6 Best Uses of Predictive Analytics for Cross-Sells and UpsellsFebruary 8, 2012 By Heather Fletcher
So says Stephen H. Yu, vice president of data strategies at Papillion, Neb.-based data provider Infogroup.
Different divisions handling different channels or various departments in charge of various products can cause problems for customers if companies decide to organize data based on those guidelines alone when creating predictive models. So instead, Yu suggests, “Companies should go through a paradigm shift towards 'customer-centric' marketing. Even the best-designed databases and models will turn out to be ineffective if marketers use such tools with 'division-centric' minds. That is how one customer ends up getting confusing offers in [a] short timeframe from the same company.”
This is just one bit of advice about how marketers can best use predictive analytics to identify cross-sell and upsell opportunities. More suggestions come from Yu and:
- Ozgur Dogan, vice president and general manager of the Data Solutions Group at Columbia, Md.-based marketing agency Merkle;
- Jeff Hassemer, vice president of product strategy at New York-based marketing services provider Experian Marketing Services;
- Paul McConville, senior vice president of sales and marketing at Vienna, Va.-based data provider TARGUSinfo;
- Stephanie Miller, vice president of email and digital services at Indianapolis-based marketing software provider Aprimo, a Teradata company;
- Barbara Nelson, a product manager of analytic and segmentation products for Little Rock, Ark.-based data solutions firm Acxiom Corporation;
- Wilson Raj, global product marketing principal for customer intelligence at Cary, N.C.-based business analytics software and services provider SAS; and
- Jesse Roberts, senior data strategist at Costa Mesa, Calif.-based marketing agency Rauxa.
Miller adds: “Use pattern analysis on all your digital data—from clickstream to tweets to email response data to score customers, campaigns and channels. This will help you improve your segments and personas. It will also help you identify your best customers, prospects and upsell opportunities, as well as the best offers to send to each group.”
2. Go ahead and get started. Miller says: "The end goal is to automate the offer placement based on analysis and predictive models for your particular customer base. However, every marketer can get started by using pattern analysis in your existing response data to identify the factors that lead to purchase behavior. Use that data (even through manual integrations at first) to improve your segmentations and send more relevant offers.”
3. Segment to aid model performance. Roberts says modeling can predict response rate, but can't explain why that's happening. So models shouldn't be considered customer profiles.
So leverage predictive modeling and segmentation tools simultaneously, Roberts advises. “When combined, you’ll find variances in performance by decile per segment,” he says. “This informs model depth selection on any [key performance indicator] basis: response, cost per response, conversion, cost per acquisition, cumulative values, etc.”
Dogan says: “Start with segmentation that captures the unique needs, product and channel preferences of distinct audience groups in the marketable customer universe. Use predictive modeling to determine the best targets for various products and services that the company offers. Develop a next-best product optimization process that takes cross-sell/upsell propensity, as well as expected profits, into account and optimizes the contact cadence.”
Hassemer says because of the variety of data now available—from sentiment analysis to neural responses—marketers can even travel beyond segmentation to “real-time micro-segmentation” that is often known by another name-personalization.
Go one step further than optimizing the models for real-time interactions with customers—know which prospects may eventually be interested in cross-sells and upsells, McConville says. He says: “Companies need to use predictive analytics to make better decisions at the moment of interaction with the prospect in order to optimize the engagement. Doing this allows the company to customize interactions regardless of channel, route prospects to their best agents in call centers, assign their best team to interested leads, etc.”
4. Do the prep work for the finished product. Develop short-, medium- and long-term goals, Roberts says. [See sidebar.]
5. Take a customer-centric view to upselling and cross-selling. Yu says: “Companies must see beyond their divisional (or product-centric) barriers and see each customer as a person with specific interests. The model scores can be used as tie-breakers among divisions (e.g., if the person’s 'cruise' score is higher than [the customer's cross-sell] 'rental car' score, offer cruise-related product even if his/her rental car score is 'relatively' high).”
Hassemer says perhaps the “single most critical component” of success here is having a multichannel view of the customer.
As Yu's comment implies, all this involves changing the company culture. Miller elaborates: “Improving segmentation based on predictive models is great, but it gets you no practical result unless the CRM and analytics teams partner with the creative/content teams. Collaboratively, identify the top-level segments differentiated by how customers buy. Then, within each of those segments, target offers based on demographics, job title geography, past purchase and other insights from the predictive models.
“Sometimes,” she continues, “the optimization is time of day for an email message. Sometimes it's the offer. Sometimes it's which offer goes on top or is referenced in the subject line. Subscriber satisfaction will only be complete when all parties collaborate together—and share in the learnings.”
For instance, in the data gathering effort in Tip 1, Nelson says social networks can yield keywords that can later be used in customer offers.
6. Measure and adjust. Roberts says: “Understand your campaigns’ lifecycles and identify appropriate tracking periods per campaign (per channel and campaign type). Validate your analytic tools used-compare [their performance] against prior campaigns to assess toolkit degradation. Measure incremental performance. Maintain a 'no touch' perpetual holdout group to assess organic market behavior, and compare [the control group] vs. the 'treat population to assess each KPI's incremental measure—the true gain of your marketing dollars.”
Raj says, among other outcomes, these predictive models for upsells and cross-sells should yield insight into “price sensitivities that can help with pricing strategies” and “identify key event triggers along the purchase path.”