Solus Updates The 4th Pillar, Predictive Scores For Targeting Customers

A magnet attracting target customers

We would like to share an exciting update with you on Solus about Predictive Scores For Targeting Customers. But first, some context:

Systems of Intelligence

We built Solus; a personalization engine to be a System of Intelligence. This concept has been around for a few years now. In our context, they sit between Systems of Record (Data sources) and Systems of Engagement (marketing systems) and do automated decision-making and optimization. Or at least assist the decision-making and optimization. Given our focus on customer value, we knew the decision-making had to revolve around the following:

  • Who do I target (The List)
  • What value proposition do I target them with (The offering or offer)
  • What message/ tonality/ cadence
  • When do I target them?
  • What call-to-action or Response devices do I use?

Do a good job on these, and the customer retention and engagement world is yours. You might find Unlocking Insights from your Customer data worth a read too for more strategies.

The Fourth Pillar; Predictive Scores for Targeting Customers

In SOLUS we focused on building Recommender and Smart Campaign systems in the early couple of years. Then, more recently came the Insights system. We now have the fourth pillar in place – the Predictive Scores system for targeting customers.

 Four Pillars of Solus

Predictive Scores

Predictive Scores for targeting customers have been the bread and butter of the analytics world for years. They’re the go-to project for any data sciences team wanting to help marketers get better ROI. What we’ve done, is make this out of the box, incredibly usable, and very accurate.

 The scores themselves will focus on familiar goals:

  • Who is likely to buy in the next N days (7/15/30 days, for instance)
  • Who amongst the new customers is likely to Repeat
  • Who amongst the customer base is likely to Churn
  • Who amongst the lapsed base is likely to be won back

We won’t stop here, of course. We’re next setting our sights on that most elusive of concepts – CLTV.

The models behind these are ML models – gradient boosting/ random forest for the most part – and have been fine-tuned to work quickly, and smartly. What often happens in data science projects is that there are folks that love building models, but putting them into regular production is much harder as that needs a different set of chops altogether. We’ve solved this:

  • The predictive scores for targeting customers are available at a customer level as ready-made variables to use in campaign selections
  • Ready segments are available (Top 30% by Likely to Buy etc.)
  • The segments can be combined with other segments for targeting (High Propensity + High ABV, for instance)
  • A ready reporting interface with gain charts and model performance
  • Configurations through a UI to set the refresh and calibration cycles etc.

This here’s a sample chart for the models that will be part of the UI:

A chart showing predictive scores for targeting customers

 The Impact

  • The obvious impact is that of time usually taken to get this nature of tool kit in place – many months of work get instantly taken care of.
  • The business impact comes in sharper targeting = better conversion rates
  • The Customer impact comes in better relevance = lower dissonance and better LTV

We’re looking forward to putting our Predictive Modeling Software for targeting customers to work!