- Security
- A
"Element of Selling". Interview with an Internet Advertising Specialist
Immediately after the laws on changing the laws on liability for improper handling of personal data were passed, we at IDX, like everyone who deals with personal data, pulled our heads into our shoulders and pressed our ears.
After catching our breath a little, we stuck our heads out of the trench and began to look around. Here we remembered that there are people who work with anonymized personal data, for example in internet advertising, and decided to find out how they are doing. It turned out to be very fortunate that at least three employees of the company once worked together with Mikhail Benyukhis, who has been engaged in internet advertising for the last 15 years, and asked him for an interview. Fortunately, Mikhail did not refuse and here is what happened.
-
How is the modern world of internet advertising organized? Was it once banner advertising? Has anything fundamentally changed since then?
The basic principles of advertising are the same in all media, including the internet. If they change over time, they do so slowly and gradually.
As thirty and fifty years ago, the main task of advertising is to build a relationship between the advertiser and his potential clients. The key concept in the advertising industry is the "funnel", which is often described as three consecutive phases of interaction. First, the advertiser wants to inform the audience about its existence or remind them of it. Then, in some way, describe the advantages of its offer, and in the last, third phase, lead the potential client to purchase.
There are advertising strategies for each of these tasks. Depending on the strategy, advertising formats are chosen. The banners you mentioned, which were once the only format, are used less often today, but various mobile formats, video advertising, native advertising, and God knows what else have developed.
I hope this description does not offend the feelings of professionals. I am still not a real marketer, I am more interested in the technology behind advertising management.
-
What roles (functions) exist in modern advertising (besides the advertiser and the recipient of advertising messages)? How is this kitchen arranged? What are clusters, auctions, and other mysterious details of the mechanism of online advertising?
In addition to these two, there is always a third participant: the advertising platform. There are very large platforms, such as Amazon or YouTube. They sell their services to advertisers or agencies representing advertisers themselves, they don't need anyone else.
If the platform is not that large, it cannot sell its services directly, it lacks scale. And here arises an entire ecosystem of intermediaries and service companies that, roughly speaking, buy advertising slots on different platforms, package them, and resell them at a higher price than they bought. The larger the market, the more complex the ecosystem. America is undoubtedly ahead of the rest of the world here, but in other developed economies, there are niches for intermediaries with substantial tasks.
-
Tell us about the company IPONWEB, where you worked for 15 years, as far as the NDA allows.
-
does the name mean anything?
-
who were and are its clients?
-
what market share does iponweb's business occupy?
-
does the company have its own original developments (technologies, algorithms, tools)?
-
is there any serious mathematics behind these technologies?
-
what was your role in the company and how did it change over 15 years?
I joined IPONWEB in 2009, by that time the company was 6 years old. The name has been the same from the beginning, and when it appeared, it had a meaning related to the business at that time. But I did not witness those times.
When I joined IPONWEB, almost all of its activities were related to advertising technologies at Yahoo. I don't know if anyone remembers this company now, but at that time it was one of the pillars on which the internet stood. IPONWEB brought machine learning to ad placement on the Yahoo network, and it was undoubtedly a breakthrough.
Then there were more clients. Among them were platforms and advertisers, but mostly our clients were various types of intermediaries. For them, the value of advertising technology is higher because advertising is their main and only business. In this segment, we worked with many well-known companies, including, for example, Google. I have restrictions on mentioning clients, and besides, I don't remember well who can be mentioned and who can't. Google, it seems, can be mentioned.
Since IPONWEB was a technology startup, I managed to do a variety of things there, from product development to operational management. As of two years ago, when we were acquired by the French company Criteo, I was working as CTO.
-
There are several myths about advertising, please comment on them:
-
advertising will eventually turn into targeted information
-
advertising will become personalized
-
advertising will be controlled (is already controlled) by AI
For me, these are not myths, but rather exaggerations and simplifications. Targeted information is one of the standard tasks, it has been more or less successfully solved for many years. It is unlikely that all other tasks and related advertising strategies will disappear, so no, all advertising will not turn into targeted information, but this task will not disappear either.
Personalized advertising has also existed for a very long time and is widely used. Will there be more of it, I don't know, but I don't see any signs of it.
AI is more difficult for me. AI is not a term, but a collective name for an indefinite set of technologies and methods of their application. Neural networks have been used in advertising for a long time, as well as other methods of machine learning. But even if we talk about LLM, surely the same Google uses Gemini in some advertising tasks, although I have not discussed this with them.
-
How is the debugging of advertising algorithms done? By what metrics?
Oh, that's a great question! Algorithms, unfortunately, always do what you ask them to do, not what you mean. If you ask to buy a lot of clicks cheaply, a good algorithm will happily buy fraud, but very cheaply. Therefore, the formalization of the optimized metric is an important part of the description of the advertising strategy.
In the lower part of the marketing funnel with simpler metrics, you can optimize something like the number of sales or return on advertising spend (ROAS). And in the upper part, where tasks are related to creating a good attitude towards the advertised brand, pure art begins. You can't look into a person's head to understand whether he liked this ad in such a place, and whether he noticed it at all. As a result, everyone gets out as much as their own creativity allows.
And how does debugging happen? Quite standardly. We use supervised learning at this point and retrain the model on the added data every few hours. I think everyone does this, with slight variations.
-
How is user identification done on anonymized data?
For most advertising tasks, you actually only need to know that this person who is now choosing a refrigerator on Avito is the same person who was on the LG website yesterday. Information that this is a 37-year-old housewife with a higher education adds almost nothing to the prediction of advertising effectiveness.
In our industry, UUIDv4, randomly generated 128-bit sequences, are used to identify users. The only problem is that these identifiers are stored in cookies, so they are tied to a specific site and, as a rule, live for a very short time. That is, the identifier of the user who visited LG yesterday will not match the identifier of the user who visited Avito today. And this is a problem from the point of view of advertising systems.
-
How important is the "freshness" of data for advertising effectiveness?
Depends on the advertising strategy. At the bottom of the funnel, it is more important, at the top less so. If a person is already close to making a purchase, in 15 minutes it may be too late to advertise something to him. And for the first acquaintance with the brand, a few days do not matter, and often even a few weeks.
-
Is there a concept of a "consumer profile" in modern advertising? How many parameters define a profile?
Yes, but it is used in such a descriptive sense: we put everything we know about the user with such a UUID into one pile and call it a "profile". And we usually know very little, some random fragments of the digital footprint. The most interesting thing that can get into such a "profile" is information about actual purchases made. Often just at the level of the fact that this user bought something on such and such a site, sometimes with the amount of the purchase or the name of the product.
Further, any calculated parameters can be extracted from this profile. Usually there are from a couple of dozen to a couple of hundred, and their meaning is extremely technical, it is difficult to describe them in human language. In general, it does not look like your profile on Facebook at all.
Another question is that Facebook itself can calculate the same couple of hundred parameters based on your complete profile, including the list of friends and what else it knows about you. With a successful combination of circumstances, this additional knowledge can significantly increase the effectiveness of advertising, so it is very difficult to resist using it.
-
What data is better (more often) used in targeting?
Here we may have a terminological discrepancy. For me, targeting is a set of restrictions that the operator of the advertising system sets when configuring the campaign. The most common here is geography, of course: it is useless to advertise in Zimbabwe a product that is sold in Mexico, etc.
For the operation of optimization algorithms, several classes of data are important. Firstly, everything that we managed to extract from the user's profile. Secondly, everything we know about the display context: platform, page content, ad placement on it. Thirdly, everything else. For example, what exactly we are advertising, pencils may differ from airline tickets. Or browser versions: quite often in the new version something stops working that worked in the previous one. Algorithms catch this quite well.
-
Algorithms that decide who to show which ad to - how automatic are they? Can they be managed without owning IPONWEB? Can they be provoked and show Trump ads in democratic states? Or somehow compromise this soulless machine?
As a rule, the system operator can choose one of the advertising strategies supported by the system, set up targeting, and allocate a campaign budget. To go deeper, you still need to own the technology or somehow negotiate more complexly with its owner, this also happens. But you can perfectly compromise the machine with a standard set of features, if you wish. The owners of advertising systems do not like this and fight it as best they can. With predictably varying success.
-
On which platforms do which algorithms work, and does one algorithm pass anonymized UserID data to another algorithm on another platform?
Sometimes it does. I don't know about all markets, it depends quite a lot on the laws in a particular country, but in America, probably about 20 percent of ad impressions use data obtained from other platforms. It's hard to calculate exactly here, but that's the order.
In this context, an anonymized UserID is usually a one-way hash of the visitor's email address. If the advertising system serves two platforms, it can give both of them an algorithm for calculating the hash, which allows it to understand that this hash, calculated by Avito, and this one, calculated by LG, correspond to the same visitor. This knowledge is used to select the displayed advertisement.
-
Have you ever encountered different alarming situations in your life, when you talked out loud with your family about changing your car to a car of brand N yesterday, and today the whole internet is filled with ads for car dealerships of this brand? Are they eavesdropping?
Even worse, you didn't have time to talk, you just thought about it, and they are already everywhere! In the vast majority of cases, such situations have intricate but quite innocent explanations.
Personally, I have never encountered unauthorized use of recorded sound for advertising optimization purposes. It is too expensive and complicated to be commercially meaningful. Legal risks cannot be completely ignored either, even in countries with such weak regulation as Russia. And finally, if you are being illegally eavesdropped, there are surely more profitable ways to monetize the information obtained.
But if you yourself have allowed someone to use the recorded sound for advertising optimization purposes, they are probably already using it or planning to use it. So read the user agreements of applications that have access to the microphone on your phone. And to the camera just in case.
However, even without eavesdropping, there are methods of information dissemination that I personally do not like. For example, a company that processes credit card payments may pass on information about your purchase to someone who will use it for advertising purposes. Like, you bought dishwasher tablets of a new brand, and you start getting ads for laundry detergent from the same manufacturer everywhere. This is an absolutely real story in Russia today, and in my opinion, it violates citizens' rights. European laws are stricter, American ones are somewhere in between, I think.
-
Do the annoying pop-ups with Cookie warnings have anything to do with advertising technologies?
The most direct! They were invented specifically to prevent advertising systems from exchanging valuable information for them. Well, primarily advertising. There are also various other methods to use cookies for more malicious purposes, but advertising has suffered the most. This solution does not seem to me to be either effective or balanced. I want to believe that it can be done better, but for now, it is what it is.
-
Annoying pop-up ads that block content and cannot be closed immediately. Does this still work and bring any results other than user irritation? Who manages this? The site owners or the advertisers who got the right to show the ad.
Usually, the site owner manages it. Sometimes it happens that the advertising system or advertiser does something that the site owner did not expect, but this is rare.
It seems to me that if they use it, it means it works, at least it brings income to the site. I honestly doubt the great benefit for advertisers, but surely in some cases, it is useful for them too. The industry usually rejects what is completely useless.
-
About 5 years ago, some application (extension) for Chrome exploded in popularity, which was designed to kill all algorithms in the world because it specifically clicked on ALL ads it saw in the browser in a hidden mode from the user. Thus, all targeting systems were destroyed, and for the user, it simply cut out banners from the page. One of us used it until Chrome banned it and removed it from the Chrome Store. Question: how dangerous are such things for the market, or are they useful for the market, or has everyone already learned to cut off such gadgets?
Cutting off in the sense of detecting is not that difficult, a good system will do it automatically and quite quickly. It is unclear what to do after that. The platform believes that the ad was shown, and even clicked on it, so the advertiser has to pay. The advertiser believes that his ad was not delivered to a potential client, so he does not owe anything to anyone. The contracts between all participants in the transaction, including intermediaries and service companies, say different things. Some of them agree with the platform, some with the advertiser, some have a third position altogether. And the court is likely to interpret this situation as fraud on the part of the user. I'm not a lawyer, but it seems there have been similar precedents. It is better not to bring it to all this, so it is right that this extension was banned.
-
What will happen to digital (online) advertising "tomorrow"?
As usual, everything is complicated. It seems that the potential of the existing formats is mostly exhausted. Volumes are still growing, but not fast. The market is consolidating, mainly growing super-large platforms like YouTube and TikTok. Many are hoping for the next global market change with the development of dialog navigation, this is about all sorts of chatgpt and other fashionable things. But it is still more fashionable than concrete.
Personally, I do not expect major changes in the next 5 years. It seems to me that the volumes will continue to grow slowly, the global leaders will remain more or less the same, consolidation will also continue. New formats will appear, especially in the lower part of the funnel, but the old ones will also be used.
Generative models will be more widely used for ad selection and performance evaluation, also without fanaticism. Perhaps they will be able to combine them with reinforcement learning, although this is an even more distant prospect. How this might affect the user experience is so unclear that I won't even try to fantasize.
-
Is advertising spam or not always?
It seems to me that this is a matter of definition: if it does not smell like spam, then it is not advertising, but "marketing communication".
But seriously, since advertising has existed for 150 years, there is probably some benefit to it. Advertising tries to entertain, inform, and help in making choices. As a consumer, I like entertaining advertising the most. For example, the annual Coca-Cola Christmas advertisement. You may not like it, but it is still a skillfully made artifact of modern culture.
At the same time, some element of pushing is inevitable in any advertisement.
Write comment