Retelling the Data Story

THANKS TO a predictive model generated from a gutted and rebuilt customer database, The Parable Group garnered an impressive 31.9% response rate for a catalog mailing.

Furthermore, during this initial use of the model, the top 10% of customers had a 70% sales rate.

The mailing, which was sent out last November, consisted of 2.7 million catalogs Parable mailed on behalf of its independent member stores. Roughly 780,000 customers had been modeled.

Participating Parable stores realized a return on their investment of some 58%, with average per-catalog revenue of $18.50. Not bad for a piece that cost 45 cents to go into the mail.

The gains were all the sweeter for the headaches involved in setting up the model. Long before the model run that produced those results, the Parable database required a great deal of massaging.

The San Luis Obispo, CA-based company provides products and marketing materials to nearly 1,000 independent Christian merchandise retailers around the United States. Although nearly 300 of them are branded under the Parable name, they are still independent retailers. Their data comes to Parable's database in a variety of formats, and with different degrees of completeness.

This would make any data analyst's life difficult. But when Melissa Lundie, Parable's information and marketing strategy manager, examined the database last year, she found that some customers were listed as living thousands of miles away from their local outlet. For a retail chain, which considers distance from an outlet to be an attribute akin to recency, frequency and monetary value in importance, this was a serious problem.

It turned out that the outside contractor charged with maintaining the file was recording the location of a given customer's most recent purchase as the “local” store, even if the customer lived in California and visited a store in Maryland, says Jim Wheaton, principal of The Wheaton Group, a Chapel Hill, NC database strategy and design firm that assisted Lundie.

The new database also replaced attributes that had been set up as ranges, such as recency, with a series of open-ended fields, allowing for greater modeling flexibility.

The model was built from data supplied by fewer than 30 retailers. Only outlets that had recorded at least 75% of all sales made during the previous two years participated in the initial model build.

After bringing the database in house, Parable began running test models to determine which product categories were indicators of additional purchases.

“[Parable] has several of these magic categories,” Wheaton says. “And there is an additional variable — people who have bought across multiple product categories. If you have purchased from category ‘A,’ and only category ‘A’, that may suggest you won't be a good customer. But if you have purchased from category ‘B,’ you may be an excellent customer.”

In November, the company was confident enough in the model to roll out a live test. By this time 60 of Parable's stores had enough data to participate.

The company has not analyzed the unmodeled returns, but acknowledges that sales rates are considerably lower.

Parable's recent database efforts have focused on expanding the number of stores that can use the predictive model, despite not having the necessary data history. To facilitate this, Parable has run a series of “sensitivity models” in which it used just six month's worth of historical purchasing data.

While there was a falloff in the models' accuracy, their performance curves were still “highly acceptable,” according to Wheaton, although less reliable than for those with the fuller scope of data.

The Parable Group does not directly benefit from the mailings. The stores are independently owned, and Parable makes its money from supplying marketing material and database services. At first blush, this form of analysis might seem counterintuitive. Smarter mailings would seem to be an incentive to cut down the amount of Parable services a given store uses.

Parable's chief operating officer Tim Blair maintains that the program's goal is not to cut down mailings — and therefore reduce revenue for Parable — but rather to provide stores with the reasons and the means to solicit their customers more often.

“We don't want them to mail less, we want them to mail more often. They should have more advertising dollars to spend,” Blair says.

He continues, “What we are working on is not necessarily profitability, but where they can use their dollar. The average Parable Group store has $1.2 million in revenue a year. If they are going to spend between 3% and 4% a year on advertising, Parable wants to help them spread the budget around most effectively.”

To Blair's mind, this means soliciting everyone on the list at least once or twice a year, and targeting the higher deciles more often.