The Magic of Machine Learning in Real Time

Part of the magic of real-time bidding is found within machine learning. This involves using sophisticated algorithms to “learn” complex patterns based on large amounts of data in order to make optimal advertising decisions. The importance of machine learning is cost avoidance and value creation. Cost avoidance is simple to understand: data-driven optimization strategies help reduce waste by identifying the most relevant impressions while selecting the best ads (better creative message, better offer, etc.), which in turn improves performance and ROI (define). Value creation, on the other hand, happens when buyers and sellers of commoditized offerings are more efficiently brought together for a transaction.  To create machine learning magic, two ingredients are required: scale and prediction. Scale speaks to the need to make more users/impressions available through the auction marketplaces, and increase the number of advertisers bidding on these users. The bigger the scale, the more sophisticated the data-driven prediction can be. The second ingredient refers to the idea that prediction needs to be “accurate enough.” Amazon and Netflix have demonstrated that when you provide an accurate prediction of what consumers want, the business grows in two ways:

1. Better inventory control and more purchases/utilization.
2. Better targeting becomes a custom delight feature when it’s perceived to be quite accurate by average consumers.

The ability to deliver relevant choices in real time based on what consumers reveal about themselves creates a virtuous cycle: 

Ads/recommendations become more accurate → consumers are willing to share additional information about themselves → the machine learning algorithm for ads/recommendations becomes even smarter

In digital advertising, the machine learning prediction ultimately boils down to two parts: identifying your target audience and reaching them efficiently. The first part requires machine learning at the user level to learn the most optimal audience segments to target; the second requires machine learning to drive real-time bidding strategy with precision. For instance, demand side platform technology allows advertisers to have global control over how many times each user sees the ads (i.e., frequency capping) and how they see them. This begs the obvious question, “what is the optimal number of ad repetitions?” The answer might be an average of five times over a period of seven days. Problem solved? Not quite.

Read More: ClickZ

BrightRoll Launches Video Ad Exchange

Earlier this year, video ad company Adap.tv launched a video ad exchange in partnership with Gannett Co. and Publicis Groupe’s VivaKi digital unit. Now, its OneSource platform will have some competition from a rival video ad marketplace started by video ad network BrightRoll.  As with the online exchanges that have emerged in recent years for display and other types of advertising, the goal is to bring increased efficiency to the video sector by giving publishers a way to unload unsold inventory and media buyers an automated system for reaching particular audiences across a wide range of sites.  “What we’re essentially releasing is a video advertising business in a box for buyers of online advertising,” said BrightRoll CEO Tod Sacerdoti, who added that the new exchange dubbed BRX is an outgrowth of the company’s efforts to further automate its own ad network over the last 18 months. BrightRoll is the third-largest U.S. video ad network based on streaming video ads viewed — at 333,492 in June, according to comScore.  “We realized everything we were building was applicable to other media buyers and sellers,” said Sacerdoti. A BrightRoll study earlier this year found that half of publishers surveyed reported that at least 20% of their online video advertising inventory is never sold, suggesting the potential for an automated, auction-based marketplace for pre-roll ads.

Read More: MediaPost

Forbes Sells Investopedia To ValueClick For $42 Million

After less than two months on the block, Forbes Media has sold financial education site Investopedia to lead gen provider ValueClick (NSDQ: VCLK) for $42 million. In June, Forbes retained the Jordan, Edmiston Group, Inc. three years after it bought the Canadian-based site.  The announcement comes a few weeks after Forbes purchased freelance journalism site True/Slant, which was shut down last week and will remain live as an archive site only.  Forbes had been an investor in True/Slant and before the purchase, it had hired site’s founder, Lewis DVorkin, as consultant to help restructure its digital offerings. The quick sale of Investopedia is a the first step in the struggling publisher’s latest digital reinvention. Despite the fact that Forbes was eager to sell the Edmonton, Alberta-based Investopedia, the company claimed that the site’s profits and users have grown in the past three years since it was acquired.  In a release, ValueClick CEO Jim Zarley said that Investopedia gives the company “great content, organic traffic and established advertiser relationships in the important financial services advertising vertical.” He also believes that the addition of Investopedia will be able to help build up its ValueClick Brands and ValueClick Media offerings.

Read More: PaidContent.org

 The Rise Of Real-Time Bidding Is The Biggest Online Advertising Story Of 2010

AdMeld, a New York City based ad inventory optimizer, just closed on a $15 million round of venture funding, in the latest sign that the real-time bidding (RTB) market for display advertising is on fire.  Last month, Google paid a reported $70 million for demand-side platform Invite Media. And just a few weeks ago, brand safety startup AdSafe, which will increasingly work with RTB platforms, raised $7.5 million.  The rise of RTB is the biggest story of 2010 in online advertising, and has been written about extensively in ad industry publications. But people outside of advertising don’t seem to know anything about it.

Read More: BusinessInsider

A Peek Inside the M&A Playbooks of Technology’s Top Acquirers

Last night a group of M&A gurus from the corporate development teams at top tech acquirers Google, Microsoft, Yahoo, Cisco, Facebook and Twitter gathered to share insights into their business with a group of startups at a fancy-pants Los Altos Hills, Calif. mansion. Though Facebook and Google might have been the most notable active acquirers lately, everyone on the panel said they are out shopping. They each have a bit of a different style, and a bit of a different target startup. Below are the most notable bits from each participant:

Google‘s Amin Zoufonoun said that he looks at three types of acquisitions: a proven product and team, an uncertain big bet, or market and tech leadership (like YouTube and DoubleClick). He said recent acquisitions by Google and other companies like Apple point to the fact that mobile is not a core part of the DNA of many tech giants. As for advice, he warned startups that they always underestimate how long it takes to close an acquisition; for Google, deals usually take three to four months. As for areas he’s interested in, Zoufonoun said he thinks music is overhyped (an interesting comment given Google is reportedly looking to make a play in this space), and mobile user interfaces are underhyped.

Cisco‘s Derek Idemoto talked up the value of post-acquisition integration. His company has been incredibly acquisitive, with 140 deals in the last 20-odd years. Idemoto bragged that 75 percent of acquired employees are still at Cisco after four years. He said he thinks video is underhyped, and that he’s particularly interested in data. “The most, and most relevant data might win,” he explained.

Read More: Gigaom.com

Why Agencies Must Lead The Technology Charge

Countless articles have been written in recent years putting agencies in the hot seat to adapt their business models or die. Why? Never-ending budget cuts and the digitization of the marketing landscape have produced two key trends currently threatening the livelihood of the traditional agency:

• Media has become digital, multi-channel, multi-platform, and decentralized. These elements are forcing media publishers to be more creative in how inventory is packaged and sold (e.g., bundling offers cross channels from print, online, to mobile). Furthermore, media companies are tired of losing revenue to agencies for the production of creative assets and are thus building and buying their own capabilities in house.
• Innovations in technologies, from brand monitoring, audience targeting, and media planning and buying technologies, to social media and mobile content solutions, drive when and how brands connect with consumers. Many of these technologies are being developed outside of the agency ecosystem.

Read More: iMediaConnection

Big Money Bet on Display Ad Tech

The banner ad is the Web’s original advertising format, but many have viewed it as a disappointment. Prices for display ads quickly tumbled, and marketers fell in love with targeted search options.  That’s not to say display units are on their way out. On the contrary, tens of millions of dollars in venture capital is flowing into ad technology. Investors are betting that a market the Interactive Advertising Bureau pegged at $8 billion in 2009 can quickly grow five times or more with the help of better machinery. (See also: “Display Ads Aim for a Banner Year.”)   “If you take the logic behind targeting to the extreme, it’s all about discovering hidden tier-one inventory,” said Terence Kawaja, managing director of GCA Savvian Advisors. “There’s a lot of inefficiency in inventory pricing.”  Inventory aggregator AdMeld is the latest company to benefit from this belief, closing a $15 million Series C round of funding that brings its backing to $30 million. Norwest Venture Partners led the round, which included AdMeld’s previous VCs as well as a strategic backing from Time Warner Investments.  AdMeld operates a tech platform that publishers use to maximize the amount of money they make from display ads. Publishers like Discovery, Fox News, Reuters and Pandora use its yield-optimization software to determine how best to package display ad inventory for audience-based buys.

Read More: AdWeek

Simplifying The Narrative

Josh Chasin of comScore can definitely count me among his fans.  He wrote a great article late last year on the limitations of CTR as a metric.  A couple weeks back he wrote another great one that I have been looking for a moment to comment on.  Between the upcoming product launch and the 1 year old I finally found a little time, somewhat belatedly.  As I read it, the main theme of Josh’s most recent article was that as an industry we have inhibited the migration of brand-focused budgets online with complex and conflicting narratives, which cause advertisers essentially to throw up their hands and look for reasons not to spend.  I couldn’t agree more.  In fact, I don’t think Josh would object to framing this as a different angle on the same idea I discussed in a post last year (Josh – feel free to comment if I am taking your name in vain).  Regardless of the angle we each take on the story, we’re clearly in violent agreement that the narrative needs to be simpler.  Josh is also quite correct that the 30-spot is an extremely compelling creative format, next to which a hastily-assembled static banner can look, well, flat.  However, as I have previously noted, within 5 years about 80% of households will have the capability to fast forward through that compelling creative.  Online creative formats get more compelling every year – it’s not hard to imagine a well-made pre-roll, rich media or even animated flash creative execution comparing favorably to a TV ad that is watched at 10X normal speed with no sound.  Even before DVRs reach their inevitable tipping point, the research shows that online advertising drives sales at least as well as TV.

Read More: Brand.net

The Web’s New Gold Mine: Your Secrets

Hidden inside Ashley Hayes-Beaty’s computer, a tiny file helps gather personal details about her, all to be put up for sale for a tenth of a penny.  The file consists of a single code— 4c812db292272995e5416a323e79bd37—that secretly identifies her as a 26-year-old female in Nashville, Tenn.   The code knows that her favorite movies include “The Princess Bride,” “50 First Dates” and “10 Things I Hate About You.” It knows she enjoys the “Sex and the City” series. It knows she browses entertainment news and likes to take quizzes.  “Well, I like to think I have some mystery left to me, but apparently not!” Ms. Hayes-Beaty said when told what that snippet of code reveals about her. “The profile is eerily correct.”  Ms. Hayes-Beaty is being monitored by Lotame Solutions Inc., a New York company that uses sophisticated software called a “beacon” to capture what people are typing on a website—their comments on movies, say, or their interest in parenting and pregnancy. Lotame packages that data into profiles about individuals, without determining a person’s name, and sells the profiles to companies seeking customers. Ms. Hayes-Beaty’s tastes can be sold wholesale (a batch of movie lovers is $1 per thousand) or customized (26-year-old Southern fans of “50 First Dates”).  “We can segment it all the way down to one person,” says Eric Porres, Lotame’s chief marketing officer.

One of the fastest-growing businesses on the Internet, a Wall Street Journal investigation has found, is the business of spying on Internet users.  The Journal conducted a comprehensive study that assesses and analyzes the broad array of cookies and other surveillance technology that companies are deploying on Internet users. It reveals that the tracking of consumers has grown both far more pervasive and far more intrusive than is realized by all but a handful of people in the vanguard of the industry. 

• The study found that the nation’s 50 top websites on average installed 64 pieces of tracking technology onto the computers of visitors, usually with no warning. A dozen sites each installed more than a hundred. The nonprofit Wikipedia installed none.

• Tracking technology is getting smarter and more intrusive. Monitoring used to be limited mainly to “cookie” files that record websites people visit. But the Journal found new tools that scan in real time what people are doing on a Web page, then instantly assess location, income, shopping interests and even medical conditions. Some tools surreptitiously re-spawn themselves even after users try to delete them.

• These profiles of individuals, constantly refreshed, are bought and sold on stock-market-like exchanges that have sprung up in the past 18 months.

The new technologies are transforming the Internet economy. Advertisers once primarily bought ads on specific Web pages—a car ad on a car site. Now, advertisers are paying a premium to follow people around the Internet, wherever they go, with highly specific marketing messages.  In between the Internet user and the advertiser, the Journal identified more than 100 middlemen—tracking companies, data brokers and advertising networks—competing to meet the growing demand for data on individual behavior and interests.  The data on Ms. Hayes-Beaty’s film-watching habits, for instance, is being offered to advertisers on BlueKai Inc., one of the new data exchanges.  “It is a sea change in the way the industry works,” says Omar Tawakol, CEO of BlueKai. “Advertisers want to buy access to people, not Web pages.”  The Journal examined the 50 most popular U.S. websites, which account for about 40% of the Web pages viewed by Americans. (The Journal also tested its own site, WSJ.com.) It then analyzed the tracking files and programs these sites downloaded onto a test computer.  As a group, the top 50 sites placed 3,180 tracking files in total on the Journal’s test computer. Nearly a third of these were innocuous, deployed to remember the password to a favorite site or tally most-popular articles.

But over two-thirds—2,224—were installed by 131 companies, many of which are in the business of tracking Web users to create rich databases of consumer profiles that can be sold.  The top venue for such technology, the Journal found, was IAC/InterActive Corp.’s Dictionary.com. A visit to the online dictionary site resulted in 234 files or programs being downloaded onto the Journal’s test computer, 223 of which were from companies that track Web users.  The information that companies gather is anonymous, in the sense that Internet users are identified by a number assigned to their computer, not by a specific person’s name. Lotame, for instance, says it doesn’t know the name of users such as Ms. Hayes-Beaty—only their behavior and attributes, identified by code number. People who don’t want to be tracked can remove themselves from Lotame’s system.  And the industry says the data are used harmlessly. David Moore, chairman of 24/7 RealMedia Inc., an ad network owned by WPP PLC, says tracking gives Internet users better advertising.  “When an ad is targeted properly, it ceases to be an ad, it becomes important information,” he says.  Tracking isn’t new. But the technology is growing so powerful and ubiquitous that even some of America’s biggest sites say they were unaware, until informed by the Journal, that they were installing intrusive files on visitors’ computers.

The Journal found that Microsoft Corp.’s popular Web portal, MSN.com, planted a tracking file packed with data: It had a prediction of a surfer’s age, ZIP Code and gender, plus a code containing estimates of income, marital status, presence of children and home ownership, according to the tracking company that created the file, Targus Information Corp.  Both Targus and Microsoft said they didn’t know how the file got onto MSN.com, and added that the tool didn’t contain “personally identifiable” information.  Tracking is done by tiny files and programs known as “cookies,” “Flash cookies” and “beacons.” They are placed on a computer when a user visits a website. U.S. courts have ruled that it is legal to deploy the simplest type, cookies, just as someone using a telephone might allow a friend to listen in on a conversation. Courts haven’t ruled on the more complex trackers.  The most intrusive monitoring comes from what are known in the business as “third party” tracking files. They work like this: The first time a site is visited, it installs a tracking file, which assigns the computer a unique ID number. Later, when the user visits another site affiliated with the same tracking company, it can take note of where that user was before, and where he is now. This way, over time the company can build a robust profile.

One such ecosystem is Yahoo Inc.’s ad network, which collects fees by placing targeted advertisements on websites. Yahoo’s network knows many things about recent high-school graduate Cate Reid. One is that she is a 13- to 18-year-old female interested in weight loss. Ms. Reid was able to determine this when a reporter showed her a little-known feature on Yahoo’s website, the Ad Interest Manager, that displays some of the information Yahoo had collected about her.  Yahoo’s take on Ms. Reid, who was 17 years old at the time, hit the mark: She was, in fact, worried that she may be 15 pounds too heavy for her 5-foot, 6-inch frame. She says she often does online research about weight loss.  “Every time I go on the Internet,” she says, she sees weight-loss ads. “I’m self-conscious about my weight,” says Ms. Reid, whose father asked that her hometown not be given. “I try not to think about it…. Then [the ads] make me start thinking about it.”  Yahoo spokeswoman Amber Allman says Yahoo doesn’t knowingly target weight-loss ads at people under 18, though it does target adults.  “It’s likely this user received an untargeted ad,” Ms. Allman says. It’s also possible Ms. Reid saw ads targeted at her by other tracking companies.  Information about people’s moment-to-moment thoughts and actions, as revealed by their online activity, can change hands quickly. Within seconds of visiting eBay.com or Expedia.com, information detailing a Web surfer’s activity there is likely to be auctioned on the data exchange run by BlueKai, the Seattle startup.

Each day, BlueKai sells 50 million pieces of information like this about specific individuals’ browsing habits, for as little as a tenth of a cent apiece. The auctions can happen instantly, as a website is visited.   Spokespeople for eBay Inc. and Expedia Inc. both say the profiles BlueKai sells are anonymous and the people aren’t identified as visitors of their sites. BlueKai says its own website gives consumers an easy way to see what it monitors about them.  Tracking files get onto websites, and downloaded to a computer, in several ways. Often, companies simply pay sites to distribute their tracking files.  But tracking companies sometimes hide their files within free software offered to websites, or hide them within other tracking files or ads. When this happens, websites aren’t always aware that they’re installing the files on visitors’ computers.  Often staffed by “quants,” or math gurus with expertise in quantitative analysis, some tracking companies use probability algorithms to try to pair what they know about a person’s online behavior with data from offline sources about household income, geography and education, among other things.  The goal is to make sophisticated assumptions in real time—plans for a summer vacation, the likelihood of repaying a loan—and sell those conclusions.  Some financial companies are starting to use this formula to show entirely different pages to visitors, based on assumptions about their income and education levels.  Life-insurance site AccuquoteLife.com, a unit of Byron Udell & Associates Inc., last month tested a system showing visitors it determined to be suburban, college-educated baby-boomers a default policy of $2 million to $3 million, says Accuquote executive Sean Cheyney. A rural, working-class senior citizen might see a default policy for $250,000, he says.  “We’re driving people down different lanes of the highway,” Mr. Cheyney says.  Consumer tracking is the foundation of an online advertising economy that racked up $23 billion in ad spending last year. Tracking activity is exploding. Researchers at AT&T Labs and Worcester Polytechnic Institute last fall found tracking technology on 80% of 1,000 popular sites, up from 40% of those sites in 2005.

The Journal found tracking files that collect sensitive health and financial data. On Encyclopaedia Britannica Inc.’s dictionary website Merriam-Webster.com, one tracking file from Healthline Networks Inc., an ad network, scans the page a user is viewing and targets ads related to what it sees there. So, for example, a person looking up depression-related words could see Healthline ads for depression treatments on that page—and on subsequent pages viewed on other sites.  Healthline says it doesn’t let advertisers track users around the Internet who have viewed sensitive topics such as HIV/AIDS, sexually transmitted diseases, eating disorders and impotence. The company does let advertisers track people with bipolar disorder, overactive bladder and anxiety, according to its marketing materials.  Targeted ads can get personal. Last year, Julia Preston, a 32-year-old education-software designer in Austin, Texas, researched uterine disorders online. Soon after, she started noticing fertility ads on sites she visited. She now knows she doesn’t have a disorder, but still gets the ads.  It’s “unnerving,” she says.

Tracking became possible in 1994 when the tiny text files called cookies were introduced in an early browser, Netscape Navigator. Their purpose was user convenience: remembering contents of Web shopping carts.  Back then, online advertising barely existed. The first banner ad appeared the same year. When online ads got rolling during the dot-com boom of the late 1990s, advertisers were buying ads based on proximity to content—shoe ads on fashion sites.  The dot-com bust triggered a power shift in online advertising, away from websites and toward advertisers. Advertisers began paying for ads only if someone clicked on them. Sites and ad networks began using cookies aggressively in hopes of showing ads to people most likely to click on them, thus getting paid.  Targeted ads command a premium. Last year, the average cost of a targeted ad was $4.12 per thousand viewers, compared with $1.98 per thousand viewers for an untargeted ad, according to an ad-industry-sponsored study in March.  The Journal examined three kinds of tracking technology—basic cookies as well as more powerful “Flash cookies” and bits of software code called “beacons.” 

More than half of the sites examined by the Journal installed 23 or more “third party” cookies. Dictionary.com installed the most, placing 159 third-party cookies.  Cookies are typically used by tracking companies to build lists of pages visited from a specific computer. A newer type of technology, beacons, can watch even more activity.  Beacons, also known as “Web bugs” and “pixels,” are small pieces of software that run on a Web page. They can track what a user is doing on the page, including what is being typed or where the mouse is moving.  The majority of sites examined by the Journal placed at least seven beacons from outside companies. Dictionary.com had the most, 41, including several from companies that track health conditions and one that says it can target consumers by dozens of factors, including zip code and race.  Dictionary.com President Shravan Goli attributed the presence of so many tracking tools to the fact that the site was working with a large number of ad networks, each of which places its own cookies and beacons. After the Journal contacted the company, it cut the number of networks it uses and beefed up its privacy policy to more fully disclose its practices.  The widespread use of Adobe Systems Inc.’s Flash software to play videos online offers another opportunity to track people. Flash cookies originally were meant to remember users’ preferences, such as volume settings for online videos.

But Flash cookies can also be used by data collectors to re-install regular cookies that a user has deleted. This can circumvent a user’s attempt to avoid being tracked online. Adobe condemns the practice.  Most sites examined by the Journal installed no Flash cookies. Comcast.net installed 55.  That finding surprised the company, which said it was unaware of them. Comcast Corp. subsequently determined that it had used a piece of free software from a company called Clearspring Technologies Inc. to display a slideshow of celebrity photos on Comcast.net. The Flash cookies were installed on Comcast’s site by that slideshow, according to Comcast.  Clearspring, based in McLean, Va., says the 55 Flash cookies were a mistake. The company says it no longer uses Flash cookies for tracking.

CEO Hooman Radfar says Clearspring provides software and services to websites at no charge. In exchange, Clearspring collects data on consumers. It plans eventually to sell the data it collects to advertisers, he says, so that site users can be shown “ads that don’t suck.” Comcast’s data won’t be used, Clearspring says.  Wittingly or not, people pay a price in reduced privacy for the information and services they receive online. Dictionary.com, the site with the most tracking files, is a case study.  The site’s annual revenue, about $9 million in 2009 according to an SEC filing, means the site is too small to support an extensive ad-sales team. So it needs to rely on the national ad-placing networks, whose business model is built on tracking.   Dictionary.com executives say the trade-off is fair for their users, who get free access to its dictionary and thesaurus service.  “Whether it’s one or 10 cookies, it doesn’t have any impact on the customer experience, and we disclose we do it,” says Dictionary.com spokesman Nicholas Graham. “So what’s the beef?”

The problem, say some industry veterans, is that so much consumer data is now up for sale, and there are no legal limits on how that data can be used.  Until recently, targeting consumers by health or financial status was considered off-limits by many large Internet ad companies. Now, some aim to take targeting to a new level by tapping online social networks.  Media6Degrees Inc., whose technology was found on three sites by the Journal, is pitching banks to use its data to size up consumers based on their social connections. The idea is that the creditworthy tend to hang out with the creditworthy, and deadbeats with deadbeats.  “There are applications of this technology that can be very powerful,” says Tom Phillips, CEO of Media6Degrees. “Who knows how far we’d take it?”

Read More: WSJ.com (Entire Article Here)