Jumping the 3 Big Hurdles to Predictive Modeling, Part 3: Make the Outputs Useful

March 5, 2019 Brady Walker

To ensure their predictive models drive meaningful results, retail analysts must tailor their modeling outputs to end-users’ unique needs. Prioritize scalability, applicability, and intelligibility.

By now, you know the drill: predictive modeling is an integral part of any marketing campaign aimed at acquiring new customers, forging strong relationships with repeat buyers, or preventing high-value customer churn.

Finding ways to achieve this final goal is particularly important, as a retailer can increase its overall revenue by as much as 5% simply by reducing churn among its highest-value customers by just 1%.

Granted, incorporating predictive modeling into your retail marketing strategy is challenging, but it’s almost always worthwhile in the end. Once you’ve pored over Part One and Part Two of this series—which cover how to prep your data and how to build the right model for your business needs, respectively—you’re ready to put the finishing touches on your burgeoning predictive modeling program.

The Third Challenge: Making the Outputs Useful

Even if you laboriously craft the perfect model for your brand’s needs, it’s not going to drive real results unless you implement it properly. To avoid getting in your model’s way during the implementation process, be sure to take the following three steps to position your predictive capabilities in a way that supercharges your business.

 

1. Set up the Model to Run at Scale

Time and again, we see retailers build powerful models that, unbeknownst to their engineers, are not equipped to last. Things fall apart, and predictive models that haven’t been designed to scale are no exception. Keeping your models relevant involves, among other things, feeding them fresh data at regular intervals. Only then can you transition from ad hoc predictive modeling to a fully-integrated predictive platform.

Let’s say you want to send automated emails to a specific customer segment: frequent shoppers who exhibit a strong affinity with your “fashionista” customer persona. This particular demographic frequently rejuvenates its wardrobe (and is willing to pay the price to do so), so you strategize holding their attention with information that should be of interest to them—say, routine trend updates with links to your brand’s products.

To accomplish this, you’ll need your model to pull data related to customer lifetime value, time since last purchase, engagement frequency, and behavioral profile. Adopting this panoramic perspective will ensure your target segment is sufficiently up-to-date before you upload it into your email service provider.

Models that haven’t been designed to run—and, if necessary, evolve—with such regularity can easily break down. Querying data taxes your system, and stitching data together and running ETL jobs (the extract, transform, and load database functions) at scale requires immense processing power.

As a result, you’ll want to make sure your data science team has tested your model under the appropriate conditions, within comparable timeframes. Or, if you’re investing in third-party software like Custora’s customer intelligence platform, don’t neglect to inquire about your prospective tools’ scaling capabilities.

 

2. Make the Model Accessible

To encourage organizational buy-in, your predictive model’s outputs need to be accessible to a broad swath of your team. Without easy accessibility, employees won’t be able to put to good use those beautiful predictive insights. If they’re not turning insights into actions, you just wasted a lot of time and resources, and you won’t be able to solve the business problems you set out to address in the first place.

Ideally, you should aim to integrate your model in a point-and-click manner, anticipating end-users’ needs. A truly smart model—a model with genuine utility—produces outputs that preemptively answer your questions before you’ve even fully formulated them in your head.

Ultimately, for your brand to reap the rewards of a cutting-edge predictive model, the model’s outputs must be intelligible, accessible, and ready for use in day-to-day activities like marketing campaign planning and reporting.

 

3. Make the Model Interpretable

After accessibility comes full-scale integration. Before introducing your model, outline a plan for its implementation. Debrief employees about their relationship to it, explaining how it functions, how it will affect their current work, and how they should make use of it moving forward.

At this stage, communication between teams is crucial. Expose the mechanics operating “under the model’s hood” so that all users will grasp the model’s impact. Using clear (though not necessarily “dumbed down”) terminology, elucidate what insights the model produces, how it arrives at its outputs, and how it will drive results for the brand. This will ease the process of aligning your teams and will prime your business for the launch of your new tool.

And at last, you’ve seen your project from start to finish. The problems are solved, the kinks are worked out, the hurdles are cleared. Go ahead. Cross that finish line.


This is Part Three of a three-part series. If you haven’t already, check out Part One and Part Two. If you’d like to learn about this topic in more detail, check out our webinar of the same name, Jumping the 3 Big Hurdles to Predictive Modeling.

 

Previous Article
Don’t Launch that Loyalty Program Until You Have the Right Customer Intelligence
Don’t Launch that Loyalty Program Until You Have the Right Customer Intelligence

When every retailer basically uses the same loyalty program structure, it’s not a value-add for your custom...

Next Article
Jumping the 3 Big Hurdles to Predictive Modeling, Part 2: Building the Model
Jumping the 3 Big Hurdles to Predictive Modeling, Part 2: Building the Model

Retailers should make a concerted effort to build predictive infrastructures whose validation frameworks, m...

×

Thank you!
Error - something went wrong!