing Department, believes that social media can help enterprises interact well with consumers, and also make the whole marketing more accurate. In Han Meirui's view, Starbucks has no as wide a range of channels as Coca Cola, so the publicity must be more targeted, The deconstruction of big data by social media can solve this problem.
Therefore, when Starbucks already has a large number of offline users, it does not take the increase of new customers as the first starting point to carry out social marketing. Instead, it focuses on maintaining old customers and realizes the growth of new customers through the praise of old customers. Because in the consumer decision-making chain, marketing driven by consumers themselves becomes more and more important.
Nowadays, the channels and scope for consumers to obtain information have greatly increased. They are no longer at the mercy of enterprises, but pursue more personalized products and services, and make judgments and share at any time according to various information collected, so as to expand the impact of personal experience to a wider range of groups.
In the era of social media, big data is still the engine and the platform for users to constantly transform. Accordingly, marketing has changed from independent to systematic engineering, and the role of data in the whole marketing process must also change from reference tools to driving engines. The data driven precision marketing engine will subvert the traditional marketing decision-making mode and marketing implementation process, and bring revolutionary impact to the network marketing industry and even the traditional industry.
Each marketing will form a circular effect. By positioning user groups, analyzing user content preferences, analyzing user behavior preferences, establishing audience segmentation models, formulating channel and creative strategies, trial launching and collecting data, optimizing and determining channel and creativity, officially launching and collecting data, adjusting launching strategies in real time, and completing launching evaluation effects, the complete data application process continuously controls marketing quality and effects, Realize the transformation from effect monitoring to effect prediction.
"Although social media makes the whole advertising marketing more precise, it is also necessary to decide whether to adopt precision marketing according to the characteristics of products and services." Zheng Jingwei said that some FMCG are not suitable for precision marketing, and traditional media such as outdoor, television and newspapers still have a strong attraction for FMCG.
It is worth noting that the deconstruction of big data by social media inevitably brings privacy problems. When users use email and social networks, they probably know that their information will be recorded. When users make comments or share photos, videos, etc., they decide what resources and advertisements Internet operators will recommend to you; When users run around the world with smart phones, mobile phone manufacturers have already collected all your information in their own databases through positioning systems, and used this information to build maps and traffic information.
Previously, these records would hardly affect ordinary people, because their number was so huge that people would not pay attention to some information unless they were deliberately looking for it. However, with the continuous progress of big data technology, this situation is quietly changing. This is also a challenge for enterprises and consumers in the era of "number" change.
3、 Multi use of reliable identity recognition may be more conducive to improving advertising effect
Having written so much, is big data accurate advertising useless? No, to doubt the truth is to apply it better. There is no mistake in the core "programing" and "targeted delivery" of big data advertising, which represents the trend of mobile Internet development, and also fully matches the demand of commodity or service advertising dissemination to meet specific markets and user groups. The problem is that there is still a huge gap between the actual capacity of big data and the declared ambition. In other words, it is not as good as it looks.
Therefore, we should return to the original purpose of advertising - to better communicate with consumers, to look at precision launch, rather than superstitious big data precision launch such gimmicks. So what's the most important? Obviously, it is not an unreliable collaborative filtering rule, nor is it a Lookalike without knowing the reason. Since the most important thing is to reach the target consumers, reliable identity recognition should be the core of accurate advertising.
What is reliable identification? For WeChat, it is reliable to judge key activity cities and analyze gender. But if WeChat tells you that you can judge whether the user is a middle class white-collar worker or a rural farmer through social interaction, it must be unreliable. Because an elegant woman who claimed to be traveling in a French winery in her circle of friends might be going out to buy fried dough sticks and soybean milk.
Sometimes the media used by the user itself reveals the identity of the customer. For example, those who often use financial software are more reliable in terms of payment ability, while more than 80% of those who use Pregnancy APP should be expectant mothers, and those who often use honey sprouts must be mothers whose babies are not yet born. Some big data companies have given examples, and the effect of media orientation and comprehensive analysis orientation is almost the same, which shows that media orientation is effective, but other demand orientations are equivalent to random selection.
Because big data itself does not focus on cause and effect, but only on correlation, the collaborative rules confirmed by big data insight can also be regarded as reliable rules. For example, the group of game paying users can basically be identified as heavy users with 10 or 20 million IDs.
To accurately identify customer identity, the collection and integration of multiple data sources is inevitable. All kinds of insights and correlation analysis around customer identity are also required courses for capacity improvement, which may be the core competence that big data advertising companies should continue to cultivate.
Source Statement: This article is original or edited by Shangpin China's editors. If it needs to be reproduced, please indicate that it is from Shangpin China. The above contents (including pictures and words) are from the Internet. If there is any infringement, please contact us in time (010-60259772).
