UPDATING THE NEW DM: Yes Virginia, B-to-B Modeling Can Work
I’m often asked if modeling works as well for business-to business marketers as it does for their consumer counterparts.
The genesis of this question is the concern that B-to-Bers have less overlay or external data available to them than consumer marketers. As far as prospecting goes, this perception is correct. Consumer marketers – especially those with large files of potential customers – can almost always benefit from models based on the census data associated with five- and/or nine-digit ZIP codes. And the very largest mailers can profit from the use of household data combined with census data. In addition, consumer marketers frequently can make good use of such commercial products as Prizm or MicroVision, which define geographic areas in terms of census, financial and other available information.
B-to-B marketers do have less “firm-o-graphic” data to work with, but the age of a firm, its number of employees, and such information as SIC codes are among the key statistics that can be used as the basis for a meaningful market segmentation. Remember, not every segment or predictive model has to include dozens of variables. In fact, the reverse is true. Models built on just a handful of variables are the strongest of all. They’re the ones that last and perform well over long periods of time.
What about models that are used to predict the expected behavior of your own customers? Are B-to-B marketers still in a relatively weak position? It depends.
For Example…
Suppose each business customer was associated with only one name on the database, and that person was the decision-maker in the firm. Then the B-to-Ber would be in the same position as a consumer marketer, and could use the same modeling techniques and variables to predict future behavior.
Unfortunately, that isn’t the case for many B-to-B marketers. If your database is organized at the site level; if there are multiple customers – or more accurately, there are names of multiple employees at each site; if each employee plays a different role in the purchase process, and that role changes depending on the product, then the analysis and modeling task is more difficult.
More difficult, but certainly not impossible. Let’s stick with the basic facts of modeling. Models generally are built around some event that happened in the past and the assumption that future behavior (yours and theirs) will be similar to past behavior. Here are some facts that apply to most B-to-B promotions:
– In the campaign that is being used as the basis for a model, a different number of promotion pieces was mailed into each site.
– The number of pieces may or may not be related to the number of names identified at that site.
All of the above can be quantified, and the bottom line is:
– The site either responded one or more times, or it didn’t.
– The response may have been an inquiry or an order.
– If it was an inquiry, the conversion status is known.
– The value of the ultimate sale is known.
– Site purchase history (recency, frequency and monetary data, products, etc.) is known.
– Some firm demographic may or may not be known.
– The average response rate (for an inquiry or sale) is known.
– The value of the average order is known.
That campaign definitely can be modeled. The model may contain some variables a consumer mailer doesn’t have to be concerned with, such as the number of pieces divided by the number of named employees, or that the response rates increase by the square of the number of pieces, but all of this can be handled by the usual modeling analysis procedures.
In the end, the B-to-B marketer will have a model that will assign an expected probability of response and an expected average order size to each site, assuming he or she were to fundamentally repeat the same mailing strategy. Of course, the model’s purpose would be to help reshape that strategy by changing the frequency of contact, the offer, the sequence of mail and phone contacts, the creative and so on. But that’s a topic for another day.
