Pretend you work in the marketing department of an online university. You have a list of hundreds of thousands of potential students that you could target with the school’s next direct mail campaign, but know the majority won’t respond. Constrained by a limited budget, you need to show results. How do you choose your targets, and how many flyers can you send – at great expense – before you risk losing money?
That was the dilemma faced by a client of Redmond, Washington-based data technology services firm Versium Analytics Inc., which answered with a question of its own: what if, within hours, we could build a customized predictive model for you, use this model to score to each of your potential targets, rank them in descending order, and prove that 80 per cent of your future students would be among the top 20 per cent, and that only two per cent would be among the bottom 40 per cent?
The university accepted Versium’s offer, the list of potential students was scored by the company’s LifeData Predictive Lead Score service, which was officially released on Jan. 27 – and the university saved 40 per cent of its marketing budget while retaining a 98 per cent enrollment rate compared to the previous year.
“And if you cut 40 per cent of your expense while still capturing 98 per cent of your converters, your ROI [return on investment] goes up considerably,” Versium CEO Chris Matty says.
The difference between traditional lead score offerings and Versium’s predictive lead score, Matty says, is that the former are usually based on simplistic rule-based models, while the latter is drawn from the company’s trademark LifeData analytics system, which contains more than one trillion consumer and business data attributes sourced in one of three ways: commercially; publicly; and privately, using the company’s own search engine infrastructure.
“It’s about a third, a third, a third,” Matty says. “In the same way that Google indexes the entire web, we index four core aspects – people, businesses, assets, and locations.”
For example, social media data – what people publicly like, tweet, and follow – can be legally harvested because it’s in the public domain. Government statistics are also fair game, and Versium outrightly purchases the rest – though “that’s the least interesting data, simply because if it’s commercially available, that means anybody else can buy it,” Matty says.
The company then uses matching technology to assign the data harvested to individual records – both people and households – and machine-learning to analyze it.
The result: While traditional rule-based marketing might divide potential customers based on between five and 20 attributes – the number of married women living in suburban homes, for example, or the average price and number of cars owned by the residents of a certain area – Versium’s machine-learning-based system can process as many as 3000 or 4000.
“We have a 400-million record on file of what cars people have driven going back 20 years,” Matty says, giving an example. “And someone who owns a hybrid is different from someone who drives a sports car.”
For Versium’s potential clients, the best part of its predictive lead score service is that businesses can use it to build custom models projecting consumer behaviour in a matter of hours, no expert contractors required. Leads are scored in real time, and organizations like the online university can easily upload and score their existing prospect lists, with the system identifying which of a company’s potential customers are most likely to buy.
“Thanks to machine-learning, you can assess each individual’s likelihood of converting, and get really granular, and that is a complete shift,” Matty says. “Very few people understand it, but the value is huge – we’ve seen increases on returns as high as 900 per cent.”