ActionNext℠ Automated Predictive Modeling Engine

Havas Helia has a 25-year heritage of developing successful predictive models for customer behaviors across the marketing lifecycle, from response to conversion to activation, cross-sell, up-sell and retention. Based on this experience, we created a proprietary predictive modeling process based on a series of statistical techniques that yields consistently robust solutions for clients and has won industry awards, including Most Innovative Solution in the Association of National Advertisers DMA Analytics Challenge. However, for prediction problems with many choices, such as next-likely-product in a retail context, our methodology wasn’t scalable. So, we developed ActionNext℠, an automated version of our technique that embeds our expertise into a predictive modeling engine that efficiently loops across any number of customer choices.

Cool, right?

We’ve now successfully used ActionNext℠ to solve many of our clients’ business challenges. The following case study shows how ActionNext℠ helped one of those clients achieve incremental ROI by personalizing the offers in a major seasonal direct mail campaign.

Case Study

Situation

Our retail client relied on the Christmas holiday season as the key to its annual performance. To maximize sales among Loyalty customers during the holidays, they sent a high-impact direct mail piece each year, designed to cut through the clutter.

In 2017, our client wanted to maximize their revenue and learn whether they could enhance campaign performance through personalization. As they did annually, our client wanted to target their most receptive loyalty customers, but this year, they wanted to test a direct mail piece with personalized products and offers against their traditional, one-size-fits-all DM promotional piece.

Solution

As the retailer’s CRM agency, we held their Loyalty customer database. We began by integrating all available data and pulling variables from it in order to put together a historical dataset. This allowed us to predict two main behaviors of interest for the holiday campaign: purchase likelihood and product choice.

We then developed a predictive model for purchase likelihood that we applied to all Loyalty customers to determine who should receive a marketing touch. Then we randomly assigned those qualified customers to two marketing treatments, either the control (one-size-fits-all) or the personalized test direct mail package. For the personalized test package recipients, we applied the ActionNext℠ model to select the top 20 products each customer should receive out of 100 total. The control (one-size-fits-all) recipients would see all 100 product offers.

Results

These efforts impacted campaign results in two key ways:

1. targeting Loyalty customers who were most likely to purchase after receiving a direct mail promotion
2. featuring products most likely to resonate with each Loyalty customer for the test package recipients

Targeting model impact: Loyalty customers who were randomly allocated to the no-mail holdout group spent an average of $97 during the holiday period, compared to customers who did not qualify for and therefore did not receive a direct mail package, who spent an average of $47 during the same period. This represents a lift of 106% from targeting the right customers.

ActionNext℠ Next-Likely-Product model impact: 300,000 Loyalty customers who received the personalized mail package spent an average of $114. By comparison, 2,700,000 Loyalty customers who received the one-size-fits-all package spent an average of $112. This statistically significant lift of $2 per Loyalty customer translated into $600,000 in incremental revenue. Leveraged at full volume across the 3,000,000 Loyalty customers targeted, this lift promised an incremental $6MM in future holiday campaigns.

Presenting only the 20 most relevant offers to each customer proved to be more efficient than delivering a package of 100 offers for customers to browse through in order to find what best met their needs. Recipients of the personalized package were also 23% more likely to redeem a coupon than control package recipients.

Key Takeaways

ActionNext℠ enabled us to employ our award-winning predictive modeling methodology efficiently at scale. In this case, the ActionNext℠ engine developed 100 product models in approximately 16 hours, which translates to a speed of 6 models per hour, or one every 10 minutes. Developing each model in a traditional manner would have required 8 hours each. That’s a 98% reduction in time.

And ActionNext℠ isn’t limited to enabling personalization in the direct mail channel. Because it predicts purchase likelihood across an array of product choices, it’s channel-agnostic, so it can be used to personalize any CRM touchpoint. Based on its success in this holiday pilot campaign, the retailer leveraged ActionNext℠ for personalization in email, SMS, social/display campaigns (via customer onboarding), as well as all direct mail campaigns across the calendar year. In fact, the 2018 holiday direct mail campaign yielded even stronger improvements we can attribute to ActionNext℠, with coupon redemption rates and incremental spend per customer both 29% higher than recipients of the one-size-fits-all direct mail version.

Other ActionNext℠ Applications

More of our clients have benefited from ActionNext℠, including a gaming console brand in need of game recommendations for their players, and a boutique hotel brand that wanted to add ancillary revenue from various sources during their guests’ stays.

The gaming console company wanted to personalize weekly email newsletter content based on titles each player was most likely to purchase. We leveraged ActionNext℠ to develop and refine a game recommendation engine across more than 300 titles, evolving over time to dynamically react to player preferences and game lifecycles. Recipients could click through the game title creative elements to download games directly from the online store.

The hotel chain wanted to personalize marketing across channels including email, website, social/online display, and in-person at check-in. We used ActionNext℠ to develop models for products and services such as spa treatments, room upgrades, Bed & Breakfast packages, parking, and family packages to enhance guests’ stays. The chain was then able to tailor content across guest touchpoints, including at the front desk to increase revenue and engender loyalty.

To learn more about ActionNext℠ and how it can help you improve your customer experience, email hello@havashelia.com.