2011 Nov 01

New Metrics for Social Media
David M. Raab
Information Management
November / December 2011

Today’s marketers are increasingly – perhaps excessively – focused on exploring social media. Efforts to date have largely aimed at attracting attention, viagra buy both through viral promotions (think: YouTube videos) and longer-term programs to build relationships (friends, fans and followers). But more sophisticated marketers are already looking at applications beyond message delivery. Here are some challenges they’ll face.

– response tracking. Marketers want to know how many people have seen their messages, how many have reacted, and what those reactions were. At the simplest level, this means tracking actions such as page views, recommendations such as “likes” and “plus signs”, registrations as friends or followers, and content sharing by emailing a link or posting it on their home page. Services like bitly and ShareThis make it easy to track sharing behavior, including re-sharing by people who received the original share. Some services can trace the re-shares back to the original sharer, providing a measure of individual influence.

– audience profiles. Traffic counts are interesting, but marketers care even more about whom they are reaching. Direct audiences include visitors to your Web site, readers of the company blog, and social media connections. Some direct audiences come with an ID that can link to an existing profile, either within the same system (a Facebook friend) or by matching to external data (a LinkedIn profile). In other cases, audience members remain anonymous but it’s still possible to gather data such as location or company (based on IP address), interest (based on search terms), or approximate demographics (based on the referring Web site). Indirect audiences include people discussing or reading about you in forums beyond your control, including blogs, news sites, and interest groups. Vendors like Quantcast and Compete.com build profiles of Web site audiences; in other cases, a forum host may make this data available.

– monitoring. Marketers may give a higher priority to talking than listening, but the good ones do both. In the social media world, monitoring involves scanning blogs, public forums, and discussion groups for mentions of the company, its products, and competitors. The simplest forms of monitoring count these mentions, which can be useful to measure mindshare vs. competitors, track public attention during a crisis, and read the awareness generated by an outbound campaign.

– content analysis. More advanced monitoring goes beyond counting to evaluate the content of social messages. This can identify topics and report on positive or negative attitudes. Content analysis is sometimes done manually, but this gets expensive when message volumes are large. Automated content analysis relies on semantic techniques to make sense of natural language. These systems already do a good job at some tasks, such as extracting keywords to identify topics and products mentioned. More subtle interpretations, such as understanding positive or negative sentiments, are still problematic.

– connections. Social networks often expose formal relationships among individuals, such as whether they are friends or follow each other. But this data can be difficult to process effectively using standard relational databases. Alternative database engines have been designed for social analysis, including Cassandra, FluidDB, and Neo4j. Their key capability is to navigate a network of social connections, making it (relatively) easy to identify friends of friends or friends with shared attributes. This supports analyses and selections that are very difficult or resource-intensive using standard database technologies.

– traffic analysis. Formal relationships are just the start of social analysis. It’s often more interesting to understand the interactions among related individuals: how often they message each other, the size or duration of those messages, whether messages are broadcast to a group or directed at individuals, whether there are patterns and how these change, and so on. Traffic analysis is often used by military, security and law enforcement agencies to infer organizational relationships. But it can also be used by marketers to identify influencers and to track dissemination of marketing messages.

– influence. The number of connections someone has, the number of messages they transmit, and the number of visitors they receive are crude measures of influence. More subtle metrics include how often someone’s messages are retransmitted, linked to, or commented upon, as well as who is doing the retransmitting, linking, and commenting, and how those patterns are changing over time. Vendors including Klout, Social Report, and PeerIndex provide various influence measures.

– case management. Social media are increasingly used for customer service. In fact, since “service” interactions on social media are often visible to the public, the distinction between marketing and service has almost vanished. Social media systems therefore increasingly include a case management component, allowing company staff to identify, track and interact with individuals over time. These interactions could also be managed through a conventional customer service system, but the public nature of social interactions means additional supervision is needed to protect the company’s image. Social media cases also extend beyond service interactions to include sales conversations and conversations with influencers, such as press, bloggers, and expert users.

– cross platform identities. Most individuals maintain separate identities for different social media systems, as well as other channels such as email, telephone, and postal address. Even though major vendors including Facebook and Google offer unified sign-in services, substantial fragmentation is likely to continue. This means that marketers who want to build complete relationship profiles will still need to capture and cross-reference individual identities across platforms. Systems are already in place to handle this sort of identity resolution outside of social media, so it’s most likely that social identities will be managed within the same framework.

* * *

David M. Raab is a consultant specializing in marketing technology and analytics and author of the B2B Marketing Automation Vendor Selection Tool (www.raabguide.com). He can be reached at draab@raabassociates.com.

2011 Sep 01

Marketing Data Quality Enters a New Stage
David M. Raab
Information Management
September / October 2011

Data quality has never been easy, click but it used to be simple: marketers defined quality as the completeness and accuracy of information in their systems. This reflected the simple goal of the marketing database itself, doctor which was to create a central location holding all information about each customer. This unified view provided the foundation for all customer treatments, ensuring a consistent, coordinated experience across all channels throughout the customer life cycle.

But marketing is no longer so simple. Companies now interact with customers and prospects in many situations where the individual is not tied to an existing marketing database record. These include anonymous Web site visits, social content consumption or creation, appearance at known geographic locations, and membership in behaviorally-targeted Web audience segments. Although individuals can sometimes be identified in those situations, the programs are valuable even when they cannot. Pragmatic marketers will, and should, continue run them regardless of whether they fit the standard database marketing paradigm.

For the marketing database, the result is something like the transition from academic portraits, which stressed hyper-accurate portrayal of the details of each individual, to cubism, which presented a simultaneous view of loosely-connected fragments. Accuracy and completeness are reasonable criteria for evaluating a classical portrait by Ingres, but make little sense for judging a Picasso. As with modern art, modern marketing databases require a revolution in terminology to discuss them productively.

The revolution in marketing databases comes down to one word: effectiveness. An effective marketing program is worth doing even if you can’t link the resulting revenue to a customer profile in the marketing database. Of course, the desire for marketing effectiveness is no newer than the search for Truth in art. Both have been with us forever and both have seemed equally elusive. But while art remains a matter of taste, it’s now possible to assess the value of many marketing programs with enough precision to be useful.

The new, quasi-anonymous marketing programs are still data-driven. You may not know precisely who is reading your Tweets, passing by your store, visiting your Web site, in a cookie-based audience segment, or abandoning your shopping carts. But you know enough to send them a targeted message and to track the results. This knowledge is embodied in data: and where there’s data, there’s data quality.

What has changed is that the relevant data quality metrics are tied to the specific programs, while metrics for customer profiles are not. For example, the effectiveness of a marketing program based on sending messages to people passing by your store or looking for a new car depends greatly on how quickly you can react. If they’re already walked past your door (a matter of minutes) or bought their car (could be hours, days, or weeks), then it’s largely too late, although you might influence their next purchase of the same item. For those programs, speed is a critical data quality metric. In fact, several types of speed may matter: how quickly the data is gathered, how quickly it reaches the marketing system, how quickly a message can be transmitted, and how quickly you can see any response. The unit of measure may also vary: it could be milliseconds if you’re trying to snare a pedestrian passing by your storefront, while minutes or hours may suffice to influence a considered purchase like an automobile.

By contrast, the traditional measure of completeness will matter much less for many situation-based programs. You may reach only 2% of the people passing your store, but so long as you can profitably contact those you reach, the program is worthwhile. Things are different in a database-driven program where missing data could result in expensive messages to inappropriate recipients.

Other program-specific quality metrics may include:

– reliability: much of the new data will reside outside of your company, either because suppliers want it that way (for example, social networks may be unwilling to share member profiles) or because it changes too quickly to import (e.g., current location data). Reliability measures, such as how often the data is unavailable, how consistent are connection speeds, and how often there are changes in access methods or data formats, are critical in assessing an external source’s value.

– consistency: external providers often consolidate information from multiple sources. These inputs may themselves change, impacting accuracy, coverage, currency, or other aspects that affect the utility of the data for a marketing program.

– specificity: many of the new marketing programs gain their effectiveness from the precision of one piece of data, such as current location, product interest, recent purchases, or attitudes. The level of precision may vary as topics change or as sources evolve.

– clarity: categories such as product interest or content topic may be hard to interpret because different terms can describe the same concept. Marketing programs that rely on matching such inputs to offers need quality measures that track the ability map inputs to a known, stable taxonomy, or to quantify potentially vague measures such as “social influence”.

Part of the challenge in this new style of data quality is to understand which metrics apply to which programs. This requires investigatory analytics that quantify each attribute and then assess its impact on results. These analytics must also separate the impact of quality-related factors from other factors, such as offers and creative execution. Once the key quality-related metrics are identified, data quality programs can use them to monitor existing sources for changes, to assess potential new sources, and to prioritize the search for improvements.

This change in the mechanics of data quality implies a deeper change in the relationship of quality efforts to the marketing database and to marketers themselves. A central marketing database supporting many different programs encourages generic quality measures such as accuracy and completeness. But as marketers increasingly use specific data sets for specific programs, quality measures can be tied directly to each program’s unique requirements. This means that quality managers must work more closely with marketers to understand those requirements, to develop appropriate measures, and to correlate results with quality changes. This closer relationship is more demanding but will ultimately provide greater insights into marketing results and new opportunities for improvement.

* * *

David M. Raab is a consultant specializing in marketing technology and analytics and author of the B2B Marketing Automation Vendor Selection Tool (www.raabguide.com). He can be reached at draab@raabassociates.com.

2011 Jul 01

Marketing Attribution Beyond the Last Click
David M. Raab
Information Management
July / August 2011

As online advertising consumes a larger share of marketing budgets, order measuring its impact correctly becomes more important. Information management professionals will inevitably become involved because they control so much of the data needed to construct sound answers. So, whether you work in IT or marketing, it’s worth taking some time to understand the issues.

– fractional attribution. If attribution experts agree on anything, it’s their disdain for the most common approach to online marketing measurement. This method, called “last click attribution”, assigns all credit for each sale to the last message received. The obvious problem is that this ignores the impact of all previous messages. One common alternative, assigning all credit to the first message a customer received, has the same flaw. Most discussion among online attribution experts is aimed at finding logical ways to split the credit among all messages.

This often involves assigning weights to messages based on their type (a display ad is assumed to have less impact than an email), when they were received relative to a purchase (later messages get higher weight than earlier messages), or simply by assigning an equal weight to each message. Weights can also be based on more elaborate statistical methods, but this requires structured tests that few marketers actually conduct. Regardless of how the revenue is allocated, the portion attributed to each message can be compared with the cost of the message to calculate a Return on Investment.

– incremental attribution. While fractional attribution methods often feel arbitrary, most marketers can find a rule they feel gives reasonable results. But the technique has a more fundamental flaw: it assumes that all revenue is the result of marketing messages, and that the sum of revenues created by each message equals total revenue. Neither assumption is correct. Most businesses have a base volume of revenue they would earn even if they did no advertising, at least in the short term. And sales that are driven by advertising are often the result of several messages working together; taking away one message might have no impact or a very high impact depending on the circumstances.

Marketers who recognize these issues are increasingly turning from fractional attribution to incremental attribution, which attempts to calculate the change in revenues resulting from a particular message. This can be measured directly through structured tests, although these are often hard to execute. A more common approach is based on stages in the sales process. A typical set include includes the initial contact; learning more about the product; and making the purchase. Messages are classified by the stage they support and buyers are tracked as they move through the stages. Marketers use this structure to compare the effectiveness of different messages in moving buyers from one stage to the next. This lets them estimate the incremental impact on cost and revenue of sending on different messages, allowing a meaningful ROI calculation.

– cross channel attribution. Fractional and incremental attribution are both designed to measure messages across multiple marketing channels. But most implementations consider only the several digital channels, including paid search, display ads, email, mobile and social media. This is better than measuring one channel but still ignores offline activity. That’s a big problem because online marketing often influences offline sales, just as offline activities influence online sales. Combining online and offline measurement is more challenging than merging online channels because many offline channels don’t lend themselves to tracking individuals. You probably don’t know whether a particular person saw last night’s TV ad or what they bought in the grocery store this morning. And even when you can track an offline individual, perhaps through a loyalty program or credit card, you often can’t link their offline data to online identities such as cookies and email addresses.

These tracking problems are not insurmountable. Many vendors offer databases that merge online and offline identities with varying degrees of success. Surveys and research panels can provide detailed information on a sample of individuals, even though their actual identities are unknown. But most offline attribution programs still feed aggregate data for marketing spend and sales into marketing mix models, without attempting to track specific individuals. Although less precise, these can still show major correlations between online and offline activities. This is often enough to substantially change the value attributed to different marketing programs.

– long term impact. Attribution discussions are generally framed in terms of assigning credit for a single sale. But one message can actually affect multiple purchases and may also impact behaviors such as payments, service requests, and recommendations. In other words, serious marketing measurement must include multiple time periods and multiple transaction types as well as multiple channels. This usually requires tracking individuals over time, or at least tracking groups that can somehow be separated – for example, by making different offers in different cities and seeing how future behaviors in those cities diverge. This sort of analysis is much easier in industries where companies have direct customer relationships, such as bank accounts or service contracts. Even in those situations, it can be difficult to isolate the impact of a single marketing treatment. Many analyses therefore look at short-term results and use simulation models to project their long-term consequences.

Final thought: The common thread linking all attribution issues is the importance of data capture. Although online activities generate enormous volumes of data, much of it can only be used for attribution if tied to identifiers that persist over time and are recognized across channels. This doesn’t happen by accident: systems must be designed to capture identifying data in ways that are both useful to marketers and respectful of their customers’ privacy. Industry vendors are constantly offering new ways to achieve this, but the true strengths and weaknesses are rarely apparent – and may not even be understood by the vendors themselves. It will increasingly be the job of information management professionals to assess these methods and understand how they fit into the broader picture of marketing measurement.

* * *
David M. Raab is a consultant specializing in marketing technology and analytics and author of the B2B Marketing Automation Vendor Selection Tool (www.raabguide.com). He can be reached at draab@raabassociates.com.