Marketing is always more effective when it is more targeted. As a result of integrating data and algorithms, marketers are now able to deliver a personalized customer experience at scale.
“By enhancing the customer experience, retailers can unleash entirely new approaches to customer engagement and interaction. With intelligent automation, they can identify customers’ anticipated needs at precise times and capture the right moment with the right offer in the pursuit of competitive advantage.”
That’s the observation made in The coming AI revolution in retail and consumer products, a global study co-sponsored by the IBM Institute of Business Value and the National Retail Federation.
A range of segments
There are various ways to target specific customers. Approaches range from lumping customers into very broadly defined categories to getting a lot more fine-tuned about the segments and responsive to individual customer behavior. In collaboration with Google, Deloitte published Digital transformation through data: a guide for retailers to drive value with data that took a closer look at these gradations.
It ranked them as follows:
- Limited segmentation: All users are analyzed in broad segments.
- Basic segmentation: Uses standard characteristics (e.g., gender, geography) for segmentation.
- Detailed segmentation: Segments are based on personal characteristics and behavior.
- Dynamic segmentation: The UX / UI can respond to a customer’s in-session behavior as he or she exhibits different segment characteristics.
Achieving the detailed level depends on much more data than the static kind that is used for basic segmentation and advancing to the dynamic level requires a level of automation that will enable recommendations and responses to go out in real-time.
The coming AI revolution in retail and consumer products invoked the women’s clothing store, Avenue Stores LLC as an example of dynamic segmentation. It explained that it brings together “data across multiple touchpoints, including in-store activities and market trend analysis, to learn and reason about what customers want and when they want it.” On that basis it can reach out to customers with communication tailored to their situation in real-time, which makes it possible to capture their attention while in “‘shopping mode.”
Marketing for loyalty
Sometimes the goal of personalizing marketing is not to entice customers into buying something from you now. Possibly, a sale won’t happen due to your physical store being closed down now, as is the case for many retailers under coronavirus, or general shifts in seasonal demand. But that doesn’t mean you should cease communication.
The Deloitte report noted that businesses have taken a new approach to fostering loyalty that contributes to the lifetime value of a customer, “shifting to propositional loyalty and developing experiences that delight customers or solve their problems.”
Being in touch with your customers to let them know you’re there for them without pressuring them to buy can pay off in winning their loyalty and business later. In this case, your automated messaging doesn’t have to respond to segment your audience, as you would be working off a general form of communication.
When you don’t have history
But what if you do need to sell your products now? Marketing recommendations can work even on the more basic level, not just for new customers for whom you have no history to flesh out a profile but for the type of marketing communication that depends on general trends. For example, a very broad segment of all people in the United States can work for promotions tied to events shared by all due to the calendar, whether it’s Mother’s Day, Memorial Day, July 4th, etc.
You don’t need to know much about your customer other than that they’ll know what these days are because they are on their calendars due to living in the United States for the trending algorithm to work well. That makes using this approach ideal for customers for whom you don’t have first-party data.
It doesn’t matter so much what they are normally interested in or what they’ve bought before when you’re sending out a marketing message about buying their mother something before May 10th. However, if you do have information about the customer, say you know they’ve ordered flowers for their mother last year, then you can combine the trending recommendation with what you know about their behavior.
Contextual clues
One level up from no history is working on what the customer indicates about her tastes and interests in the moment of browsing. That’s what fuels contextual recommendations that Amazon and other e-tailers have made us expect: the “people who bought that also bought this” recommendation, which can work to cross-sell and upsell.
You don’t have to know where the shopper is now or what they’ve bought in the past. The algorithms will create a general segment based on the selection of items alone and can make suggestions accordingly. Context can even be expanded beyond that to whatever may be indicated by the customer’s behavior in the moment.
As mentioned just above, Amazon’s platform is what normalized this kind of shopping experience, and it announced at the end of last year that it was bringing this capability to its AWS customer: “We’re pleased to announce support for contextual recommendations, through which you can improve relevance of recommendations by generating them within a context, for instance device type, location, time of day, etc.” It also goes on to explain the advantage of “Personalization for new/unidentified users even when the past interactions of these users are not known.”
Historical precedent
More data is always helpful in achieving greater segmentation, and a shopper’s browsing and purchase history is a great indicator of their taste and interest. The data science applied by one team for making predictions based on browsing history is explained in Targeted Advertising Based on Browsing History.
As many shoppers browse more than they buy, it would be helpful to discover what made them abandon the item or even the cart. Enough data history may reveal the answer.
For some it could be the price, and they may be interested but not enough to buy without a discount. All that is needed for them to complete the purchase may be a discount code or a message that the item is now on sale.
Other shoppers may hold because they are very cautious and are waiting on reviews from those who have experience with the item. For others, it could have just been a form of virtual window shopping, and they were never really serious about the purchase. While you could sometimes post product reviews that are not original to your site, as Target does for many cosmetic items reviewed on the brand site, there’s not much you could do about shoppers who are not going to buy.
The more history you get, though, the more insight you’ll have to make the right recommendations. You also can combine this approach with contextual clues to inform the algorithms that recommend items based on what was purchased by customers with similar purchase or browsing histories.
Data plus predictive marketing equal results
The more data you have to work with, the better the outcomes from your algorithmic predictions and automation. The proof of that is in tangible results. One of the examples the IBM report shared was of a retailer that used intelligent automation for more agile campaign launches. It had great stores of customer data that went untapped due to the difficulty it had in integrating it into its campaign. When it shifted to “AI-powered campaign automation software” it was able to advance from one-size-fits-most campaigns that took four days to “Rich, personalized campaigns” that were up and running in just 36 hours. These more rapidly deployed messages garnered better open rates than emails used in the past.