David Raab Article Archive

Marketing has always been more of an art than a science. Sure, there are plenty of statistics available: advertising budgets, audience figures, market shares, research reports, survey results, and sales by product, region, time period, and just about every other category imaginable. But these statistics are largely descriptive: they tell what happened, but not why. What really matters in marketing is the insight that leads to a desirable product or an effective advertising campaign or a superior customer management strategy. Statistics can measure the performance of new products, campaigns and strategies, and sometimes can even stimulate the thinking that leads to a new insight. But the famously high failure rates of new product introductions and new advertising campaigns show how little is really known about what it takes to make a marketing success. (Failure rates for new customer strategies are less well known, although probably just as bad.) If the engineering of new airplanes failed as often as the marketing of new soft drinks, few of us would be willing to fly.

Still, since marketers do have a lot of statistics available, they have long been heavy users of basic business intelligence tools for reporting and analysis. In fact, marketing systems have been leaders in incorporating such capabilities–several of the earliest multidimensional database technologies, for example, were tools for analyzing market data. Marketing systems have also incorporated predictive modeling and statistical forecasting functions that are still fairly new to general purpose business intelligence software. Some marketing software has also offered the mix of personal portals, key performance indicators and dashboard presentations that could be considered the state-of-the-art in business intelligence technology.

The most advanced marketing business intelligence functions have been part of systems for customer relationship management and database marketing. This makes perfect sense: because those systems capture customer responses directly, they come the closest to being able to measure the cause-and-effect relationships between marketing efforts and results. Those systems also typically incorporate a denormalized database that, although included primarily to support list selection requirements, also provides an excellent platform for many business intelligence applications.

But even the response analysis capabilities of CRM and database marketing systems provide a very limited view of the impact of marketing activities. Most analysis is still conducted on a campaign level, so even basic relationships betweem campaigns are largely invisible. In the relatively rare instances when analysis is conducted at the level of individual customers or of customer groups, the systems do little more than compare the metrics of one group versus another. This still gives no real sense of what causes any differences. In other words, two groups had been given different marketing treatments, the analysis will show which treatment was superior but not why. As a result, further refinement of the treatments requires either complex multi-factor test designs to isolate the impact of individual elements or, more likely, continued trial-and-error.

CRM and database marketers may complain that such tests are difficult and time-consuming. But they are an unobtainable luxury for marketers in other fields, who lack direct contact with their customers. For these marketers, business intelligence has largely been limited to conventional reporting and analysis of disconnected statistics. While better than no information at all, this is far from the comprehensive picture of business dynamics that business intelligence systems can provide to other business areas.

The equivalent business intelligence system for marketing would be based on a model that shows how changes in inputs interact to produce different results. Specialized models of this type do exist, for example to predict magazine circulation given assumptions about new subscriptions and renewal rates, or to predict credit card portfolio results based on acquisitions, interest fees and default rates. A generalized model would incorporate a broad range of factors, including ad spending by channel and customer segment; pricing and trade promotion policies; product features and costs; and competitive and macroeconomic effects. As even this brief list suggests, the details would have to be tailored to each company’s situation. But such tailoring is needed in any business intelligence implementation.

The trick to such a system is determining the relationships between inputs and results. These relationships structure the presentation of historical data–for example, the items displayed when a manager wants to dig into the causes behind a change in sales. The relationships also determine the inputs to whatever predictive model is used to simulate the results of alternative strategies.

Defining such relationships is a matter of custom analysis and intuition. Even the handful of advanced marketing optimization systems, from vendors including Fair Isaac, Veridiem and SAS, rely on expert users to develop the business models that are then loaded into their software. The role of those products is not to develop such models but to simplify their execution and reporting of actual results.

Some portions of the model development may eventually be automated through advanced analytics that infer relationships from historical data. But there are limits to what can be done: in marketing more than other business activities, the relationship between inputs and results is unpredictable. The intervening factor is effectiveness–spending the same amount of money on, say, advertising will have greatly different results depending on whether the actual campaign has powerful creative and intelligent media placement. This is an element of uncertainty not present in most other business investments. When I buy an oil tanker, for example, I have a very good idea of how much oil it can carry, how fast it will sail and what it will cost to run. I also know that if I buy another oil tanker next year, I’ll get very similar results. I can’t say the same about an investment in marketing.

In short, marketing is subject to inherent uncertainties that are not present in other business activities. This limits the precision of marketing business intelligence systems, even as it makes the control they provide more important. So we can expect continued expansion of the limited choices in business intelligence software for marketers. But, compared with other business intelligence tools, marketing systems will pay less attention to structured analysis of historical data and more to estimating results of the future.

* * *

David M. Raab is a Principal at Raab Associates Inc., a consultancy specializing in marketing technology and analytics. He can be reached at draab@raabassociates.com.