Marketing Analytics In Online Social Spaces

A GREAT MARKETING DIVIDE sep-arates creative types and analytic types. An ad agency lives or dies by its creative

work, and the creative director is usually the star. This is because creative can af-fect many aspects of a company’s brand and brand is what any company lives or dies by. In contrast, analytics has tradi-tionally been seen as primarily part of a direct response model or at least focused on more mundane matters such as ROI

(Return on Investment). The CEO would not think to ask the director of marketing analytics how the company’s reputation is holding up. At best, analytics would be used to design and evaluate the results of a focus group or survey. Marketing in on-line social spaces presents analytics with an opportunity to cross this great divide. There are three aspects of the social en-vironment that create the opportunity:• What was hidden is now visible.

Conversations about brand

preference, customer experience and other aspects of brand health are now regularly posted on blogs. A company no longer has to wonder what people think of them. They just need to listen.

• What was anecdotal is now accessible to automated capture. Online content may be downloaded and analyzed. Previously, market research relied on focus groups, surveys and clipping services.

• Analytic tools draw ever-deeper layers of meaning from data. Natural Language Processing and other technologies for handling large amounts of unstructured data are starting to render meaningful insights.

This article does not pretend to fore-cast what will happen in analytics for social marketing or the size of the op-portunity that is unfolding. Instead, I will walk through some basic concepts and

A GREA

arates analyticlives o

work, and the crea

early experiences based on my role as head of analytics at 360i, an indepen-dent digital marketing agency based in Atlanta. The lack of agreed upon defi-nitions and rapidly changing landscape means that I will miss much that is impor-tant. Even the term “social marketing” is sometimes used to mean marketing in online social media and sometimes used to mean marketing for social causes. What I can offer are those items and is-sues that are most in the mind of an an-alyst at a digital marketing agency. The foremost item is measurement.

UNDERSTANDING THE PERFORMANCE of digi-tal assets requires a different approach to traditional online media. The IAB (Inter-active Advertising Bureau) has standards for rich media metrics and WAA (Web Analytics Association) has developed standards for Web analytics, but met-rics for digital assets, such as widgets or games, are only recently finding agreed upon standards. For example, distribu-tion and interaction are the two key areas of measurement for widgets. Distribution

measures reflect the unique characteris-tic of this kind of media’s ability to “travel” though the Web. They include:• Placements is a measure of the

number of different “locations” that the widget has moved to. There are a number of variations on this theme, including unique placements, active placements and new placements.

• Grabs is the number of times that the widget has moved to a new placement. This is a measure of the action of creating a placement. It also has variations including grab rates and attempts.

• Transmission Rate is a key measure of virality, the ability of a widget to spread.

• Hubs is a measure of the domains that provide access to the widget for placement. Certain domains can be import distribution points for marketers to recognize and cultivate.

As a widget expands across the Web it creates a directed network that emanates from seed points. By contrast, interaction metrics are somewhat similar to more tra-ditional Web analytics metrics including: • Views – the number of times that

a widget is viewed is generally measured as the number of times the widget is loaded. Since we don’t know

the exact position of the widget on a page, this can lead to overstatement when the actual impression would require the user not just to load the page but also to page down.

• Clicks – the number of clicks a widget receives is a stronger measure of interaction.

• View Time – the amount of time the page the widget is on is viewed by the user.

• Interaction time – generally measured from some interaction start point, such as a mouse over, to the point at which the widget interactivity is terminated.

• Interaction details – most widgets offer the user a number of options for interaction. This is particularly true of multifunction and game widgets. Each action, as well as the ordering of all actions, provides insight into the appeal of the widget and changes that can be made to enhance that appeal.

• Ad supported metrics – widgets that carry ads must track interaction information about the widget itself and for any ad that the widget displays.

As with all metrics, the devil is in the details, and variances in the definition of these metrics can create significant differ-ences in our understanding of a widget’s performance. For example, terminating an

interaction metric on the last widget-based action will produce a smaller time value than terminating it when the user navigates away from the page hosting the widget. So measuring digital assets requires the kind of thoroughness and precision that analyt-ics professionals are familiar with. Let’s look at a campaign that leveraged digital assets to maximum benefit.

DIGITAL WORD OF MOUTH (DWOM) is de-fined by the Word-of-mouth Marketing Association as “giving people a reason to talk about your goods and services, and making it easier for that conversation to take place.” One of the most effective techniques in DWOM is to distribute con-tent, in the form of digital assets, to key influencers among the population you are trying to reach.

One of the problems with content on the Web is that it is easily “borrowed,” and the benefit derived from that content goes to the “borrower” rather than the creators. Working with 360i, NBC found a way to deal with this issue for some of its premier content. NBC shared SNL (Sat-urday Night Live) clips with major blogs and media sites but wrapped that content in a portable video player, which could be embedded on sites outside of YouTube and other video sites. 360i developed a

It’s fast, it’s easy and it’s FREE! Just visit: http://analytics.informs.org/

DWOM program to promote the clips on major blogs and online media outlets. In addition, 360i optimized the player for natural search.

The result was that the DWOM cam-paign drove 77 percent of all SNL vid-eo interactions and 63 percent of daily unique users. SNL videos were featured on hundreds of top sites including Pop-crunch, Defamer, Huffington Post, TV Squad and The New York Times. Video views reached almost 1 million on mul-tiple days during the campaign and hit 3.1 million views following the season finale.

But digital assets are just one building block in effective social media marketing. The following section describes a more comprehensive social media strategy.

IN 2007, H&R BLOCK engaged 360i to create a social media strategy for the 2008 tax season that would drive consid-eration for H&R Block’s online tax prod-ucts and build up an online presence for the brand. The goals of the campaign were to change perception of H&R Block as simply “brick and mortar” to a digital brand with a strong online presence, in-crease awareness of H&R Block’s online tax products and build online communi-ties that engage consumers with fun and educational information on taxes.

The campaign centered on creating a fun and educational online experi-ence for consumers looking to find out more about taxes and H&R Block’s tax solutions. Working together, H&R Block and 360i developed portable assets that proved taxes could be fun and en-gaging, including games and quizzes, which were then spread across the Web for consumers to find and share with friends.

In addition, 360i created a robust so-cial media strategy to distribute these as-sets and to support H&R Block’s fictional “brand evangelist,” Truman Greene. The social media program also included:

• creating a brand presence in social spaces;• giving consumers a platform to

interact with H&R Block on MySpace, Facebook and eHarmony;

• highlighting H&R Block’s digital products through interactive and sharable widgets; and

• starting conversations with online influencers [1] using Twitter.

Figure 1: The DWOM campaign drove SNL video views to record numbers and hit 3.1 million views following the season finale. Figure 2: H&R Block implemented a holistic social strategy based on listening to their customers and

delivering engaging, educational content.

Figure 3: Block of Fortune is a fun and engaging widget that provided helpful advice and information about taxes to consumers.

1. An online influencer is a person whose opinion matters to a sizable online audience, especially if it is in a particular area of expertise. A blog such as Fashionista can influence its readers to purchase fashion products or also blog about them. In the case of Twitter, an influencer would be someone with a large number of followers.

W H AT I S A N O N L I N E I N F L U E N C E R ?

The effort raised awareness about H&R Block’s online tax programs that strongly positioned them as a digital brand with a strong Web presence. The campaign results: • total brand awareness increased 52

percent;• word-of-mouth awareness increased

55 percent;• Internet advertising awareness

increased 171 percent.

Most importantly, the social marketing strategy put H&R Block directly in touch with its target customer. Many consumers began to ask questions about the brand’s products directly on Facebook, MySpace and Twitter. In these fun and useful com-munity-based environments, H&R Block became the go-to brand for knowledge about online tax preparation.

THE MORE STRATEGIC APPROACH has more strategic goals such as brand aware-ness and engagement. eMarketer esti-mates that U.S. advertisers will spend $4.7 billion on display ads and another $3.1 billion on other branding-oriented ads including rich media and video in 2009. Traditional brand measurements typically rely on a survey methodology but the Web offers alternatives.

At GMAC Insurance, I worked on a project for evaluating brand metrics explic-itly from customer behavior. GMAC part-nered with Insweb, an online Insurance agency that presented our offers together with competitive offers. By appropriately masking our competitors, Insweb was able to let us see when we won and lost competition for a customer and when the customer’s decision was based on price. We estimated that we could command a 7 percent average premium in the mar-ketplace and up to 30 percent for cus-tomer segments with clear preference for the GMAC brand. In an industry that runs an underwriting profit margin of around 5

percent, a premium of 7 percent is a key competitive advantage.

AUTOMATED TECHNIQUES FOR deriving in-formation from text data have come a long way in the past 10 years. However, most large advertisers still want to know that a human, preferably an expert in digital pub-lic relations rather than an intern, has read the blog posts as well. The major steps in mining text are: 1) information extraction, 2) analysis and 3) summarization.

Technorati and Google blog search provide broad search capability for the blogosphere. Most social networks allow

Figure 4: Online listening takes on many forms. A simple one-question poll can be more effective than extensive surveys that tend to have signifi-cant selection bias.

Figure 5: A Google blog search or Technorati gadget on your home page is a great way to keep track of your competitors, your executives, market senti-ment or new developments in your area of expertise.

Figure 6: A sample Radian 6 dashboard shows historic post counts and up to the minute tweets and video.

varying degrees of access to data though developer APIs. Tools such as Radian 6 or Buzz Logic provide great listening capability and summarization in dash-boards and widgets with self-explanatory names such as River of News or Influ-ence Viewer.

360i Advanced Analytics is also ex-perimenting with NLTK, an open source Natural Language Processing toolkit for the Python programming language. Our first use of NLTK was to build out “long tail” keywords for Paid Search cam-paigns. First we tapped Web server logs and social media sources for prospective

keywords. Promising keywords must be assigned CPCs (Cost per Click) that we are willing to pay in order to make a rea-sonable return. The Bayesian classifier in NLTK does a good job of matching pro-spective keywords with those in the ex-isting Paid Search lists. The trick is to design effective feature extractors. These are small Python functions which identify features of the text such as contains a brand term or location specific.

Accurately assessing sentiment, even simple sentiment polarity such as “positive” or “negative” is more difficult than processing keyword lists. Slang, sarcasm and irony, remarkably creative spellings, and tortuous negations con-spire to hide meaning. For example, a retailer may be interested in under-standing various aspects of shopping

behavior. Specific social shopping sites such as Kaboodle or Zebo do not have easy data extraction features, so we went to Twitter to see what people were doing as they shop. A simple extract from the Twitter API for the keyword “shopping” on Oct. 16, 2009, produced 1,090 tweets. Nineteen of these tweets contained the word “earn,” which is a feature of tweets from tweeters look-ing for marketers to earn money from home. After cleaning up the data set we start discovering useful tweets: • “shopping via world wide web.”• “At Target...I love shopping so early in

the day...I got the store to myself!”

• “Happy Friday early release day - Shopping at 1:30 - here I come”

And some less useful:• “Shopping”• “shopping jobs”• “Can I go shopping nooow pleaaase”

One of the nice things about the Twit-ter API is that the tweets can be tied to demographic information if the tweeter has provided it, so we can start to build a picture of which age groups, genders and geographies have distinct behaviors. Customer segments, sometimes repre-sented by online persona descriptions, may closely tie to online behavior.

It’s fast, it’s easy and it’s FREE! Just visit: http://analytics.informs.org/

Figure 7: Online Persona are a good way to summarize customer segments derived from demographic and online behavior in an accessible way for humans. The information provided by members of online social networks makes this straightforward.

Figure 8: A quick check in on Sunday, Nov. 29 shows no instances of “earn” in 1,441 tweets. I guess marketers get Sundays off. Five percent of tweets indicated they were online and 2 percent offline on Nov. 29 versus 3 percent and 1 percent, respectively, on Friday, Oct. 16, which may tell us more about tweeting behavior than shopping behavior.

The Yahoo Buzz graph shows the build of search volume and the spike for Black Friday, the first shopping day after Thanksgiving.

SOCIAL NETWORKING SITES TEND to value themselves based on their membership rather than visits to a Web site. One of my first a ssignments in the social net-working sphere was to develop models for monitoring and predicting the growth

of what was to become a large (80 mil-lion member) social network.

The key metric for viral growth is trans-mission rate. When the transmission rate is greater than one for an active popula-tion, growth becomes exponential. We fo-cused on refining the key touch-point for transmission: the invitation e-mail that an existing member sent to their friends to encourage them to join. Over the summer of 2005 this was consistently improved until, in September, we went viral. In the

next three months we picked up as many active members as we had in the previ-ous 12. Despite, or perhaps because of this, there was a consistent growth in en-gagement statistics such as time on site.

DIGITAL MARKETING ANALYTICS has tradi-tionally focused on monitoring Web site performance with tools such as Omniture Site Catalyst. Paid search has provided a

fertile field for forecasting and optimization models focused on tight ROI goals. Market-ing in social spaces requires rigorous and meaningful measurement but also opens up more diffuse challenges around brand health and consumer intent. !

Kevin Geraghty ([email protected]) is vice president of research and analytics for 360i, an independent digital marketing agency based in Atlanta. A graduate of the O.R. program at University College, Dublin, Ireland, he was a co-author of the 1985 Edelman Award finalist paper “Revenue Management Saves National Car Rental.” In 1996 he founded Revenue Research, Inc., a revenue management consulting firm.

Figure 9: Virality depends primarily on transmission rate. Tweaking the campaign factors that affect transmission rate is the key to campaign success.