Many of you have probably heard the term “dimensions” used in the context of marketing. Whether they’re geographies, customer segments or products, dimensions are ways you can slice the data according to different categories.
Dimensions are used extensively in some areas of marketing analytics, such as marketing mix modeling. Another example is Web analytics, which involves fine-grained analysis of traffic flow by multiple dimensions such as traffic source or visitor segment. Dimensions also are available as reporting options within some platforms.
However, reporting and analytics for many digital channels have historically relied heavily on a hierarchical or pre-aggregated view of data, with dimensions only being used sparsely, and typically just one dimension at a time.
Now, all of this is changing: Digital marketing, along with marketing analytics, is moving toward an increasingly multidimensional approach. With the proliferation of devices and the increasing complexity of digital strategies, static hierarchies and pre-determined drill-down paths are becoming insufficient for analyzing marketing results.
Dimensions And Hierarchies
Hierarchies are approachable to everyone, since they resemble the nested folders (which are also heading the way of the dodo with the rise of search-based navigation) that we’ve all become accustomed to using when organizing files within a computer. But when you compare hierarchical data to dimensional data, you quickly realize that dimensions offer far more flexibility in the ways you can analyze results.
For example, when drilling down, hierarchies force you down a pre-determined path where you have to first split by a particular category such as geo, and then drill down further into a different category like products. In comparison, a dimensional breakdown lets you drill down or pivot in any order, whether it is by geo, product, brand, segment, or any other defined dimension.
This flexibility has always been important for business intelligence, where entire data structures, such as OLAP cubes, are designed around multidimensional organizations. Over the past few years, dimensions have become increasingly important in marketing as well. Here’s why dimensional analysis will become a critical skill for marketing analysts in the coming years.
Dimensions Are Powerful
A dimensional view of data provides analytical flexibility far above and beyond what hierarchies are capable of. Below are a few examples of why dimensions are powerful.
Dimensions give you the freedom to slice and dice data from many angles. Dimensions give you the ability to drill down and find root causes in any situation. For example, let’s say that there’s a revenue spike resulting from a product promotion across multiple geographies. If your hierarchy is split first by geo-targeted campaigns and then by product-specific ad groups, your campaigns will show a similar lift across the board and it will be difficult to pinpoint the root cause. On the other hand, with dimensions you have the choice of splitting first by product, so you can quickly identify the promotion as the cause of the spike.
Dimensions are critical for mapping marketing efforts to business outcomes. For analyzing marketing impact, dimensions are often a must. Marketing elasticity (the effect of marketing on sales) can vary significantly by brand, geography, product, and other variables. Marketing mix modeling typically requires splitting both marketing and sales data by equivalent dimensions for accurate marketing impact assessment and invest and divest recommendations. Even simpler analyses, such as split tests, require making informed choices on which dimensions to hold constant and which to randomize for the test to yield informative results.
Using dimensions can exponentially increase your learning opportunities. The potential for insight generation is exponentially greater (literally) for dimensionalized data compared with hierarchical data. In the example of geo-targeted campaigns and product-specific ad groups, the only insight you can obtain easily using a hierarchy is which products perform well for each region. But with a dimensionalized view, you can also see which regions are performing better for each product. With increasing amounts of categories, the number of possible comparisons for a dimensional view increases exponentially, while for hierarchies it increases linearly (if even that, since depth of hierarchies is usually limited).
Dimensions Are Becoming A Necessity
Recent trends in digital marketing are making the use of dimensions all but unavoidable both for execution and analytics. Below are just a few of the tailwinds.
Digital execution is becoming more sophisticated. The increasing sophistication of digital execution means that the number of total ad units will increase exponentially when using a hierarchical structure. Arguably, device proliferation was the forcing hand that put dimensional execution in the spotlight. An example of this is Enhanced Campaigns in Google that was introduced two years ago, which forced device targeting to be executed as bid modifiers (a dimensional adjustment) instead of breaking apart device-targeted campaigns.
Execution is increasingly tied to business data. Programmatic execution and data feeds have made it possible to incorporate business data to drive execution. A prime example of this are the various types of product ads we see today: Google’s Product Listing Ads, dynamic creatives showcasing individual product units, and most recently Facebook’s Product Ads. As more and more business data is brought in to inform execution, the dimensions present in business data will flow into marketing.
Integrated (multi-channel) digital campaigns are now ubiquitous. Today’s integrated marketing is about optimizing investment mix and timing in order to maximize total marketing impact. However, one of the challenges of cross-channel measurement is the lack of consistency in organization: Some data may be organized on audience segments, others on geographies or product lines. In order to bring all this together, it is necessary to harmonize the disparate data using a consistent dimensional taxonomy.
There is increasing pressure to demonstrate ROI. Focus on marketing accountability has been steadily rising over the past few years, and will continue to rise for the foreseeable future. As I mentioned previously, dimensions are oftentimes a must for measuring marketing impact accurately, whether using a sophisticated marketing mix model or a basic split test or correlation of investment vs. revenue deltas.
Getting To Know Dimensions
Dimensional analysis will be a must-have in the next-gen marketing analyst’s toolbox. So if you haven’t done too many multidimensional analyses before, how do you get started? Below are a few tips.
Data Preparation and Extraction:
If your platform provides reports of campaign structure, plus one or more additional dimensions, such as device, start using these additional dimensions in your analyses instead of relying on just your existing campaign structure.
If your campaign management tool allows tagging or labeling, tag ad groups across campaigns (or ad units across ad groups) to create new dimensions and expand the scope of possible insights.
When querying data, stop pre-aggregating everything into a single-dimensional view. Leave a few key columns (if using SQL) or fields (if using NoSQL) in the query output for further analysis.
Testing and Analysis:
Become comfortable with multivariate testing as well as A/B testing. Note that both A/B testing and multivariate testing have their place, so one is not necessarily superior over the other.
Graduate from t-tests and start using more generalized ANOVA tests where appropriate.
Become familiar with some basic multidimensional regression models such as fixed effects models and random effects models.
Comments