This is a story about the age-old conflict between marketing and sales. It’s also about how the two sides came together to achieve some extraordinary things.
It started when the sales team at AT&T asked for a sales-likelihood model for help in long-distance acquisition. The firm’s marketing department, which jealousy controlled all targeting, couldn’t have been less enthusiastic.
“Their attitude was, ‘been there, done that,'” said Pete Huff, group director, AT&T Corp.
Among other things, marketing thought the focus should be on contact management, and that sales should contact every lead it was sent, no questions asked.
But sales, which had to make monthly numbers to make, didn’t see it that way. And Huff, whose unit produced the lists, was caught in the middle.
“My first meeting with the sales director, he said, ‘So you’re the list guy. I’ve been waiting to meet you,'” Huff said. “But once I understood what he wanted, it didn’t seem that unreasonable or overly ambitious. At the bare minimum, the guy wanted to understand the potential of the lists we were sending them in terms of expected yield and sales per hour.”
The two sides might have remained stuck in their silos forever, but an outside consulting firm was brought in and it agreed that a “sales-likelihood model made a lot of sense,” Huff said. “This broke the tie. Marketing relented.”
(If only that had happened back in the late ’90’s, when marketing built a model without consulting with sales. Except for trials, that project never saw the light of day.)
Once started, the modeling process took 18 months, and there was at least one return to the drawing board for the consulting firm. In the first pass at the model, “statistical metropolitan areas were one of the chief predictors when we’re sitting on a behavioral goldmine,” Huff explained.
But those kinks were ironed out, and AT&T ended up with some good if “fairly parsimonious” models, without too many predictors or ones coming from unusual sources.
The next step was validation in the field, using two external telemarketing agencies. That required close management.
“One of the vendors wanted to sort the list once they got it, in reverse order by last contact date, as is standard practice,” Huff said. “We didn’t want to bias the test.”
In the end, the results were impressive, and that led to full deployment.
The typical post-model sweep might cull 3 million names from a prospect universe of up to 70 million. However, the vendors could not always handle that number, and at times up to one million names would be dropped.
At first, determining which leads not to call was based on “old-fashioned business rules” — i.e., it was done by segment. But the segments were soon supplemented by deciles based on model scores. These were applied across the segments.
A similar change was made in the apportioning of leads to the telemarketing vendors. That, too, had been done by segment, but now lists were “randomized across all vendors,” meaning that everyone received the same list.
This move “eliminated the list as an easy excuse for sub-par performance,” Huff said. “And it didn’t require another dog fight with marketing. Because all of their choices were intact at VVD, we did not need to ask them for their permission.”
The result of all these actions? Sales per hour jumped from .35 to .45 over a three-month period. And the internal tensions were reduced (if not eliminated). A contributing factor was formation of a list council to discuss current list issues. It met twice and “got marketing back to the table for the first time in months,” Huff said. The marketing guys, who are smart if a little siloed, “conceded that we objectively could gauge quality of a list.” The final verdict? Marketing made a key contribution to increasing the sales per hour.
Sales also took a different view of lists. “There was less emotion,” Huff continued. “They were starting to say, ‘This is what we’ve got, what do we do with it?”
Network
