• Aucun résultat trouvé

FROM RETROSPECTIVE ANALYTICS TO MORE PROSPECTIVE ANALYTICS WITH DATA MINING

Dans le document Big Data, Mining, and Analytics (Page 170-173)

Mining and Analytics in E-Commerce

FROM RETROSPECTIVE ANALYTICS TO MORE PROSPECTIVE ANALYTICS WITH DATA MINING

The previous section focused on developing analytics to quickly display activities that are ongoing for particular e-commerce-based activities.

This gives analysts a solid understanding of what is happening to their organization and provides insights as to strategic initiatives relative to that information (e.g., hot spots and ad display or traffic origin and paid search optimization). The next step in analytics involves leveraging data resources to generate models that provide more mathematically and sta-tistically based information on possible reasons why things are happening

and what is likely to happen by identifying patterns and associations between variables.

In Chapter 3, we included a host of data mining methods that can be deployed in order to identify noteworthy patterns, relationships, segments, and clusters in data resources. In this section we will refer to more directed data mining approaches in leveraging data in the e-commerce spectrum.

Defining E-Commerce Models for Mining

E-commerce-based data models can take on a variety of forms. For many analysts, the term mining an e-commerce initiative refers to mining web-related data to determine optimal strategies for cross-selling products or simply displaying strategic information to visitors as they are navigating

FIGURE 7.2

(See color insert.) Website activity analytics. (From Google Analytics. Google and Google logo are registered trademarks of Google, Inc. Used with permission.)

a website. However, the evolution of the electronic commerce and com-munication world has opened up new data sources that can provide strate-gic information that doesn’t necessarily involve website-related activities.

Elements such as traditional email marketing or market correspondence through social media or mobile applications provide a robust source of data to better understand target markets (who they are, what they like, and what they do).

Better Understanding Email Tactics

Many email marketing campaigns focus on the acquisition and trans-mitting of emails to a population that hopefully includes a segment of a company’s target market, where content involves an email subject head and a message that delivers a chunk of product or service information to the recipient. Given the low cost of this tactic, many marketers emphasize email volume (numbers of email transmissions) and frequency (repetitive transmissions) with the hope that something will stick, where stick refers to recipients that actually open, read, and respond to email. The ultimate metric that depicts campaign success usually involves a conversion by a recipient (e.g., someone who opened the email and clicked on a link ask-ing for more information, or purchasask-ing or subscribask-ing to some offerask-ing).

Despite the seemingly low cost of this initiative, the process of acquir-ing relevant and authentic email addresses can be a time-consumacquir-ing and resource-utilizing activity, so why not better understand what types of emails are more effective in generating conversions from recipients, rather than just make do with an expected 0.01% response rate and potentially be viewed as a spammer? A strategist must consider the components of a traditional email from both the creator and the receiver of the message.

The electronic communication can be broken down into three main com-ponents: the sender ID, the subject heading, and the general content of the message. General strategic tactics are to focus on providing pertinent con-tent in the general concon-tent area; however, this may be shortsighted given the possibility that many receivers don’t ever open emails from unknown sources. Email marketers should consider format styles of sender ID and subject headings in optimizing email activities in junction with other rel-evant data (e.g., descriptive information of the recipient to focus on target market). These variables can be analyzed/mined to determine their effec-tiveness as measured by select performance or target metrics. These met-rics can include whether an email was opened, and or if it resulted in an

interaction by the recipient. An additional piece of information that can yield descriptive value to the model is to categorize the type of email con-tent (e.g., types and number of links and pictures and general structure).

The modeler can utilize directed mining techniques such as logistic regression or neural networks to estimate the likelihood an email will be opened or the likelihood a message may result in an interaction (click on a link). Generally, the entity/organization that is conducting the email cam-paign has some information on the email recipient, through either previ-ous interactions through web activities or product purchases. Descriptive information on the recipient in conjunction with email attributes provides the driver variables to corresponding performance metrics. Other opera-tional type variables, such as the number of emails sent, could also be included in the model. Consider the data illustration in Table 7.1.

The 0/1 binary enables analysts to model the likelihood that emails with a particular type of recipient and frequency, sent along with a particular sub-ject heading on the message, will be opened. This mining analysis provides significant value, as it can yield strategic information on the type of email subject heading that communicates to the market segment and invokes a response. However, marketing strategy should not incorporate deceptive tactics via email IDs and subject headings, where both variables should be relevant to the organization and the message to be communicated. Although the cost of emailing remains inexpensive on a relative basis to other market-ing initiatives, the effectiveness of the tactic should not be taken lightly. The combination of basic data resources and appropriate analytic capabilities can quickly yield insights as to better understanding the marketplace.

TABLE 7.1

Dans le document Big Data, Mining, and Analytics (Page 170-173)