Why do product promotion and advertising require calculation?

Regarding the topic of "Why does advertising need to be calculated", traditional advertisers and Internet people have different understandings and views. When we delve deeper into the significance of computational advertising, we find that: through computation, we can select the best personalized ads and achieve the highest input-output ratio.

There is a Dr. Li in Shanghai who proposed a general formula for predicting the comedy effect of crosstalk. He became an instant hit after teaching Guo Degang on Oriental TV. Regardless of whether the formula is reliable, this person really made everyone laugh: Crosstalk can be calculated using formulas, are you kidding?

Recently, I encountered such a thing. This semester, I opened a course called "Computational Advertising" at Communication University of China. I thought everyone would enthusiastically discuss algorithms and data. However, as soon as I got on the podium, the teacher at Communication University of China reminded me: I'm afraid you have to first explain why advertising requires computation?

Why should advertising be calculated? ! For Internet people, this seems to be the same as the question "Why do you take off clothes when taking a shower?". It has never been asked and never needs to be answered. Faced with such profound philosophical questions, I fell into deep thought, and when I woke up I suddenly realized: it turns out that there is a huge cognitive gap between advertising practitioners from different backgrounds.

For scholars, agencies and CMOs of traditional advertising, advertising is about surrounding the client with beautiful women and luxury cars in the afterglow of the sunset in Cannes, relying on imagination and insight to fool users all over the world into being proud to pay for it. It is something that shabby little shops on the street and shady programmers cannot covet.

For Internet engineers, product managers and CGOs: advertising is the black technology that drives the wealth engine of the new era. It uses artificial intelligence to print money like pie in the ground. Not to mention artists, all practitioners with advanced mathematics scores below 80 should be ashamed.

In order to bridge this gap and strive for the Nobel Peace Prize, I decided to use this article to give a brief explanation of the question "Why does advertising need to be calculated?" If you understand this article, your understanding of the principles of digital advertising should be better than 90% of advertising company CEOs.

Traditional advertising divides the execution process of advertising campaigns into several parts: planning, creativity, delivery and closing, such as the head, tail and body of the Trojan horse below. It is just “putting” the horse’s body, because traditional media can neither be attributed nor personalized, so it can only be treated as a black box without understanding it deeply.

With the advent of digital media, the secrets of Trojans have come to people's attention: how to select the best personalized advertisement through calculation every time it is displayed to significantly improve the effect is a new revolutionary problem. Therefore, in the Internet advertising circle, everyone began to focus on the inside of the Trojan horse.

Personalization means serving different ads to different users, that is, deciding what ads to place based on the user. This is already a cliché concept, but we would like to remind everyone that the basic principle of personalization is not so much "the right one wins" as "the highest price wins" , which is also the core calculation problem we will discuss.

From a business perspective, personalization can first allow users to see product promotions that they are more interested in, greatly improving conversion rates; secondly, it can finely segment traffic to serve a large number of small customers at the same time. What changes have these brought to the market?

The most important change is that the media no longer haggle with big clients, squeezing the urine out of them. Instead, they try their best to lower the entry threshold for clients, let them bid independently, and then allocate traffic based on the logic of the highest bidder wins.

In other words, the sales model of online advertising has changed from mainly negotiated contracts to mainly open bidding, which is crucial.

Under such business logic, the core goal of the advertising platform is no longer the budget, but the number of customers! Number of customers! Number of customers!

This is easy to understand: if there are enough customers, the price can be raised to a higher level, and the media can make money without doing anything. If you think about why major sales offices spend money to hire middle-aged women to queue up when a property is launched, you will have a direct understanding of this.

As a result, the digital advertising market structure has undergone a dramatic change compared to traditional advertising: in January 2019, Facebook’s advertisers reached seven million! This is probably an order of magnitude larger than the total number of advertisers in human history before the advent of the Internet. In addition, most of the revenue comes from the army of performance clients, while the budgets and voice of big brand clients are already a thing of the past.

CMOs of traditional large companies often ask me how to embrace digital advertising? In my opinion, the most important thing is to first understand such market changes: times have changed, small customers are the market leaders, and the media will no longer follow you around! That’s right, they even welcomed you from afar and treated you to four dishes and a soup, but that’s just a cover for the small customers to see!

Next, let’s take a look at the basic calculation logic of the auction model where the highest bidder wins.

1. Basic calculation logic and division of labor

What is the high bidder? It is the ad that brings the greatest profit to this display. Because the decision must be made in an instant before the ad is launched, the display revenue is estimated and called the "expected cost per mille (eCPM)". As the revenue for a single display is too small, it is customary to multiply it by one thousand for convenience. When we talk about computational advertising, the core is to calculate the eCPM and find the highest bidder.

How is eCPM calculated specifically? If we break down the process by which advertising produces its effects, it will become clear.

As shown in the figure below, some users will click on the ad after seeing it. This proportion is the click-through rate μ . Users who click the ad may generate conversions, bringing the average click revenue v to the customer. Obviously, as long as you calculate these two quantities separately, multiplying them together is eCPM.

Who will make the estimate?

Generally speaking, since the click data is on the media and the conversion data is on the customer's side, the most reasonable division of labor is for the media to estimate the click rate and for the customer to estimate the click revenue.

Correspondingly, many of the mainstream computational advertising platforms we see are charged on a cost per click (Cost per Click, CPC) basis.

There is a new question here. The relationship between clients and media is a game. They are not obliged to report the actual click revenue, but they bid according to what is beneficial. The question is: How to guide customers and get them to tell the truth as much as possible? Otherwise, the eCPM calculated by multiplication will be wrong!

This question has long been answered in game theory: if the auction adopts a two-price system, that is, the first place winner is charged the same price as the second place bid, then when the Nash equilibrium is reached, the bidders will bid according to their own benefits, otherwise it will be disadvantageous to themselves.

If there is a group of positions being auctioned together, the second price can be generalized to the more general VCG pricing, and the details will not be elaborated. The two-price and VCG mechanism is the mainstay for maintaining the stability and transparency of the bidding advertising system and maximizing social value.

The customer has given click revenue, so it’s time to calculate the click rate. What factors determine the click-through rate of a particular ad display?

First, users. Different users have different interests and needs, so different ads are suitable for them.

The second is the scenario, that is, what the user is doing at the moment. Irrelevant advertising will naturally have a poor effect. In other words, click-through rate is a function of advertising, users, and scenarios.

The two factors in the above formula are user ( u ) and scenario ( c ). Knowing only their ID is useless: what you want is not that his name is Zhang San, but what he is interested in. Therefore, quantitatively describing the characteristics of users and scenarios has given rise to another computational problem - audience targeting, or user profiling, which requires some user data.

Audience targeting is to infer a user’s interests and intensity based on his behavioral data in various scenarios. Relying solely on the data generated by the advertising system itself is not enough to effectively characterize users. We often need more valuable data assets such as search, shopping, and application lists. With the user's interest tags, they can be brought into the click-through rate model to more accurately determine his feedback on a certain advertisement.

Of course, this also brings serious problems - the wanton collection and abuse of user data.

Data has created the prosperity of the Internet, but also brought threats to user privacy. How to process data and find a technical balance between personalized results and user privacy is a topic called Differential Privacy. However, I personally believe that the indiscriminate GDPR standards like those of the EU are a case of throwing the baby out with the bathwater. If they are strictly implemented, they will completely destroy Europe's Internet industry.

At this point, we have figured out the typical decision-making logic of computational advertising: calculate the eCPM of each candidate ad based on the current user and scenario, and select the highest one for display. Among them, the media calculates the click-through rate based on the user tag, and then multiplies it by the click revenue provided by the customer to obtain the eCPM. In addition, in order to encourage customers to bid truthfully, a two-price or VCG mechanism should be adopted.

In addition to this typical logic, there are two product trends: one is the expansion of demand-side calculations downstream, which is for programmatic; the other is the expansion of supply-side calculations upstream, which is for oCPM/CPA.

2. Open computing to customers: programmatic

If advertising decisions are further opened up to customers, they can use their own data and algorithms. For example, if the client knows his or her members very well and can calculate their click-through rate and click revenue more accurately, then just let the client do the calculation and then report the eCPM to the media.

Don’t forget that advertising decisions must be made for every display, so customers must set up servers and use programs instead of people to make placements. This is programmatic advertising. Since the eCPM estimation is done by the client, programmatic advertising is generally settled on a CPM basis.

Google is a representative of programmatic advocates. Its full range of programmatic product lines are widely emulated by other large media. For large clients like JD.com and Toutiao, which have abundant data assets and superior technical capabilities, using programmatic methods to access media platforms and purchase traffic has become a standard approach. It can be said that the prosperity of programmatic advertising has greatly promoted the use and transaction of first-party (customer) and third-party data.

However, programmatic advertising has a natural problem, which is that only big customers can afford it, which is completely contrary to the principle of customer quantity first mentioned above. As you can imagine, this expands the competitive advantage of large customers, but it helps little with the number of customers, so the media does not necessarily benefit much.

You can think about it, why is search advertising not programmatic? Once you understand this, you will have a clear idea of ​​the advantages and disadvantages of programming. Of course, in China, as long as bidding is used in the background for advertising that requires customers to manually place orders, such as search, it is also vulgarly called programmatic advertising by many people. This statement is completely wrong. He should be given a thousand four hundred slaps in the face by having his socks pulled off and slapped left and right.

3. All calculated by the media: oCPM/CPA

Contrary to the open idea, you can also leave the calculation of click-through rate and click revenue to the media, and settle with customers directly based on conversion (Cost per Action, CPA) or sales (Cost per Sale, CPS). In fact, media oligarchs who have traffic and data hegemony are likely to do better than their clients in estimating click revenue - provided that the clients provide the conversion data to the media.

This relatively closed product approach is represented by Facebook: instead of opening programmatic interfaces to customers, they want to take over even the calculation of click revenue. Don't forget that Facebook has a wealth of information about users' behaviors and interests, so it would be strange if it couldn't accurately calculate its customers.

The problem is that the conversion process is varied and not very controllable. For example, if your website is often down, how can I settle accounts with you based on the results? In response to this, Facebook creatively adopted the oCPM (optimized CPM) solution: settlement based on CPM, but optimized based on CPA. Once you are sure the data is feasible, you can completely take over and become a CPA.

Small and medium-sized customers are looking forward to such a solution, because they no longer need to optimize advertising, but only set goals and wait for results. As for the testing costs paid by oCPM, they are limited and controllable. Therefore, under the oCPM/CPA product mechanism, the cost for small customers to enter the market is reduced to an extremely low level.

Final Words

Here are some words of hindsight:

The latter product direction seems to be more successful, and it is also more in line with our philosophy of customer number first. Judging from the results, Facebook's customer base has exploded, and it has surpassed Google. Of course, product differentiation is not the only determining factor. In addition, these two product directions are not contradictory and can be adopted separately for different customer groups.

This is a summary of the secret hidden in the belly of the Trojan Horse. I hope this article will help you quickly understand the mysteries of modern online advertising "computation". If we continue writing this much nonsense and unfunny content, readers will probably curse us, so we’ll stop here.

Related reading:

1. Which advertising copywriting materials have timeless creativity?

2. How to write search advertising promotion creatives?

Author: Bei Ming Cheng Hai Sheng

Source: Computational Advertising (ID: Comp_Ad)