Programmatic Advertising: Real-Time Marketing T he holy grail of advertising and marketing is to deliver the right message to the right person at the right time. If this were possible, no one would receive ads they did not want to see, and then no advertising dollars would be wasted, reducing the costs to end users and increasing the efficiency of each ad dollar. In the physical world, only a very rough approximation of this ideal is possible. Advertisers buy television and radio spots, newspaper ads, and billboards based on broad demographics, and the context in which the ad will be shown. The Internet promised to change this traditional method of buying ad space by allow-ing advertisers to gather personal information on consumers through the use of cookies placed on the user’s browser, which tracked behavior and purchases online and could be matched with offline information as well. Advertisers could then use this information to target ads to just the desired individuals they were seeking, based on personal character-istics, interests, and recent clickstream behavior. From the beginning, e-commerce was a trade-off for consumers between privacy and efficiency: let us know more about you, and we will show you only the advertising and products you are interested in seeing and would be likely to respond to. For brands, the promise was scale and cost: let us know who you are looking for and we will find millions of people on thousands of websites that fit your criteria. E-commerce was supposed to end the mass advertising that began in 19th century newspapers, 20th century radio, and exploded with the growth of television. The latest rendition of these promises from the ad tech industry is programmatic adver-tising, which it touts as an automated algorithmic platform that allows large brands to bid for ad space (web pages) on hundreds of thousands, and even millions, of websites, in coordinated campaigns, measure the results, and extend brands to tens of millions of consumers with unprecedented scale. But in 2017, it has become clear that the promise of programmatic advertising has not been realized and has many risks for brands. In fact, it has injured many brands, and the ad tech industry is reeling from advertiser criticism that the existing online ad ecosystem is murky, opaque, not accountable, and in some cases fraudulent. Contrary to the rosy promises of the online ad industry, most notably the ad giants Google, Facebook, and Twitter, most of the display ads shown to website visitors are irrel-evant, sometimes hilariously so, to visitors interests. For this reason, the click-through rate for banner advertising is stunningly low, well under 1%, and the price of generic display ads is less than $1.00 per thousand views because of their low response rate. Check this out: visit Yahoo (one of the largest display advertisers on earth) on any device, look at the prominent ads on screen, and ask yourself if you are really interested in the ad content at that moment in time. Often, it is an ad for something you have recently searched for on Google or even already purchased at Amazon or other sites. These ads will follow you for days as you are re-targeted across the web and on mobile devices. Researchers have found that only 20% of Internet users find that display ads on websites are relevant to their interests, and depending on the type of ad (sidebar, native inline, pre-roll video, or pop up) are viewed unfavorably by 50% to 78% of visitors. To understand how we ended up in this situation, its useful to review briefly how the Internet ad industry evolved. Digital display advertising has progressed through three eras. In the early 2000s, a firm with a website interested in ad revenue (a publisher) would sell space on its site to other firms (advertisers), usually through an ad agency or via a direct relationship. These were primarily manual transactions. By 2005, ad networks emerged. These networks allowed advertisers to buy ad space on thousands of participat-ing sites in a single purchase and allowed publishers to sell to advertisers more efficiently. Prices were negotiated among the parties. This was very similar to the manner in which ads on cable TV were sold. By 2011, even larger ad exchanges emerged and began using automated real-time bidding for ad space. This provided advertisers access to an even larger pool of publisher ad spaces that numbered up to the millions of websites. Prices and ad placement were automated by algorithms and adjusted based on real-time open auctions, in which advertising firms and brands indicated what they were willing to pay to advertise to consumers meeting specific criteria. Google, Facebook, Twitter, and others developed their own proprietary automated bidding platforms. Collectively, these are called real-time bidding (RTB) programmatic advertising platforms. The result today is an extraordinarily complex ecosystem of players, and sophisticated technologies (called the ad technology stack). In programmatic ad platforms, scale has increased dramatically. In 2017, there are
thousands of advertisers and millions of web pages where ads can be placed. The ads are chosen and generated based on the users browser cookie history and information about the web page, so that ads can supposedly target the right consumers. The content of the web page and the ad location on the page are also important. All programmatic advertising platforms use big data repositories that contain personal information on thousands to millions of online shoppers and consumers; analytic software to classify and search the database for shoppers with the desired characteristics; and machine learning
techniques to test out combinations of consumer characteristics that optimize the chance of a purchase resulting from exposure to an ad. All of this technology is designed to lower the cost and increase the speed and scale of advertising in an environment where there are hundreds of millions of web pages to fill with ads, and millions of online consumers looking to buy at any given moment. In 2017, advertisers are expected to spend over $32 billion (75% of all total display ad spending, including on banners, videos, and rich media) on programmatic advertising. This amount is expected to grow to over 80% by 2019. Programmatic ad platforms have since evolved into three different types: traditional
auction-based real time bidding (RTB) open to all advertisers and publishers; private marketplace (PMP), where publishers invite selected advertisers to bid on their inven-tory; and programmatic direct (PD), where advertisers deal directly with well-established publishers who have developed their own supply-side platforms (an automated inventory of available ad space). Currently, about 75% of programmatic advertising is programmatic direct and 19% is PMP. Open exchange RTB is no longer growing and represents only about 25% of programmatic marketing. The trend is towards publishers, especially well-known brands with large budgets, to reduce their dependence on the operators of the platforms, and exert much more control over where their ads appear, how visible they are, and what content they are associated with. To find out why continue reading. Currently, just 25% of online display advertising is still done in a non-automated,
traditional environment that involves marketers using e-mail, fax, phone, and text mes-saging in direct relationships with publishers. Traditional methods are often used for high value premium ads, say, the top of the screen with a video, expanding ads seen at major newspapers, magazines, and portal sites, and native ads appearing alongside or interwoven with native content. This is the world of the traditional insertion order: if you want to advertise on a specific newspaper or magazine website, call the ad department and fill out an insertion order. For instance, if you are a brand selling biking accessories, you can tell your ad agency to place ads in biking magazine websites and on social networks, targeting the readers of those magazines. In this environment, firms who want to sell products and services online hire advertising agencies to develop a marketing plan, and the agency directly contracts with the ad department of the publishers. This traditional environment is expensive, imprecise, and slow, in part because of the
number of people involved in the decision about where to place ads. Also, the technology used is slow, and the process of learning which of several ads is optimal could take weeks or months. Real-time so-called A/B testing is difficult. The ads could be targeted to a more precise group of potential customers, and to a much larger group of potential customers. While context advertising on sites dedicated to a niche product is very effective, there are many other websites or social network pages visited by bikers that might be equally effective, and cost much less. The process is very different in an open exchange (RTB) programmatic environment.
Ad agencies have access to any of several programmatic ad platforms offered by Google, Yahoo, AOL, Facebook, and many other pure ad platforms. Working with their clients, the ad agency more precisely defines the target audience to include men and women, ages 2435, who live in zip codes where mountain biking is a popular activity, have mentioned biking topics on social networks, have e-mail where mountain biking is discussed, make more than $70,000 a year, and currently do not own a mountain bike. The ad agency enters a bid expressed in dollars per thousand impressions for 200,000 impressions to people who meet most or all of the characteristics being sought. The platform returns a quote for access to this population of 200,000 people who meet the characteristics required. The quote is based on what other advertisers are willing to pay for that demographic and characteristics. The quote is accepted or denied. If accepted, the ads are shown to people as they move about the Web, in real time. As people visit various websites, the automated program assesses whether they meet the desired characteristics, and displays the mountain bike ad within milliseconds to that person. The programmatic platforms also track the responses to the ads in real time, and can change to different ads and test for effectiveness based on the platforms experience in near real time. The programmatic platforms claim they use algorithms and machine learning programs which can identify over time the most effective ads on the most productive websites. At least this is the promise. In PMP platforms, a group of publishers invite selected advertisers to bid on ad space,
often using the publishers own customer data. Generally, the publishers know more about their customers than the ad platforms algorithms and databases can provide. For instance, the leading online newspapers might combine their inventory of ad space (web pages) and invite premium big budget brands to bid on the space. This gives the publishers much more control over who advertises on their pages, and gives advertisers a shot at getting premium ad space, better page placement, and better results from more precise knowledge of the consumer. This is reflected in higher costs for the advertisers. In the programmatic direct model, a single publisher directly contracts with selected brands and advertisers for guaran-teed placement of ads, and like PMP methods, offers both parties more control and precision. Brands and ad agencies bid for this space in a semi-automated environment. In some cases, prices are negotiated directly between the publisher and the brands or their ad agencies. The risks of RTB are essentially that brands lose a great deal, if not all, control over
the presentation of ads, including what websites they appear on, where on the screen they appear (above or below the fold or scroll), how long the ad is present on screen, who is doing the clicking on the ads (real interested persons or bots or fake people), and the content of the website. For instance, in early 2017, JPMorgan Chase had ads appearing on an estimated
400,000 websites a month using programmatic RTB auctions. It became suspicious when only 12,000 sites produced any clicks. An intern was assigned to visit each one to see if they were appropriate for the bank. The intern discovered that 7,000 were not, leaving 5,000 for whitelisting as pre-approved websites. JPMorgan Chase has not experienced any fall-off in the visibility of its ads on the Internet since it eliminated 355,000 websites from its ad campaign. Going forward, JPMorgan Chase intends to winnow the list to only 1,000 whitelisted sites. One of the blacklisted sites advertising the JPMorgan Chases private client services turned out to be a website called Hillary 4 Prison. In 2017, YouTube came under intense fire by leading brands because their YouTube
ads were appearing next to offensive material, promoting racism, hate and terrorism. As a result, JPMorgan Chase, Verizon, Gerber, AT&T, Johnson & Johnson, Lyft, and Procter & Gamble (the worlds largest advertiser) all pulled ads from YouTube. In addition to malicious sites, there are millions of fake sites on the Web set up for
the sole purpose of displaying ads and generating revenue. Facebook has an estimated 84 million fake accounts, and Twitter has over 21 million. Many of the fake sites are bots that
generate clicks but have no real people viewing the ads. While RTB platforms try to prevent this behavior, they can easily be defeated. The result is large ad expenditures but fewer legitimate clicks, and lower conversions. Analysts estimate that the top 50 online media publishers account for only 5% of all ads shown on the Web. This means 95% of the ads are being shown on niche websites with small audiences, or completely fake sites, with fake visitors. In general, ad platforms have little idea, if any, of where the ads are appearing, the content of the websites, or who is clicking. Top brands with large budgets no longer believe the ad platforms’ claims that they use algorithms and machine learning to weed out fake sites, hate sites, and sites that feature porn. Its anyones guess how much this all costs brands, but the National Association of Advertisers estimates the cost at $6.5 billion a year.
Case Study Questions
1. Pay a visit to your favorite portal and count the total ads on the opening page. Count how many of these ads are (a) immediately of interest and relevant to you, (b) sort of interesting or relevant but not now, and (c) not interesting or relevant. Do this 10 times and calculate the percentage of the three kinds of situations. Describe what you find and explain the results using this case.
2. Advertisers use different kinds of profiles in the decision to display ads to customers. Identify the different kinds of profiles described in this case, and explain why they are relevant to online display advertising.
3. How can display ads achieve search-enginelike results? 4. Do you think instant display ads based on your immediately prior clickstream will be as effective as search engine marketing techniques? Why or why not?
6.7 REVIEW KEY CONCEPTS
Understand the key features of the Internet audience, the basic concepts of consumer behavior and purchasing, and how consumers behave online.
Key features of the Internet audience include the number of users online, the intensity and scope of use, demographics and aspects, the type of Internet connection, and community effects.
Models of consumer behavior attempt to predict or explain what consumers purchase, and where, when, how much, and why they buy. Factors that impact buying behavior include cultural, social, and psychological fac-tors.
There are five stages in the consumer decision process: awareness of need, search for more information, eval-uation of alternatives, the actual purchase decision, and post-purchase contact with the firm.
The online consumer decision process is basically the same, with the addition of two new factors: website and mobile platform capabilities and consumer clickstream behavior.
Identify and describe the basic digital commerce marketing and advertising strategies and tools. A website is the major tool for establishing the initial relationship with the customer.