The history of internet advertising can be traced back to May 3, 1978, when the first spam email was sent to 400 users of the computer network ARPANET, deployed at that time in just four locations: the University of Utah, UCLA, UC Santa Barbara, and Stanford Research. Fifteen years later, by 1993, when ARPANET had developed into the Internet and the spread of the Web enabled multimedia websites, the market of banner ads appeared. This new market originally relied on direct selling of banner slots offered by website publishers to advertisers, but this approach started to lose its efficiency very quickly when there was a surge in the number of websites. It became operationally difficult or even impossible for advertisers to run ad campaigns and manage budgets across thousands of publishers. On the other hand, publishers needed a robust and centralized way to sell their inventory at scale.
The challenge was taken up by ad networks that acted as brokers between publishers and advertisers. DoubleClick, launched in 1996, offered a platform that enabled advertisers to run ad campaigns across a wide network of websites, dynamically customize a campaign according to its performance, and measure the return on investment. This created a perfect environment for automatic decision making because the measurements and adjustments could be done dynamically. However, it was not really programmatic at that time. The online search engines, meanwhile, were also struggling to improve their advertising capabilities. Advertisers paid for the number of times their ad was demonstrated by the search engine – the cost per thousand impressions (CPM) model – similarly to banner ads. This approach was inflexible from the pricing perspective, causing certain revenue losses for search engines, and also from the targeting perspective because irrelevant ads were not penalized in any way. The breakthrough happened in 1998 when the GoTo.com search engine introduced an automated auction model with two innovative features:
The per-pay-click (PPC) model improved both revenues and ad relevancy because advertisers who were willing to pay for top ad spots for specific search queries generally offered more relevant and better resources. This model was adopted by Google in 2002 with one principal improvement: the ad was selected based on Google’s expected revenue, not the bid amount. Google measured the click-through rate for each ad as a ratio between clicks and impressions, and the expected revenue was estimated as
revenue = bid price x click-through rate
This was a programmatic self-learning technique that optimized the business objective, both in terms of revenue and relevancy, because click-through rates tend to be low for irrelevant ads, so even highbudget advertisers were not able to clog up the bandwidth.
The trajectories of ad networks and search engines converged in 2007–2009 with adoption of the auction model across the board. Advertisers and publishers became connected by ad exchanges that accepted real-time bids for individual ad impressions, and a new era of real-time bidding, commonly abbreviated as RTB, thus began. The advent of RTB exchanges gave impetus to programmatic tools for advertisers – datamanagement platforms (DMPs) and demand-side platforms (DSPs) – that provided the ability to collect data about the behavior of Internet users and make bids on RTB exchanges depending on the estimated propensity of a given user to respond. The success of RTB was impressive: the share of inventory sold by DoubleClick (acquired by Google by that time) through RTB rose from 8% in January 2010 to 68% in May 2011 [Google Inc., 2011].
Reflecting on the history of RTB, we can conclude that one of the most prominent achievements of programmatic advertising is a framework that enables owners of consumer bases, originally the publishers of web content, to provide personalized marketing services to parties who are limited in their ability to interact with consumers, originally the advertisers of products and services. The infrastructure that sits in between the publishers and advertisers is typically provided by an independent party and includes the following:
Later on, this pattern started to spread across other industries. Other types of consumer-base owners, such as retailers and mobile operators, were also looking for an efficient way to commercialize their data and relationships with consumers, and other types of service users, such as banks, product manufacturers, and insurance companies, were willing to know more about their customers and have more channels for communication with them. For instance, a manufacturer of consumer packaged goods can use a retailer’s channels, such as stores and eCommerce websites, to offer personalized discounts to consumers to promote new products and increase their market share.
Consequently, advertising services and data services started to transform into the more generic model illustrated in Figure 1.2, which represents a multipurpose marketplace of services and data that connects actors from different industries. The range of services offered by such a marketplace can go far beyond advertising, covering areas like credit scores and insurance premiums. The heterogeneity of this environment, where one constantly deals with someone else’s data, often in real-time, leads to overwhelming complexity of data flows and operational decisions, and programmatic methods are probably the only way to tackle it.
