Co-op Databases Step Up Modeling

May 29, 2007 8:57 PM  By

The co-operative databases have sharpened up their modeling with a new generation of models for catalogers. New modeling initiatives from the co-ops are capable of producing better response and bigger list universes for catalogers.

Lafayette, CO-based Abacus is rolling out AbacusOne, which takes the best names across several different models. It is a simple concept: Run a variety of models and the names that rise to the top several times will outperform the names that only are selected by a single affinity or synergy or CHAID (chi-squared automatic interaction detector) model. The results show that catalogers’ prospect list universes of names responding above breakeven are significantly larger than the universe of names from building a series of models in rounds.

Westminster, CO-based Next Action’s Next Gen approaches its newest generation of modeling creatively, starting by response modeling a cataloger’s response files compared to their actual mailed files and doing a sophisticated matchback across an entire mailing to see which names they could have served up across the entire prospecting universe. Then NextGen builds a whole series of models to see which models would deliver the best performance, resulting in increases in response and larger profitable list universes.

Wiland Direct in Longmont, CO, aims to put the customer’s business priorities first when developing models. First, it prioritizes the cataloger’s business need for prospecting names in terms of power (to serve up the best responding names first), stability (placing great importance on producing models that respond the same time after time, and depth, (to maximize the available list universe). Wiland can deliver stability or confidence that the results will be the same cycle after cycle, by allowing the segment size of the models to vary depending on how many good names are available when the model is created.

So instead of forcing the segment size to a predetermined quantity of, say, 30,000, the model looks at the strength of the previous models and the quality of the names available and delivers the same quality of names in each segment rather than delivering the same quantity of names in each segment. So you might get 15,000 names in segment one rather than 30,000 names delivered by the previous model, but you can be more confident that these segment one names will respond much like the segment one names you mailed the last time.

Jim Coogan is president of Santa Fe, NM-based consultancy Catalog Marketing Economics.