Introduction

If you’re a digital marketer, you most likely share at least two common goals with all other marketers: working toward a holistic view of customers and predicting how customers will respond to your marketing efforts. In other words, you are ready to move beyond traditional Web analytics and seek true digital intelligence about your customers so you can answer such questions as:

• Why do customers interact with your brand?
• How do customers engage with your digital properties (single or multidomain) across multiple experiences and touch points?
• What type of content do customers and prospects spend the most time engaging with and, more importantly, why?
• Which channels work together to attract and create higher-value traffic segments from an integrated marketing perspective?
• What types of interactions have predictive value in leading to conversions?
• What will happen to digital traffic if more advertising dollars are spent on one media channel versus another– and vice versa?

Finding the answers to these questions has been difficult. Many digital-savvy marketers have long been frustrated in their attempts to achieve a comprehensive, forward-looking understanding of customers. Part of the challenge has been the siloed nature and varying structures of digital data sources. Another part has been the limitations of Web analytics tools that aggregate and report on what happened in the past, but lack the sophistication of predictive marketing analytics.

The bottom line is that obtaining a holistic view of customers is a classic big data problem because it requires you to capture, prepare, manage, integrate and analyze huge amounts of digital data. The good news? It’s getting easier to do those things, thanks to developments in advanced analytics, data visualization and big data processing power.

In this paper – which is based on a webinar hosted by the Direct Marketing Association (DMA) and sponsored by SAS – we will take a look at how these technological advancements can enable you to become more predictive and prescriptive in your digital and integrated marketing efforts.

A History of Fragmented Data and Analytics Tools

Marketers have long had the challenge of stitching together different types of data and data sources to achieve their vision of integrated marketing. This includes:

• Internal first-party CRM and transactional data
• Digital analytics data from Web analytics and advertising offerings like Google Analytics, DoubleClick, Adobe Omniture or Right Media.
• Third-party data from data management platforms like Nielsen, BlueKai or X+1.

Unfortunately, it has been difficult to take advantage of such data sources because they have been stuck in silos. And marketers have struggled to bring together available online, offline and third-party data in a way that is logical, efficient and easy to use.

Until recently, many marketers depended primarily on third-party tools designed to use aggregated data to create reports that described what happened in the past. Obtaining an omnichannel, integrated view was extremely difficult. As a result, it was practically impossible to get a data-centric, comprehensive view of the customer that could feed integrated marketing analytics or, more specifically, prescriptive approaches.

While data-driven marketers and analysts have used powerful advanced analytics for many years to perform sophisticated analyses – such as regression, decision trees or clustering – they have been limited to using offline data, primarily due to restrictions on access rights to online data from third parties. On the flip side, Web and digital analytics tools primarily aggregated and reported on historical information and didn’t enable predictive analysis.

For the most part, Web and digital analytics tools are designed with the visualization of data as the primary driver for users, because data visualization enables a faster, deeper understanding of the insights and trends hidden within data in a more consumable manner. Their ease of use and visual appeal have helped marketers get a better understanding of the important trends and insights within data. And yet data visualization largely has been very descriptive in nature – that is, primarily about reporting, business intelligence and descriptive statistics.

For example, data visualization tools have provided basic charting techniques that allow marketers to see the distribution of males versus females, age bands, or the number of customers in specific states or geographies. This level of analysis will always be important; however, marketers today are asking bigger questions and need more sophisticated capabilities to take advantage of ever-increasing amounts of available consumer data.

Evolving Tools Enable Prescriptive Analysis

There have been simultaneous advancements in advanced analytics, data visualization capabilities and the availability of big data. What’s more, predictive analysis is becoming more accessible to a wider community. In the past few years, there’s been a notable upswing in new and incremental abilities to process very large amounts of information. Data repositories – from Hadoop environments to traditional relational databases like SAP, Teradata and Oracle – are getting bigger, stronger and faster. And now that it’s possible to handle very large amounts of information, we can approach digital data differently, no longer limited to using siloed digital data for basic retroactive analysis, ad hoc reporting and alerts.

Advancements in how we deal with big data allow us to take advantage of more sophisticated analytics. We’re starting to progress from descriptive analysis (What happened?) to diagnostic analysis (Why did it happen?), predictive analysis (What will happen?) and prescriptive analysis (How can we make it happen?). Predictive analytics and data mining thrive on detailed data. When we can bring together very granular digital data streams that highlight consumer behavior and feed that into predictive models, we can improve our approaches to segmentation, ad targeting and customer experience management. This will enable marketers to take advantage of predictive digital analytic scoring and business process rules together to meet the challenge of prescriptive marketing within their automation platforms – including outbound, inbound and personalization systems.

The biggest value of data-agnostic, advanced visualization platforms is that they allow you to see things you could never before see using traditional digital tools. They are also extremely easy to learn, use and communicate results with. As the famous mathematician John W. Tukey said in his 1977 book Exploratory Data Analysis, “The greatest value of a picture is when it forces us to notice what we never expected to see.” Today we have an attractive opportunity to watch predictive analytic and visualization technology mesh together with positive implications for integrated marketers.

Predictive Visual Analytics – an Example

To illustrate how advanced visual analytics can help organizations improve their approach toward digital intelligence, let’s go through an example that brings together data from online and offline sources:

• Online data. Includes a Web analytics feed from SAS.com combined with digital advertising data from Google DoubleClick.
• Offline data. Includes third-party marketing lifestyle information about digital visitors, associated with their location.

Now let’s look at how traffic arrives at SAS.com, both from a historical and predictive perspective.

Historical View

Let’s say that a manager asks, “What did our Web traffic look like over the last few months?” We can get the answer in just a few clicks (see Figure 1).

Predictive View

Now suppose the manager asks,“What’s going to happen to Web traffic in the next two weeks?” In one click, we can show a forecast of expected site traffic of any duration – no coding required.

What’s more, the technology uses champion-challenger forecasting. That means that multiple forecasting algorithms are applied to the data in near-real time, and the algorithm that is most statistically accurate in fitting the data is selected for the visualization. In other words, you get the most accurate result, no matter what your quantitative skill level is (see Figure 2).

[Download PDF to see Figures 1 & 2]

Improving the Forecast

We can also improve how this model predicts future website traffic by providing more information from which it can learn. In Figures 1 and 2, the visualization only represented visitors by date. Now we’ll add more data from originating visitor traffic sources – paid search, organic search and direct visitors who came to SAS.com without the stimulus of an advertisement.

By adding these three segments to the forecast model’s consideration, we can see (in Figure 3) that the confidence interval (i.e., best- and worst-case scenarios) of the prediction gets much tighter, showcasing accuracy improvement in the model’s prediction compared with the earlier iterations.

As a digital marketer – and more specifically, a digital advertiser or media planner – you have a limited amount of control over organic search traffic. You have more control over paid search, which is an ad-centric channel. What if we increased our paid search ad budget by varying amounts? What effect would that have on overall site traffic? That’s actually very easy to answer.

In Figure 4, we have simulated a 35 percent increase in paid search advertising. Let’s see how this change will affect the traffic pattern forecast for the entire website. With today’s ever-changing ad budgets and short time windows, having the ability to simulate increases or reductions in ad spending in different marketing channels can be very valuable.

[Download PDF to see Figures 3, 4 & 5]

Now we have two numbers representing website traffic in Figure 5. The baseline was the original prediction – 1,085. If we increase paid search by 35 percent, we can expect 1,323 visitors to the site. That means that a 35 percent increase in ad spending on paid search is predicted to produce a 22 percent increase in overall traffic over the next two weeks.

Based on how your organization manages budgets and decisions, you could explore different what-if scenarios. For example, you could determine if the impact of increasing paid search advertising by 25 percent or 45 percent would be worth the investment. This would be valuable information, indeed, for a manager or director.

Illustrating Correlations and Regression

As a second marketing application, let’s look at correlations. Correlations are beneficial when you don’t know much about the data and you want to improve your understanding of unexpected relationships or trends.

So let’s explore what factors may influence the quality of website conversions. For example, some conversions may generate high revenue, and others will generate low revenue. Once we understand the drivers of conversion, we may want to determine what drives higher-quality conversions.

Visitors come to SAS.com looking for information related to analytics, business intelligence or perhaps customer intelligence. When a visitor looks at a set of pages related to one of these product areas, it is indicative of that visitor’s contextual interest. One possible assessment we can make with correlation analysis is to determine if contextual interests in specific product areas have any relationship to conversion quality. And that’s just skimming the surface of what we can learn.

We can easily drag and drop in data from contextual product interest categories, traffic source origination, site engagement depth, segmentation, ad impressions and even third-party inferred data. Then in a few seconds, we will have a graphical depiction of where relationships are hot – shown below in white/pink/red.

[Download PDF to see Figure 6]

Now we can see a correlation between visitor engagement and conversion quality on our website. To dig deeper into this relationship, we can double-click the corresponding box highlighting the relationship, which launches a visual regression analysis, shown in Figure 7. The Y axis shows the quality of a site conversion, and the X axis demonstrates the depth of a visitor’s engagement. Note that traffic is color-coded, with different colors representing traffic frequency segments.

Now let’s look at this from a segmentation perspective – specifically, visitor traffic source. Traffic sources can vary – social media, blogs, search, display, etc. – and we may want to identify trends in uniquely behaving traffic sources. In Figure 8, we can quickly see that direct visitation represents the highest volume of traffic to the website. But we can also see that the intensity of the regression line decreases when we focus only on direct visitors (as compared to the overall population). That’s bad news from a marketing perspective, because it means that our largest segment of visitors needs to exhibit higher levels of engagement (pages viewed, time spent on-site) to achieve the same level of conversion quality compared with the overall population. This is a problem worthy of our attention.

To get our heads around this, we need to see a comparison. Figure 8 showcases how we can compare the analysis of the direct visitor segment on the left with a second visualization of the second most-voluminous segment – organic search – on the right. The organic search segment is behaving much more favorably, as indicated by the greater intensity of the regression line. These visitors are displaying a much lower amount of engagement – that is, they need to look at fewer pages and spend less time on the website before they start to convert at a higher-quality level.

[Download PDF to see Figures 7&8

Conclusion

In the past, siloed data gave marketers an incomplete view of their customers. The analysis tools that were available limited digital marketers to descriptive reporting of what happened in the past. With the latest advances in big data, advanced analytics and visualization, we can now perform holistic customer or prospect analysis using visualization tools that are data-agnostic, can integrate data from multiple sources and provide robust analysis options. These new capabilities enable us to be more prescriptive in our digital marketing efforts so we can determine the best action to take to meet our business objectives.