Schooling ‘Curmudgeon’ Denny Hatch on ‘New-School’ Marketing
[Editor's Note: It's not often we get an offer to "school" Denny Hatch. So we took a long, hard look at Mark Klein's article before agreeing to run it. Upon review, we think he makes some legitimate points about what may really differentiate the "old school" and "new school" of direct marketing. What do you think? Is Klein "schooling" old school marketers? Or is he just using new words for the same tried and true "old rules" of direct marketing?]
Even old-school curmudgeons want to make their marketing communications relevant. They just don't know the best way to do it.
Target Marketing's Denny Hatch argued eloquently for the old-school approach in his recent column. Hatch is a smart and experienced direct marketer who is still relying on demographics and behavioral data, which he describes as "Hobbies and interests such as pets in the house, history of travel, etc." Under the heading "Direct Marketing 101" he says to make a relevant offer, an organization needs demographic and psychographic information. Do this, says Hatch, and "You'll generate some revenue." He's right, but it won't be nearly as much revenue as you'll receive if you use transaction data and individualized marketing.
The Holy Grail for marketers has always been individualized marketing, where each person receives a unique message/offer tailored to his or her own needs and wants. This means if your company sends 5 million emails, each email is different and includes offers appropriate to the recipient. Most companies still don't realize this is now feasible, practical and relatively easy, so instead they may send just one, or if they are sophisticated, perhaps 10 different messages to their 5 million customers.
New-School vs. Old-School Marketing
Good old-school marketing is based on demographics, psychographics and segmentation. Logistic regression is often used to determine the segments. The approach is to look at a set of variables related to buyers of a specific product (where they live, what are their interests, etc.) to find a group of other customers in the population who might buy that product. It's a product-centric approach, and the associated metrics are those that typically measure what happened in the past. Relevance hopefully comes from considering those demographics and hobbies.
In contrast, "new-school" marketing uses transaction data (who bought what, and when). Analysis is done at the individual customer level, recognizing that you can't predict what a particular customer will do unless you analyze at the customer level. The key metrics are predictive and forward-looking: "Risk Score," (measuring who will stay loyal and not defect); "Likely Buyer Score," (who is likely to purchase in the next 30 days); and purchase propensities that predict the probabilities of each customer buying each product or product category. Relevance comes from the offers made based on the purchase propensities.
Using transaction data is key. Companies already own this trove; they don't need to go outside to buy it. Further, by definition of a customer, they have it for each and every customer. Demographic and psychographic data is notoriously spotty and can be entirely missing for many customers. Most importantly, transaction data has been proven to be the most predictive. When it comes to purchasing, people vote with their wallets, not with their ZIP codes.
The Who, What, When Equation
The new, predictive metrics let you target the appropriate customers and offer them products matching their needs and wants. They also tackle what Hatch thinks is unknowable. He says, "What the CEOs, techies and bots do not reckon with is the 800-pound gorilla: WHEN." In fact, new-school marketers can identify when customers are likely to purchase with high accuracy.
Figure 1 (at right) shows the correlation between Likely Buyer Score and the percent of buyers in the following month. Customers were put in buckets according to their Likely Buyer scores, and then buying rates for each bucket were measured at the end of the next month. Clearly, there is a close correlation between the forecasted probability and actual customer behavior during the following month.
Figure 2 (at right) shows the results of an even more stringent test using product purchase propensities. For approximately 1 million customers and thousands of SKUs, probabilities were calculated for individual customers to make a repeat purchase of particular SKUs. Probabilities were assigned to the many combinations of individual customers and particular SKUs. Again, the customer/SKU combinations were bucketed according to probability of purchase. The chart shows how closely the predictions matched actual purchasing behavior in the next month.
The conclusion from these charts is crystal clear: Proper analytics can predict which customers are likely to buy, when they will buy and what products they will purchase. Even cross-sell (selling a product the customer has not previously purchased) can be predicted with similar accuracy.
These results depend on using transaction data for individualized analysis, and on making relevant offers to individual customers based on that analysis. The whole analysis-campaign process can be automated, and scales to tens of millions of customers.
What the Curmudgeons Are Missing
There is still a place for the old school data the curmudgeons love, but its role is more supplemental than fundamental. Use it to enrich the messaging, but recognize that transaction data is more predictive. The new metrics, based on that transaction data, are predictive rather than just measurements of past behavior. That makes them actionable. Changes in a customer's Risk Score, for example, can be used to trigger a retention campaign message. Likely Buyer Score can tell a marketer when to send some expensive collateral. Purchase propensities inform what offer to make.
The customer relevance that both curmudgeons and new school marketers want won't come from product-centric analysis. If you want individualized marketing, you need to analyze at the individual level, not the segment level, and you need to describe or profile customers based on customer-level analytics.