A/B testing is a key tool for marketers. Now big data analytics provider Globys has announced a new kind of big-scale, machine-learning testing that it says can increase mobile customer retention by ten percent or more by going beyond A/B testing.

Called Mobile Occasions, the company said its solution gives marketers new capabilities for experimentation and machine learning. It is intended to apply scientific methods to automatically test, adapt and scale marketing treatments.

Globys describes itself as a "big data analytics company that specializes in contextual marketing and reporting for telecom and financial service providers."


Up to 'A Hundred Million Users'

Olly Downs, chief scientist at Globys, told CMSWire the Mobile Occasions solution uses "unsupervised data exploration and pattern discovery techniques together with advanced experimental design [and] optimization methods to learn, keep track of and act upon individual customer behavior."

Mobile Occasions is different from marketing automation systems, Downs said, because it can scale "for a hundred million users," employing "individual behavioral profiles." He added that the solution "extends well beyond programmatic A/B testing" in that it allows for "permutations to go beyond what the humans designing the experiments can preconceive [and] then applies machine learning for automatic adaptation and optimization."

This can involve "thousands of combinations of landing pages, or call center scripts, or SMS communications, or in-app messages," he said, in contrast to marketing automation systems that are rule-based and "require marketer intervention."

Globys Mobile Occasions - Longitudinal Customer View.png

A Customer View of transactional data over time and the impact of marketing on behavior

'Automatically Learns'

"We’re talking about a system that automatically learns the right interactions based on how customers react," Olly told us, "and also takes into account long term related impacts -- for example, how a certain communication may influence behavior that may impact profitability, versus response rate alone."

As one example, he envisioned a retailer launching a new product that wants "to know who the early adopters are, and who is most likely to tweet and tell all of their friends about it." Additionally, he said, they might want to ID the social influencers and determine how best to engage them "to drive viral promotion that accelerates adoption."

Similarly, a financial services provider might want to figure out how best to manage "each customer over their entire lifecycle." Driving certain behavior, such as buying, learning, or sharing, could involve incentives or, in some cases, no offer at all, he said.

3 Main Techniques

Three main techniques are employed:

  1.  In the Discovery phase, Mobile Occasions uses automated behavioral analytics to uncover insights.
  2. During Execution, the system provides new kinds of offers, messages, or contexts to see what is driving a specific action/behavior or to determine which marketing is no longer effective.
  3. Then, the Optimization Phase uses machine learning to tweak the performance.

How does their solution compare to, say, Google's marketing optimization?

Google "offers marketing optimization for online display advertising and search advertising only," Downs said, "and doesn’t provide rich tools for in-base marketing, beyond retargeting display and search ads." It is optimizing for short term responses, he said, such as CPM, cost per click, and, sometimes, Cost Per Action.

By contrast, he said, Mobile Occasions works on "orchestrating interactions with customers over days, weeks and months to affect long term improvements in engagement and retention of the customer." Its solution, he said, can go from testing ten messages to thousands in short cycles, and then testing personal communications – to tens of millions of customers, if need be.