Michel Ballings

Social Media and Machine Learning Uncover a Wealth of New Customers

Facebook contains more than 65 million company pages that actively engage one billion users. For organizations selling goods and services from business to business, that’s a potential gold mine of customer prospects. Sifting through the vast expanse of possibilities has largely thwarted B2B companies though, as most use social media to qualify prospects based on intuition and a limited set of characteristics. Identifying and quantifying leads this way takes up valuable time that effective sales professionals could better use to convert leads.

Michel Ballings, an assistant professor in business analytics, recently authored a study with colleagues from Ghent University that was the first to show how machine learning techniques can help companies use social media to better qualify prospects. Ballings worked with Coca Cola to test the study’s models in the field and estimated that using lead lists generated from data that includes social media factors would result in a more than $21 million gain in revenue for the company in a one-year period.

The study used a prospect information list that Coke purchased to identify more than 9,000 potential bars and restaurants with websites and Facebook pages that could be mined for data. Ballings’ research identified 73 key pieces of customer information from the commercial data, 53 from website text, and 99 from Facebook pages and posts. Using machine learning algorithms, it compared prospects to Coke’s current customers to find the closest match.

“The advantage of machine learning is that it automatically learns the model structure, as opposed to a human specifying the model,” Ballings says. “The main challenge in qualifying leads is the lack of qualifying characteristics. With big data, auto screening can simply find which characteristics present are relevant.”

Ballings then gave lists to Coke’s salespeople that ranked prospects based on each of the data sources (commercial list, website, Facebook, or various combinations) to test which source provided the highest quality leads. After six months of sales calls, he then ran the experiment in a second phase, comparing the prospects his models identified to the prospects Coke’s sales people actually converted.

“This is a more apples-to-apples comparison and a more direct way to evaluate whether the prospects we identified as high priority were the correct ones,” Ballings said. “In both phases—comparing prospects to actual customers or to successful conversionsFacebook data provided the most powerful indicators.”

The top 10 variables in both phases came from Facebook data, with likes, check-ins, and were-heres heading the top of the list. None of the commercial data variables, the means by which most companies currently identify prospects, were among the top influential variables.

“Our study challenges the current best practices for B2B customer acquisition,” Ballings says. “These commercial lists are very expensive, and public Facebook and website data provide a lot better value. It makes sense for companies to work with Facebook, Inc. to see if the approach proposed in our scientific study can be extended to a commercial solution.”

The idea for the study grew out of a partnership with a former Master of Science in Business Analytics student working at Coke, and a recent graduate assisted with the study while working there.

“Social media improves the qualification of prospects, and that translates into real, positive financial gains for companies,” Ballings says. “Machine learning techniques are increasingly driving all kinds of strategic business initiatives, and companies are very interested in our students with skills in this domain.” —Katie Williams

Other Stories from this issue