Big Tech keeps getting bigger!

It’s becoming increasingly difficult to wrap your head around just how massive some big tech stocks are getting, especially since they keep outdoing themselves.

The pandemic has pushed even more activity online, and the FAATMAN stocks (Facebook, Amazon, Apple, Tesla, Microsoft, Alphabet, and Netflix) have benefited immensely.

With many of these companies experiencing record breaking quarters, how much revenue do the big tech stocks generate per minute?


Was passiert, wenn Innovation auf Bürokratie trifft


“Deutschland scheitert in kleinen Schritten”, titelt die FAZ ein Interview mit Rafael Laguna de la Vera. Der Software-Unternehmer ist ist Gründungsdirektor von „SprinD“, der Agentur für Sprunginnovationen der Bundesrepublik Deutschland. Mit ihr soll ein bisher für Deutschland einmaliger innovationspolitischer Ansatz zur Förderung von disruptiven Innovationen umgesetzt werden. Und auch wenn da bisher noch kein Wort mit der Endung -ung drin vorgekommen ist, wiehert schon bei dieser Beschreibung der Amtsschimmel.

Dabei bringt Laguna alles mit:  seit über 30 Jahren ist er als Unternehmer und Investor im Bereich Software erfolgreich. Bereits im Alter von 16 Jahren gründete er sein erstes Software-Unternehmen (Elephant Software), mit 21 programmierte er bereits ein komplettes Kassensystem für die Getränkewirtschaft (dicomputer) und mit 31 verkaufte er seine erste Firma (micado). Aktuell ist er hauptamtlich CEO von Open Exchange, Das Unternehmen ist mit 270 Mitarbeitern ein führender Anbieter von Open Source-Software für E-Mail und Office-Productivity. Da bleibt eine Frage ungestellt: Warum tut er sich das an?

Ich empfehle das Interview zur Lektüre – es wirft ein Schlaglicht auf die Innovationsfreude in Deutschland. Oder den Mangel daran.

Zu dem Interview geht es hier.

The (r)evolution of (programmatic) digital advertising

Over the time, many people have asked me how the digital advertising industry works. Many buzz-words have changed over the years and the whole deal seemed more and more complex. But in essence, to make it very short, technology has evolved but the business is still what it is. The different digital channels have gone through almost identical phases of development. First was Banner Advertising on Websites, then followed video ads on the web, then mobile advertising and now we see the next step with CTV / OTT (digital TV / streaming platforms) and even digital out-of-home media.

What matters for all those channels is that the business is (increasingly) dominated by programmatic advertising.

Follow the numbers and you will see that the trend is clear. The share of programmatic advertising in Germany has gone up from 37% in 2017 to projected 62% in 2022. The global programmatic media spent was 43 billion US$, a share of 49%. Statista predicts a share of 90% of the total media spend in the US for 2022, up from a whopping 84,6% in 2020.[i]

But why is that so? Because there has been a programmatic transformation in digital advertising. In principle, media selling and buying is a simple job, which has been done by humans. But high volumes and low prices increasingly called for this job to be automated and the price negotiation to be moved from phone calls to programmatic methods. So, in short, programmatic advertising automates the process of media buying and selling. Injecting machine learning and AI into the process has the goal of optimizing the price and transparency and also the efficiency for both advertisers and publishers. Often this is achieved by auction mechanisms, but that is not always true since there are many shapes and forms, sometimes mixed with traditional ad selling and / or trading desks.


A look back – the history of digital advertising

Some claim that digital (display) advertising first happened in 1994 – a banner ad from AT&T on the predecessor of Wired Magazine,[ii]. In fact, this is not true, because display advertising has been around before that with sponsored content on Compuserve for instance (who remembers that!?) or advertising content on interactive videotex (or Bildschirmtext / BTX in Germany). This follows the logic that putting advertising on a screen is display advertising.

Source: Wikipedia

The revolution, however, started in fact with the Wired ad – which was what has become the digital media format of “full size banner” for many years. These banners had fixed ad slots, which again were sold by publishers in pretty much the same way they would sell their magazine ads – time based, fixed position. This was done by the same ad sales teams, of course.

Soon the market realized that there is so much more potential, and as early as 1995 the first ad server was brought to market. The first one was meant to allow salespeople to sell advertising in a traditional way but across many publications of the same publisher.

During the dot-com era, the number of website publishers rose exponentially, and also traditional publishers picked up and started web-versions of their magazines. The traditional publishers saw the internet as a threat – and believed that their traditional earnings would erode while the new digital earnings would not fill the gap. The response was to either do nothing and hope the internet would go away, or to pump out websites as if there was no tomorrow. Bottom line, the number of available ad spaces was far bigger than the demand from advertisers. The problem was the so-called “unsold” or “remnant” inventory.

This facilitated the growth of the ad serving industry. Ad servers allowed the publisher to sell premium inventory (like their front page) directly to advertisers or to go through advertising networks. So, in essence these ad networks were (and some are still around) brokers for advertising inventory, for the ad slots on the web. Bundling the ad slots of many publishers with the efficiency of one shared salesforce.

Those ad networks were the predecessors of SSPs – Supply Side Platforms (we will get to that later). Still today, ad networks manage the ad sales business for publishers, either exclusively in their name or for their unsold inventory or a mix. The magic lies in the word “network”, because with ad servers these ad networks could sell across multiple websites of multiple publishers, with or without targeting. Bundling packages became possible as did working with dynamic pricing, even though mostly by phone between the sales guy and the media buyer in an agency.

As a rule of thumb, the less “premium” the bundled placements were and the less obvious or transparent it was, where on the represented websites the ad would appear, the lower the price. This meant that an advertiser for a branding campaign had and still has the incentive to buy premium placements at a premium rate, while transactional campaigns or reach generation can run at ultra-low prices with high volumes. The response rates, in that case, obviously are as lousy as the price is low.

This created another problem: Advertisers calling their agency to complain that they have not seen their ad. Even though they checked the website they have been advertising on! It was just too small a sample size of the total inventory for them to see it and advertisers did not understand that they would not always see their ad, like on TV or in a magazine.

And when they realized how it worked, soon there was the demand to segment the audience and to reduce the scatter loss in advertising. Why not show the ad only to people it is relevant for?

The result was the rise of targeting in advertising. Since the early 2000s it has been standard in ad serving and could be geographic, based on cookie data, with a frequency cap (how often one ad should be shown to an individual) etc. etc.

I remember too well that we tried to sell these targeting features at a premium rate to our ad-serving customers, but there was a problem. The more specific the targeting becomes, the smaller the audience is. Even if a publisher (or ad network) would have been able to claim a higher price for a targeted campaign, it was rarely worth the effort – because the reach is so much smaller while it is more work to make it happen. So, the economics of targeted campaigns, in very simple terms, were not favorable with human interaction required.


RTB – the first form of programmatic advertising – enters the stage

This made companies appear which offered Real Time Bidding (RTB). Real time is relative, obviously, but back in the day we used a currency for ad tech performance which was “the ad needs to load faster than the page content”. So, these companies enabled an auction process, which was faster than the loading time for the webpage. We can assume an average of 80-120 milliseconds for the total process from the initial request (the CMS of the publisher calls for an ad) and the ad being delivered. Technically it is more complex than this:

Once a visitor comes to the website, the website CMS will trigger the publisher’s ad server. In case there is no other (premium) campaign it will send a request with website and targeting information (about the page visitor or user) to an ad exchange. The ad exchange then matches the data against available advertising campaigns. Those could for instance be tailored towards improving sales conversion. We all know the situation: We look on websites for a product, for instance a bike rack – maybe even putting it into the shopping basket – and then move on for whatever reason. This information will now be used to match an ad campaign to be displayed only to users, who have been on the shop’s / advertiser’s website before without making the purchase. Other competing campaigns could have entirely different audience targets. And we could add semantic targeting, which would mean to address an audience that is similar to another audience. Contextual targeting looks at the page content – if it is about action cameras it may be used to show an ad for the latest GoPro. Semantic contextual targeting like this, done by i.e., Semasio, would allow you not only to target people looking for shoes, but also those for high-heels, boots or flip-flops. A campaign targeted against “shoes” would then be shown to all those people, just to name some examples.

The clear benefit of this automatic bidding process including the vast targeting possibilities is that advertisers (allegedly) reach their audience without scatter loss. Only those interested in a new Toyota will see the Toyota ad. That is the concept, but it has some limitations:

  • the more targeting is applied, the less people a campaign will reach
  • it works for campaigns with a high probability of relevance, mostly transactional campaigns triggering purchase, subscription, test-drives… any call to action.
  • it works much less for general branding campaigns / campaigns which are supposed to improve brand awareness for the advertiser.

But generally, it works pretty well and yes, RTB is programmatic advertising: Programmatic, as I stated in the beginning, is a way to automate the advertising business. There are other ways to do programmatic advertising, but the vast majority of all programmatic campaigns are executed with RTB, that is at least the case in Europe, with a preference for PMPs and direct deals in the US. A PMP is a programmatic marketing term that stands for Private Marketplace – private advertising auctions, as opposed to public marketplaces. Advertisers can only access the private marketplace with an invitation, and it’s a real-time bidding environment.

So RTB is making the match between buyer and seller in digital advertising, using automated processes. But it is a bit more complex than that.


DSPs, SSPs, Ad exchanges and DMPs

A challenge is that one publisher alone would not be relevant enough, in most cases, for big media buyers to buy programmatically from them. It is important to bundle advertising spaces (inventory). Publishers do this via so called SSPs (Supply Side Platforms), which allows them to share their inventory with one or many ad exchanges. Or, in the current step of the digital marketing evolution, they have become the ad exchanges, with direct connections to the DSPs.

On the other side of the business, large media buyers leverage DSPs (Demand Side Platforms), usually connected to DMPs (Data Management Platforms).

Then again, it is pretty simple: Imagine all advertisers and publishers use one platform to meet and transact. So, all the advertising campaigns that need audience meet all the available advertising inventory. That is what the SSP does – comparable to what the ad network is doing with manpower, just automatically.

This started with publishers (and ad networks) pooling their unsold ad inventory with others to get a bigger chance at selling it, applying yield management just like airlines and hotels do. With higher demand for specific inventory, the price will go up. If there is none, the price will go down.

There are also specific ad exchanges for mobile advertising (e.g., Smaato) and CTV / OTT (digital TV).

The ad exchanges and SSPs need one ingredient to work best: Data. While they work just fine without additional data and just by volume, adding data allows for targeting and segmentation – and that means an increased price for the media buyer. The more publishers or networks are grouped, the more data an SSP can collect and aggregate in hopes to generate higher advertising revenues.

Both DSPs and SSPs are about bundling power. On the demand side, advertisers join DSPs that are connected to SSPs. Those SSPs are where the transaction happens, the DSP is simply funneling and bundling demand to make media buying more efficient for advertisers. On the flip side, SSPs are bundling the (unmonetized, so not otherwise sold) ad inventory of various publishers.

Source: Mobidea blog. In today’s industry, ad exchanges and ad networks play a much less significant role than not too long ago, so there would be a line between the DSPs and SSPs in this chart, leaving out networks and exchanges.


Imagine a visitor comes to a website. The website then sends the information to the SSP, which passes it to the DSP. The DSP says “I want that” back to the SSP along with dozens of other DSPs triggering the bidding. The information about the winning ad campaign is then forwarded to the SSP and further to the website, which then calls the ad from an ad server sending it to the user’s browser to be displayed. And all this happens in <120 milliseconds.

Obviously, an SSP has the function of selling the inventory for the connected Publishers as expensively as possible, and a DSP has the exact opposite purpose: to buy ad space for the lowest possible price.

The most relevant SSPs in 2020[iii] were

  • OpenX
  • Pubmatic
  • Magnite
  • Xandr
  • Oath (Verizon Media)
  • Index Exchange
  • SpotX (soon to be part of Magnite)
  • tv
  • Google AdX

On the demand side, the list reads like this:

  • The Trade Desk
  • MediaMath
  • Amazon (AAP)
  • Google DV360.
  • Xandr
  • Adform
  • Amobee
  • Adobe Advertising Cloud

The first pan-European DSP was the Danish ad tech company Adform, which was founded in 2002 as an ad server for the buy-side (media agencies).

Going back to the grease that keeps the system going: data. I said earlier that Data Management Platforms or DMPs plug into DSPs, but what for? Well, to make the best use of the possibilities programmatic media buying brings, the advertisers need to mine data and use it intelligently. Whatever you do on connected devices, you leave tracks, and that is what DMPs collect, store and sort to help the DSPs make better decisions. For many campaigns, the DSP is fully capable to make the decisions and prioritize right. But when more targeting criteria come into the game, it gets tricky and that is where the DMPs full strength is. It is the “big data engine” that provides you with just the ad you wanted at the time you wanted it – in theory. It is still advertising.

This piece is only meant to give basic insights into programmatic advertising and how it works. I suggest you check out the Ryte Wiki for the most common forms of targeting:

I hope this has given you good insights into the world of programmatic advertising. One thing is clear: programmatic advertising is unstoppable and we can see it gain relevance in digital TV, streaming, OTT and Video on Demand as well as Digital Out of Home. It´s a good time to be in this space.





Google stoppt personalisierte Werbung – eine historische Entscheidung

Google hat angekündigt, zukünftig für Werbung kein User-Tracking mehr zu nutzen. Künftig sollen Nutzer nur noch in recht groben “Interessen-Pools” zusammengefasst werden. Damit begegnet Google kartellrechtlichen Verfahren in den USA und stärkt den Datenschutz.
Das könnte heute, rückblickend betrachtet, ein historischer Tag werden. Noch ist unklar, welche Implikationen die Ankündigung von Google auf das gesamte Werbe-Ökosystem hat. Klar scheint aber so viel:
1- alle, die in der Lage sind, eine direkte, langfristige Beziehung zum Endnutzer aufzubauen werden gewinnen – sowohl auf der Publisher, als auch auf der Advertiser-Seite. Wir werden auch langfristig deutlich mehr Logins sehen. Für kleine Publisher wird es ohne entsprechende Allianzen schwer.
2- für alle intermediären Anbieter, deren Geschäftsmodell auf dem Aggregieren von Nutzerdaten/ IDs basiert, werden in Zukunft unüberwindliche Hürden entstehen. – Ohne den Google Tech Stack lassen sich solche Modelle nicht flächendeckend betreiben.
3- Google positioniert sich im Battle zwischen Apple und Facebook eher auf der Apple Seite. Spannend wird sein, wie stark nun der politische Druck auf Unternehmen wie Facebook, Amazon, Xandr oder LiveRamp zunehmen wird, nachzuziehen.
4- Google verzichtet langfristig sicher auf signifikanten Umsatz im B2B Segment. Aber wird damit ein klareres Spielfeld für seine B2C Produkte Search und YT bekommen.
Es ist und bleibt absehbar also spannend für die digitale Werbebranche.
Mehr dazu gibt es in diesem lesenswerten Beitrag.

Die Lernkurve mit AI – Klasse analysiert von Forbes

Hier im Bücherregal steht der Titel „There is no such thing as Artificial Intelligence“. Vielmehr sind es Algorithmen, die von Menschen gemacht sind und die mit Daten gefüttert werden. Blinde Technik-Gläubigkeit ist hier ebensowenig angebracht wie Panik, aber ein realistischer Blick auf das, was grade passiert mit AI  – die uns ja bereits in allen Lebenslagen umgibt. Diesen kritischen Blick vermissen wir oft in der Diskussion, die eher polarisiert geführt wird. Umso mehr hat uns der folgende Artikel von Forbes gefreut, der einfach nüchtern betrachtet, was beim Einsatz von AI alles schief geht. Und warum es schief geht. Und wie sich dies in der Zukunft darstellt.

Zu dem Artikel, der guten Stoff zum nachdenken und diskutieren liefert, geht es hier: „President Biden is a man, woman and 40 years old – why we need algorithmic transparancy“.