Measurements are the lifeblood of all successful direct-to-customer businesses. Knowing what data to measure, what analytical tools to use, and then using them to do things that require statistical sophistication is a tough challenge.
To these challenges we must now add the growing trend to undertake real-time analysis of customer interaction (usually on the Web, but sometimes in the contact center) for cross-sells, upsells, and special targeted offers.
The vast majority of companies (even quite large ones) still rely on spreadsheets or spreadsheet-based tools to get the job done. Powerful though they are, spreadsheets work best as a simple reporting tool. They do not offer a full range of options, as dedicated analysis packages do.
Aside from reporting, there are three other commonly used analytical approaches: online analytical processing (OLAP), statistical analysis, and data mining. Essentially, OLAP tools provide a preconfigured set of relationships, links, and algorithms by which data can be interpreted. This is the typical slice-and-dice and drill-down approach to data analysis. But if you want to discover relationships among data that are not predetermined, you need to use statistical analysis or data mining. Statistical analysis tools require statisticians to use them correctly; data mining works the same way. What’s the difference?
If all of this sounds complicated, it is. Which is why there are service bureaus that will do all this for you: 21stAZMarketing (21stcm.com), Axciom (axciom.com), Epsilon (epsilon.com), Experian/Direct Tech (dmti.com), Knowledgebase Marketing (knowledgebasemarketing.com), and Miglautsch Marketing (migmar.com), to name a few.
On the other hand, you could build your own data marts and/or data warehouses, with analytical tools to slice and dice the data. Many large enterprises do just that, most often taking a best-of-breed approach, implementing multiple third-party components to form an integrated whole. In an environment with multiple divisions, multiple business rules, and multiple sales channels (encompassing multiple countries, more often than not), this may make great sense.
Many of today’s catalog order management systems include their own built-in analytical modules. Some of these have become quite impressive platforms in the last few years, making them a very attractive option for medium-sized organizations with budget constraints.
Finally, there is a host of business intelligence (BI) applications. This is a hot-button, buzzword niche right now, and that’s unfortunate, because it enshrouds almost every system in a befuddling cloud of hype as noxious as industrial-strength mosquito spray. It also blurs boundaries and functionality, and it engenders a web of entangling alliances that can disguise the true nature of what these systems do.
Each of the leading business intelligence systems has become an amalgam of tools that provide statistical analysis, OLAP, and interactive support for e-commerce. For instance, SPSS (spss.com), for many years a leader in the statistical analysis field, has branded its analysis tools DiscoverIT/BI Solutions, which include a data mining application (Clementine), time series analysis (DecisionTime and WhatIf), and advanced modeling and reporting tools (called, simply, “SPSS”). But these are a sideshow compared to solutions for business performance management, customer relationship management (CRM), the CRISP-DM data mining and “Web mining” suite, and a half-dozen other products and services, all built up from the basic BI solutions.
SAS (sas.com) is another case in point. Another long-time leader in statistical analysis, SAS now offers enterprise marketing automation, analytical CRM for retail and several other vertical markets, and the “e-discovery” module that provides click-stream behavior tracking with purchasing, customer service, demographic, and psychographic data to render a comprehensive customer view and provide support for personalizing customer transactions.
A mainstay of the analytics world is Cognos (cognos.com). Famous for its PowerPlay OLAP application and Impromptu query tool, the company now offers a suite of systems under the name of “e-Applications,” ranging from Sales Analysis, Inventory Analysis, and Procurement Analysis to A/R, A/P, and G/L Analysis programs. These are all repackaged versions of PowerPlay and Impromptu, dressed up to be more user-friendly.
Two of the largest leading BI vendors are Microstrategy and Business Objects. The latter, founded in 1990, claims to have “pioneered the modern business intelligence industry by inventing and patenting a ‘semantic layer’ that insulates users from the technical complexity of database systems.” Although this stretches the truth like a politician caught in the klieg lights, taken at face value it would put the blame for much of the current hype in the industry squarely on the shoulders of Business Objects. So be it.
The “semantic layer” that the company presents (which is literally an English-language structured query language interface) includes a raft of components under the headings of sales analytics, customer analytics, and campaign analytics. The latter includes more than forty pre-packaged “best practice analytics” to determine the critical success factors of marketing campaigns.
Business Objects also offers a Set Analyzer that lets you select a series of customer performance and demographic variables to build customer segments interactively and on the fly from the Business Objects data warehouse, which combines data from relational databases, production systems, OLAP servers, spreadsheets, and other sources.
MicroStrategy (microstrategy.com) is perhaps the most infamous of the BI companies spawned by the e-commerce bubble. Founder Michael Saylor notoriously promised a billion-dollar contribution to the nation’s schools just days before being forced to restate company earnings in March 2000, crushing the firm’s stock value and making it one of the first kingpins to fall in the ensuing collapse of dot-com valuations. This past summer the company ran into further (rather severe) financial difficulties.
Financial issues aside, MicroStrategy has one of the very best suites of analytical tools in the BI field, MicroStrategy7, which includes several OLAP and data mining modules, set analysis (such as those Business Objects has) and e-commerce analysis tools, and support for portals, extranets, and alternative reporting platforms. It also features application integration tools (supporting XML and other standards) as well as a Software Development Kit for customized implementations.
These are not the only two games in town, of course. Kana (kana.com), which recently bought BroadVision, offers a full range of CRM and campaign management tools (along with workflow and process automation) developed in Java to support business modeling, response analysis, real-time customer interaction analysis, and customer segmentation.
E.piphany (epiphany.com) has a similar set of object-oriented, Web-based tools (as well as application integration tools) with its E.5 Suite of analytic, personalization, and interaction platforms that offer collaborative filtering (a real-time data mining technique for cross- and upselling), real-time personalization, and promotional campaign analysis (including e-mail marketing) with data mining, data modeling, and OLAP functionality.
To close the loop and give you another insight into the complex tangle of the BI world, MicroStrategy7 is the foundation for the Xchange Campaign application offered by Xchange (xchange.com). Xchange uses algorithms developed by Charles Elkin at the University of California, San Diego, along with a few other tools to drive its Xchange Analytics, Xchange EnAct, Xchange Real Time, and Xchange Campaign modules. But at its heart is MicroStrategy7. What gives Xchange an advantage? Packaging for vertical markets, plus those “secret sauce” algorithms that enhance the data analysis.
To make selection of the right tools even harder, a company like Net Perceptions (netperceptions.com) doesn’t even focus on analysis as a reporting tool but rather as almost exclusively an interactive tool (which all of the above offer in their CRM or “real-time” modules, too). Using “collaborative filtering” techniques, Net Perceptions can evaluate current and historic customer behavior on the Web or in the call center to make on-the-fly cross-sell, upsell, and offer suggestions. Smith & Hawken achieved a 16.5% increase in the average size of online orders based on the use of the Net Perceptions Retail Revelations™ suite of software and services. And there are many other examples of similar success.
The BI field will become increasingly difficult to characterize as vendors incorporate tools, applications, algorithms, and methods from a growing array of sources and providers. To gain a competitive edge, many of these vendors will offer integration services or provide their applications on an outsourced basis.
Moreover, the algorithms and tools are becoming better every year. Retailers are just beginning to get religion about using analytical platforms, which is drawing more investment and players into the field and increasing academic interest in what is, after all, an arcane statistical specialty at its core.
Meanwhile, the above-mentioned systems and vendors are a good start in getting to know the field. They will most likely all be around as the business matures — with the possible exception of MicroStrategy. Just expect their wares to be rebranded a couple more times over the next few years.
Ernie Schell is president of Marketing Systems Analysis, Inc., Southampton, PA, and author of The Guide to Catalog Management Software. He can be reached at (215) 396-0660 or email@example.com.
Statistical modeling plays five fundamental roles:
Classification modeling predicts the outcome of a given set of characteristics, with the outcome sorted by classes (X or not-X, Y or not-Y).
Regression modeling predicts a numerical outcome (for example, projected average dollar value).
Association modeling predicts the likelihood of a second event if a first occurs (if customer buys X, will he buy Y?).
Sequencing modeling predicts the likely sequence of events (will a customer buy C before B before A?).
Clustering modeling describes associations of entities (what gets sold with what, or what age groups buy what products).