Sometimes, machine learning can sound pretty high tech. But in a post on their blog, Schibsted Media Group—a company headquartered in Oslo that does business in 22 countries—makes a clear case for it. They used machine learning to develop a more targeted group for subscriptions and increased their telemarketing conversion rates by 540%, winning first prize for "Best Use of Data Analytics" at INMA World Congress 2017. (The photo is from a previous award win.)
Schibsted trained "a machine learning algorithm on a dataset of all logged-in users from a given observation period during which they do not have an active subscription... The algorithm learns the difference in behavior patterns between those that do not purchase and those that do purchase."
Variables they used:
- recency (how long since we have seen you on our site?);
- frequency (how many days did we see you during the observation period?);
- volume of content consumed;
- proportion of days visited that are weekend days (that's probably one for another blog post);
- number of devices used to visit site.
Feeding that information gave them a Subscription Purchase score that they used to determine who to target in their telemarketing. In a test, that group converted 6 times higher than a random group. Then they tried telemarketing for the first time with one of their newspapers—using this method—and the conversion rate was 8.8%, also significantly higher.
I gave a quick call this morning to my go-to for everything telemarketing, Cathy Karabetsos, president/CEO of SIPA member QCSS, to ask about this. As a vendor, QCSS, of course, doesn't have access to all that customer information. But the idea of getting a more targeted list of would-be-subscribers to call is an exciting proposition.
"It would be amazing," Karabetsos said. "We do an amazing job with the [customer] sets we do have, but this would take it to a much higher level... It's a great idea to talk to customers about gathering their open data in that way, and taking those data sets and applying it differently."
Schibsted tested more using the Subscription Purchase scores. They put an ad at the top of the news feed on their app to see which group would click on it more. The targeted group registered more than 4 times the impressions of the random group.
"The experiment proved that the users most likely to buy a subscription according to the model were both more likely to open the app to see the ad and more likely to click the ad if they saw it, and has thus demonstrated the potential of using such scores within the product," they wrote.
It's clear that once you have that Subscription Purchase score, there's a lot you can do with it. This could be "industry wide," Karabetsos said. "Taking the level of data of each individual company and seeing how [machine learning] will help them in their efforts" could significantly increase results, especially for telemarketing.
The more you can tailor your pitch to the audience, the more your telemarketing call will stand out. This implies better segregation of information (i.e., by age, profession, etc.) and customization of approaches for every audience segment to engage particular needs and concerns. Assimilating records based on specific interests, using techniques such as quizzes, drawings and other contests at your events, can be a way to both obtain opt-in for telemarketing calls and identify audience characteristics for better targeting.
Schibsted currently has the model running in production on a weekly basis for four publishers, with three more planned for roll-out in the coming quarter.