Predicting the future with marketing analytics

By Published on .

Reprints Reprints

The most potent marketing mantra these days is that everything is about databases. But having a database is one thing; understanding it and ferreting out likely prospects is quite another.

That was Joe Somma’s mindset when he arrived at Independent Health two years ago as director-market intelligence. But what he found at the midsize health solutions company, headquartered in Buffalo, N.Y., was a sales staff burdened by an inefficient process of separating good prospects from bad.

As a result, sales was spending an inordinate amount of nonproductive time culling through businesses in eight western New York counties that were not yet Independent Health customers. Somma changed that by deploying an array of predictive analytics tools to separate the wheat from the chaff.

“The idea was to take our current database and begin to mine it for its potential,” Somma said. “We wanted to do two things: to limit the time our sales folks were spending with these enormous lists, and to target those prospects who are most appropriate for our company. It was all about creating, and putting into, a pre-ordered list of companies with a special affinity for Independent Health.”

Somma employed predictive analytics modeling from SPSS to create “look-alike” models. The company’s current lineup of 7,000 business customers was intensely sliced and diced, then mapped on the remaining 33,000 businesses in western New York state for similarities.

“Segmentation is a very interesting area of analysis,” said Colin Shearer, worldwide head for industry solutions at SPSS, which was acquired last fall by IBM Corp. “Traditional marketing segmentations are usually done by an assumption of what the segments are. But that’s not too sophisticated and perpetuates what marketers think their audience is. It’s almost self-fulfilling.”

Somma, who had used analytical tools in a previous marketing position, deployed such SPSS modules as Decision Trees (to discover relationships between groups, and to predict future events), Advanced Statistics (for multivariate testing to determine product interest levels) and Neural Networks (for uncovering complex relationships in a database).

“It was sort of interesting the way we did it,” Somma said. “We did two different models—one that looked for those companies that look like our current customers and another that looked for which companies would be profitable customers for us.”

his produced a “2-by-2 matrix,” Somma said, putting Independent Health prospects into one of four cells: high look-alike companies that would be profitable; high look-alike but less profitable companies; low look-alike but potentially profitable prospects; and those companies with low scores both on their similarity to current customers and their profit potential.

“From that, we went back to our sales team and said, ‘Here’s your strategy: Call the cash cows in that first quadrant first,’ ” Somma said. “Besides getting better discipline from an ordered list, sales was really able to attack the market more effectively.”

Somma said strong performers for the company tended to reside in certain Standard Industry Classification code categories, such as education, other health care companies and some technology groups. The process also screened for companies within particular revenue and staffing areas.

“We’d take the first three digits of the SIC codes, chop off the rest and use it for input into our model,” Somma said. “In looking for prospects we’d look for those SIC areas where we had heavy penetration.”

Driving these approaches, Somma said, was a recognition of “cost per capture.” For the companies in the less-desirable quadrants, Somma said, he “peeled them like onions” for hidden possibilities and adjusted his marketing outreach resources accordingly.

In reaching out to companies with high look-alike qualities but low profitability potential, Independent Health would minimize its marketing costs by employing lower-paid telemarketing staff as well as e-mail.

“And for those prospects where the look-alike is low but the profitability potential higher, and who are a little harder sell, then maybe we’d use recommendations—say, contacting a broker or lawyer who might get us in there,” Somma said.

Sometimes the modeling steered away from company firmagraphics entirely and looked instead at employee makeup. Somma said he worked to identify those companies whose staffers were most likely to take advantage of wellness programs, a value-added service that Independent Health offers its customers.

“We’d identify these companies, turn them over to our wellness teams and say, ‘Sic ’em!’ ” Somma said.

Independent Health’s results were strong. The company spent about $500,000 in assembling its prospecting database, and another $30,000 on the IBM SPSS software. From the company’s initial use of its 2-by-2 matrix of look-alike companies with high potential for profitability, ultimate profitability was 3.5 times the company’s normal average, Somma said. And based on the upfront investment, the return is running at 4 to 1, earning $4 for every $1 spent, he said.

“And if we don’t buy anything else from SPSS, but just renew our license, we’ll be in great shape going forward,” he said.

Somma believes that predictive tools, although powerful, can’t solve everything.

“Don’t think that just because it’s a computer, and an elegant algorithm, it will produce correct results,” Somma said. “You still need good analytical understanding and intervention.”

Objective: For sales to focus on the most likely new customers
Strategy: Employ “look-alike” modeling to identify prospects that most resemble current profitable customers
Results: Return on investment of 4 to 1, and profitability 3.5 times the company average

In this article:
Most Popular