Predictive Versus Descriptive Modeling with Analytics

I wanted to post about approaches to using web analytics data that recognize its limitations, and its power.

Many organizations use historical analytics data as a basis for forecasting future growth, and establishing performance goals and budgets. This applicaton for analytics data can blur the distinction between predictive and descriptive data. Understanding this difference is critical to an effective analytics program. It generally falls to the analytics professional to ensure that the difference is clearly understood within the organization.

I'm going to start out with a couple of definitions. What do I mean when I say predictive versus descriptive modeling?

Predicitive modeling refers to a mathematical model that can accurately predict future outcomes. For instance, I know that if I apply sufficient heat to water, the water will reaach 100 degrees celsius and begin to boil (barring slight variations for altitude which are also predictable). The rate at which this happens and the amount of energy required can be mathematically described.


Descritive modeling refers to a mathematical model that describes historical events, and the presumed or real relationship between between elements that created them. For instance, yesterday when I went to the store to buy milk, it cost me $1.00 a litre, last month it was 95 cents, last year it was 80 cents.. Based on historical events, I assume it will cost me roughly $1.05 to buy a litre of milk next month.

Web analytics falls in to the latter category. It is a set of descriptive, historical statistics.

Past Versus Future Performance

I direct marketing activities for a division of a financial publisher. The company has an outstanding track record for identifying and recommending market-beating stocks. Historically, they've made recommendations that represented a lot of money for a lot of investors worldwide.

But at the end of each financial report that we send out....and those of virtually any financial advisory service or market report is a warning to readers something like:

Past performance may not be indicative of future performance.


***Web Analytics reports should carry the same user warnings***


Have any of you ever sat in a management meeting in which company representatives have said something like, "We have the data to demonstrate the relationship between ad spending and revenues, so if we want to grow sales by 20%, we just need to ratchet up our ad spending accordingly."?

This kind of thinking fails to recognize the rate of change in markets, technologies and the competitive landscape, and fails to factor in the concepts of resource scarcity and the law of diminishing returns, as well as the potential from economies of scale. That's without even considering the inherent flaws in any data gathering and processing system.

So, if the operation of complex markets can not be accurately predicted by simplified analytics models, what are the more tangible uses of historical analytics data?

1. To identify broken systems. A significant change in performance data can often indicate a technical problem, overloaded systems, broken links, faulty logic etc.

2. To select between alternatives. Analytics is particularly apt for testing market responses to different offers, creative, or sales processes with A/B or multivariate testing. It can also provide a guide as to which channels or markets tend to be most lucrative or most cost effective among existing channels. (The challenge, then, is often to find ways to expand the more lucrative channel).

3. To flag new market opportunities. A careful study of web analytics data can reveal new opportunities for cost savings, revenue generation or operational improvements.

4. To extablish a meaningful dialogue with existing and potential customers. Web analytics data can help us to learn about customer needs, desires and propensities. It can teach us the language that the customer uses to articulate their needs, so that we can respond meaninfully. It also can give us parameters to personalize the user experience to better meet their needs and create loyalty, trust, and ultimately customer satisfaction

One final VERY POPULAR use for historical analytics data

And hey,...IF your company persists in falling in to the infinite resources analytics model trap. IF they continue to hold fast to the belief that past performance predicts future performance, you can always use your historical analytics data as a roadmap to all of the external factors that caused your company to veer off its charted course.