I know, I've "Geek-ized" a pop sensation. I couldn't help it. The temptation was too strong. It seems analytics has become the talk of the town in marketing. Everyone is discussing measurement and ROI and how analytics is helping them make better decisions about marketing spend.
However, as all contestants are not created equal, neither is analytics. The discussions I've seen span a wide spectrum of tactics described as analytics — from the most sophisticated cause-and-effect analysis to the most basic calls and clicks response tracking.
I categorize analytics discussions into three categories: arithmetic, mathematics and statistics:
Arithmetic. The first, and most basic, primarily delves into the primary reporting of response and conversion metrics, such as calls received or accounts generated.
Mathematics. The intermediate, combines reporting with high-level trends and insights obtained through effective data mining.
Statistics. The most advanced, delves into cause-and-effect relationships to determine the effects of the often tangled nature of various marketing stimuli.
None of these is right or wrong, and each level acts as a springboard to the next higher level, but each provides answers that can lead you in different directions. So let's look at three examples and what each area of analytics will tell us. (Note that while the situations I present here are real, the numbers and results listed are hypothetical.)
In our direct mail campaign, response to the listed URL is less than the response to the toll-free number. Do we drop the URL?
Arithmetic answer: The toll-free number had a 20 percent response rate and the vanity URL had a 10 percent click-through rate. Given an equal conversion rate, we are getting half the sales from the URL and, therefore, we should use only the phone number.
Mathematics answer: Among the three regions where we advertise, Region A has the highest response rate at 35 percent, followed by the Region B at 22 percent, with Region C coming in lowest at a dismal 3 percent. Interestingly, the trends are reversed for click-through rates, with Region C the highest at 25 percent, followed by Region B at 13 percent and Region A at 4 percent. Given an equal conversion rate for all regions, we should switch to using only the toll-free number in Region A.
Statistics answer: The customer segmentation model has revealed that the better performance of the toll-free number in Region C was the result of higher brand perception in the offline-senior segment versus lower brand recognition and acceptance in the young-techies segment; the former contributed to higher call-response performance and the latter segment took to the Web, improving the click-through rates. We should work to improve brand recognition uniformly in all regions among all segment, to improve our direct-response results to both our toll-free number and our URL.
DRTV advertising performance is going down. Should we reduce or abandon our efforts?
Arithmetic answer: Because our DRTV response rates have decreased over the past four quarters from 8 percent cumulative (call and click) to 5 percent, we advocate no further investment to DRTV investments.
Mathematics answer: We believe the decline is potentially due to creative fatigue and lack of new offers. Our competitors are updating their creative and offers, with results that consistently perform better than ours. As such, we advocate no further investment in DRTV advertisements until we develop new creative and offers.
Statistics answer: Over the same time period our DRTV results have declined, we have reduced our direct mail spending. Our marketing-mix models show that our DRTV and direct mail programs are highly correlated. The models also show that an increase of 20 percent in DRTV spending coupled with a 15 percent increase in direct mail spending will lead to a net ROI increase of 2.15.
We tested credit card advertising in newspapers and received a great response. Should we repeat it?
Arithmetic answer: Both the conversion and response rates for the recent newspaper placements were great. Conversion increased from 10 percent to 25 percent, giving us 1,200 more cardholders. We should definitely repeat these placements.
Mathematics answer: While the conversion rate jumped to 25 percent, our analysis of new card members reveals a high proportion of subprime borrowers with a high likelihood of defaulting on their payments. Further segmentation also shows a high portion of these lower quality card members were acquired through the general category newspaper placements rather than financial category placements, which result in higher quality conversions (though at a lower conversion rate). We recommend the placements be repeated, because the present value analysis of both the lower- and higher-quality acquisitions in relation to the cost of the investment still resulted in a net positive ROI.
Statistics answer: Our three-fold regression model analyzed the interaction between our website banner, SEM (Search Engine Marketing) and newspaper spending and found a net positive impact in conversion rate when a prospect is exposed to all of these marketing campaigns. However, newspapers seem to have a dampening effect on the results, especially when reviewed in combination with banners and SEM, separately; the latter of these resulting in the highest net present value. Given the aggregate positive ROI of newspapers though, and the above information, we suggest that future placements be run in markets with limited banner and SEM overlap.
I hope these examples help elucidate the differences between the three methodologies. My nomination for the winning "American Marketing Idol" goes to … Statistics. (Surprised, aren't you?)
And no, it doesn't (well, it shouldn't) take months of data-gathering to be able to use statistics for analysis, especially once you have built the models and only need to tweak them over time. Statistics is the only one of the three that improves with time, does not need to be completely repeated every time and provides you with the most confident answer.
Satnam Singh is Vice President of Analytics for Javelin.
