SiteWit FAQs

General Questions

Installation

1. The bulk of your website traffic is monitored using tracking code snippets (Java scripts) that are generated for you during signup.  This mechanism is similar to those used by all other website analytic tools.  You can install the tracking code yourself or arrange for our technical support to assist you (or your website developer).
2. In order to track specific goals, such as contact form submissions or purchases, you will need to simply identify the corresponding goal page URL.  For example, you might associate a thank you page after a purchase with goal completion.  Of course, you will almost certainly have several types of goals.  For a bit more advanced tracking, you can install goal tracking code (as was done for the general tracking code).

You should install the goal tracking code in the processing page of a goal after it has been successfully accomplished.  For example, you should install the code after a purchase has been processed and the credit card has been charged.

Predictive Analytics

What are predictive analytics?

Predictive analytics or behavioral modeling techniques focus on making predictions about future events, typically at an individual level, drawing upon methods from reference disciplines such as statistics, marketing, data mining, and machine learning.  So, we might try to build a model that predicts whether a particular customer is likely to purchase a product, fill out a contact form, or download a brochure.  Knowledge discovery in databases (KDD) is viewed as an overarching process that includes data selection, preprocessing, transformation, and data mining activities (see below).  For instance, SiteWit uses a specialized data warehouse to preprocess your traffic and then builds data mining models based on the goals or desired behaviors specific to your website.  By automating the machine learning methods, we can deliver continuously refined models of how your best customers (and others) are interacting with your website.  The data mining models are used to score each visitor, as well as segment your traffic into categories such as loyal customers, prospects, one-time wonders, and zombies.  The SiteWit quality score is also used for pay per click management and campaign optimization.

The KDD Process for Predictive Analytics

What is data mining?

Data mining is the process of discovering interesting or even surprising patterns in large-scale databases that are useful or actionable in some concrete way.  Data mining algorithms uncover meaningful patterns in an automated, or at least semi-automated fashion.  Data mining methods are part of the broader discipline of machine learning, which focuses on pattern recognition, natural language processing, computer vision, and other challenges associated with artificial intelligence.  Several important data mining algorithms focus on inductive reasoning, learning patterns from large numbers of examples.  SiteWit uses several inductive approaches to uncover patterns in the large number of visits to your website.  Data mining algorithms are often categorized as supervised or unsupervised approaches to learning.  Supervised learning requires detailed cases with labeled outcomes, while unsupervised techniques simply analyze data to develop insights.  By defining specific goals for your website, SiteWit is able to use supervised algorithms for many predictive analytic tasks.  Though, unsupervised techniques are also used to develop insights regarding the structure of your website.  In fact, less formal learning strategies are infused throughout the SiteWit system.  We have tried to design a system that constantly learns from your website traffic patterns, adjusting key parameters, and refining the optimization and predictive analytic strategies.  By automating a collection of data mining techniques, we are able to deliver sophisticated services at a reasonable cost.  Predictive analytics provides the clearest example of data mining within SiteWit.  Goal-specific behavioral models are learned from many examples of visitor behaviors and are then used to predict future behaviors.  So, imagine we are predicting the likelihood of particular visitors purchasing on your website.  The behavioral models can be used to make a simple binary (true/false) purchase prediction, which can ultimately be compared with the actual user behavior (a true/false outcome), yielding a simple classification matrix representing the four combinations (see below).  These combinations of predicted and actual behaviors correspond to visitor segments.  It should be noted that behavioral models are used for more than simple predictions, playing a critical role in the derivation of quality scores for individual visitors.

Classification Matrix for Goal Behaviors and Predictive Analytics

How are visitors segmented into categories?

SiteWit predictive analytics are used for several purposes, such as scoring individual visitors, visitor segmentation, as well as campaign optimization.  Since the behavioral models can be used for a simple binary (true/false) prediction based on a specific goal, visitors can be classified based on the prediction and the actual outcome.  For example, whether a particular visitor was predicted to purchase and if they actually did.  This gives us four categories (or a two-by-two classification table) corresponding to the prediction (true/false) and actual goal outcome (true/false).  These simple categories are interesting and actionable!

1. Loyal Customers (True Positives)
   These are customers that have accomplished the goal and at the same time have been predicted to accomplish the goal by our data mining models.  Therefore, their behavior is the benchmark for the other customers.
2. Prospects (False Positives)
   These are customers that have not accomplished the desired goal, but have been predicted to accomplish the goal by our data mining models.  These prospects are very important to you.  The behavioral models think they should have purchased (or accomplished some other goal).  If you close even 10% of these prospects you will significantly improve your ROI.  You can tap these prospects using the SiteWit API to dynamically respond as they re-visit your website or combine this prospect list with a CRM system.
3. One-Time Wonders (True Negatives)
   These are customers that accomplished a specific goal.  However, the data mining models predicted they should not have done so.  These visitors demonstrated very different purchasing or goal behaviors, as compared with other customers.  These customers are an important source of information!  Again, you can use the SiteWit API to dynamically deliver a quick survey or instantly offer to contact them.  Why did they buy?  You can also follow-up via different channels once you integrate a CRM with your SiteWit account.
4. Zombies (False Negatives)
   These are the customers that score very low based on the behavioral models and may be wasting your resources.  They do not purchase or seem to be somewhat less interested in your products and services.  One of the goals of predictive analytics is to separate these visitors from the other more valuable visitor segments.  In addition, SiteWit campaign optimization is aimed at minimizing the negative impact of these visitors on your paid campaigns.

• Accuracy is simply the overall number of correct predictions (both true positives and true negatives), as compared with all predictions, (TP + TN) / (TP + FP + TN + FN).  In most website applications, the low conversion rates means accuracy is dominated by the large number of true negatives.
• Sensitivity or the true positive rate (TPR) is defined as the number of true positives out of both true positives and false negatives; TP / (TP + FN).  That is, the probability of predicting goal accomplishment for visitors that actually did have a goal.  A high sensitivity implies a low type II error rate (for false negatives or misses).

• Specificity or the true negative rate (TNR) is defined as the number of true negatives out of both true negatives and false positives; TN / (TN + FP).  That is, the probability of predicting no goal accomplishment for visitors that actually did not have a goal.  A high specificity implies a low type I error rate (for false positives or false alarms).  Within the SiteWit application, false positives represent the important prospects segment, so a higher number of false alarms is actually desirable (and intentionally cultivated as part of the scoring process).

• Positive predictive value (PPV) is simply a measure of how many visitors that are predicted positive are true positives, TP / (TP + FP).  This is clearly related to the overall number of visitors that really do accomplish goals, which is often a very low conversion rate for purchases and other such goals.  So, false positives are useful for identifying good future prospects.
• Negative predictive value (NPV) is a measure of how many visitors that are predicted negative are true negatives, TN / (FN + TN).  The false negatives represent visitors that actually (and unexpectedly) converted, these are upside surprises and are certainly worth investigating.
  

Search Engine Marketing (SEM) and Paid Search

How are campaigns optimized?

Campaign optimization begins with preprocessing and organizing your visitor data in our data warehouse.  Within the data warehouse, several different revenue attribution models are employed to allocate the costs and goal-specific revenue to individual clicks.  SiteWit then adjusts the analysis to fit your traffic volume.  Thousands of different traffic summaries are computed using marketing costs, attributed revenue, visitor quality based on predictive analytics, and other factors.  These individual traffic summaries are used to construct and rank campaign adjustments.  A weekly schedule is developed, with individual actions taken at varying levels, depending on the volume of traffic.  Overall campaign optimization is pursued conservatively, with adjustments made over several periods, so that more data can be used to support each action.  Therefore, you should expect more gradual improvement, rather than abrupt changes in campaign performance.

Can you accept or reject the optimization recommendations?

Yes.  You are in total control of the optimization process.  We provide recommendations on how to improve your campaigns.  However, you can accept or ignore the recommendations  SiteWit builds and maintains weekly schedules for your campaign that you can accept or ignore, minimizing the time you need to spend managing campaigns.  By using SiteWit, you can focus more energy on growing your business and less time mired in the details of managing your online marketing.  If at any time you are not satisfied with the results of campaign optimization, we will revert to your original campaign with just 1 click of the mouse (see our "1-click undo" above).

Can you see the changes made to the campaign during the optimization process?

Yes. SiteWit optimization works by developing weekly campaign recommendations based on an analysis of many different slices of your website traffic. You can review the list of recommendations, as well as the estimated impact these changes will have on your ROI. For example, a recommendation may involve stopping the campaign on weekday afternoons and/or in particular locations, or pausing underperforming keywords. Campaign optimization uses a somewhat conservative approach, making smaller adjustments a step at a time, rather than taking more drastic actions. This way each subsequent decision is informed by more historical data. Finally, a complete record of all the campaign optimization steps and recommendations is kept, along with measures of performance improvement.

Does website traffic volume affect campaign optimization?

Yes, the more data we have on a campaign the more specific the optimization can be.  Our data warehouse is designed to preprocess and organize your traffic for analysis.  If your campaign receives a large volume of traffic, any analyses can be performed at a much more detailed level, resulting in more refined campaign adjustments.  You are in complete control of the analytic strategy (or level of magnification) for each of your campaigns.  We may recommend upgrading your optimization level if we can make further improvements to your campaign and increase your ROI.  SiteWit will notify you when that happens and then you can decide if the additional optimization seems reasonable.

Search Engine Optimization (SEO) and Organic Search