Resources Home

Retailers, Your Contact Strategy Needs Machine Learning—and a CDP

Last updated December 3, 2021

Retailers, Your Contact Strategy Needs Machine Learning—and a CDP

Retailers, how effective is your contact strategy? Is it the most effective it could be?

Your goal is to communicate with consumers on the right channel, at the right time, with the right message to begin, nurture or reactivate a relationship with your brand.

As a retailer, you have an advantage few industries share—direct interactions with customers. This allows you to collect first-party data across all of your channels and use it for subsequent marketing and interactions. However, the growing list of channels, including stores, ecommerce sites, call centers as well as new modes of conversion such as connected devices and live streaming, is rapidly increasing the type and volume of customer data available. Without a clear strategy for collecting, unifying, and leveraging customer data, much of it will go to waste.

How a Customer Data Platform Can Help Retailers

Overcoming this challenge requires a customer data platform (CDP). A flexible, ideally schema-less CDP is the key to integrating different types of customer information into a single customer view in real time or close to it. As I’ve written before, a single customer view (SCV) is critical to retail success.

Once you’ve achieved an accurate SCV, the next step is to enhance and enrich that data with machine learning (ML) models. Some CDPs come with out-of-the-box prepackaged models including channel of preference, expected lifetime value, product/category preference as well as propensity to buy or more importantly, to churn, next best product, next channel, etc.

However, this no longer provides the level of customer insight needed to achieve digital transformation. You need the ability to create your own ML models within your CDP with a simple drag and drop interface to provide differentiated experiences for customers. This may include lookalike models or propensity scores for specific products like services and warranties or for specific actions like creating profiles or providing preferences.

This flexibility has far reaching repercussions on how you can optimize your marketing efforts beginning with customer acquisition. The measure of success has not changed. Represented by LTV/CAC, it is simply lifetime value over customer acquisition cost.

Many deployment platforms provide lookalike models. Typically, these platforms can take any group of customers and use machine learning to identify shoppers similar to those customers. This process is usually opaque to the retailer, with minimal levers for optimizing the audience selection.

Building an audience based on the entire customer universe or even the active customer universe is very broad. Although performance may be adequate, the audience is not optimized based on lifetime value, but rather on making a purchase. A good return on ad spend (ROAS) does not necessarily equate to a strong LTV/CAC. And, customer acquisition cost (CAC) is increasing. In a marketplace pulse analysis published in June 2021, the average cost-per-click (CPC) increased 28% between January 2020 June 2021. For Amazon, it went up 36%.

In addition, the expansion of content and streaming services is making it harder to reach customers. In a September study, Roku found that people are spending 78 more minutes per week streaming versus watching linear TV. That gap grew from 12 minutes in 2020.

Improving Audience Selection with Machine Learning

A CDP with extensive machine learning models can greatly improve audience selection. Using forecast lifetime value, you can build a lookalike model using only the customers with the highest lifetime value (LTV). Further refinements may include selecting customers with the highest LTV acquired through each channel, for example, the highest LTV customers originally acquired through facebook or a specific Google SEM campaign.

Depending on the retail sector, you may be able to select a starting audience with even more accuracy. For some sectors, trigger products or services signify shifts in customer behavior. The creation of a baby registry in home goods is a good example. The purchase of antioxidants in supplements is another. A research study at a vitamin retailer showed that retention and LTV increased after a customer first purchased an antioxidant. Building a lookalike model based on these triggers will further improve LTV/CAC.

Strong analytics come into play on identifying target platforms. With strong analytics and ML, you can identify your highest LTV customers (both historical and future). With that knowledge, you can use zero-party or third-party data to identify interests, sociodemographics, psychographics, content consumption, etc. This opens up opportunities to identify which platform may offer a better LTV/CAC. For example, you can compare a lookalike model on Facebook with a highly connected cross border audience filter and the same lookalike model applied to Viber ads. Even if that requires multiple campaigns management and comparison, it’s not a big concern with a powerful CDP.

Similarly, for current customers, a CDP with extensive machine learning models can improve retention and lifetime value. Most CDPs can deploy triggered campaigns like cart abandonment and browse abandonment, usually through email or SMS. They can also score propensity to shop or lapse models for deployment of reminders through owned channels.

In addition, flexible machine learning models can calculate next best channel or even next best time. A cart abandon contact can then be split into several campaigns, based on the channel customers are expected to engage with next. For example, a social campaign can be created and deployed from a machine learning model of all cart abandon customers expected to use the retailer’s social channel next. The same can be done for retargeting, email, SMS, etc.

Retailer marketers using CDPs that allow them to create their own machine learning models without coding enjoy even more advantages. They can score their customers with propensity models for any specific product or action. For example, an appliance retailer can build a predictive model to score all recent buyers on the probability of purchasing a service plan or insurance plan for their appliance. The retailer can then develop a contact strategy that reminds customers most likely to buy the service plan to do so. At the same time, a different contact strategy (and offer) can be used to generate interest among customers less likely to buy the service plan.

Customer Data for Better Product Launches

How else can customer data plus machine learning create unique and profitable contact strategies? The development of owned or in-house product lines is another way retailers can leverage customer data and ML. Owned product lines tend to have higher margin rates than carrying other manufacturers’ brands. Plus, the uniqueness of these products may improve repeat activity and retention.

Foot Locker recently announced the launch of its own apparel line. A strong CDP can assist efforts in identifying the subsegment of Foot Locker’s customer base that will be receptive to the new line and which ones may need to receive promotions to motivate them to try the new line. More importantly, Foot Locker can identify which customers not to invest marketing dollars on, because they have no interest in the new product line.

Product extensions targeting a specific audience can also benefit from the ability to develop custom machine learning models. IKEA recently announced a furniture set for gamers. Properly identifying gamers (both within the company’s active and inactive customer universe and through lookalike models) will be critical to the success of the marketing efforts for this product line.

Larger retailers are also expanding their product portfolios beyond their traditional roles to become more central to consumers’ lives. One of the best examples is Best Buy’s expansion into home healthcare technology. Positioning the new platform to consumers will require complex propensity models that are rarely, if ever, out-of-the-box. Retailers will either need technical support or an internal data science team to build these models. Alternatively, and more efficiently, retailers can partner with CDPs that provide that level of flexibility.

Recently, sustainability is becoming more and more prevalent, especially in sectors with known high-pollution impact, including apparel. Lululemon, Patagonia, REI, and others have all launched resale programs. Even large CPG brands like Adidas are launching their own version. These programs require extensive data collection and more importantly, strong propensity models to identify whom and how to contact regarding this program, as not all consumers will be receptive.

Lastly, retailers need to incorporate delivery preference (BOPIS, curbside pickup, etc.) into their contact strategy. This requires complex ML models that use customers’ geography and mobility in addition to historical patterns.

CDP for Retail Growth

In summary, retail brands need a CDP that provides extensive flexibility and ease of use in developing propensity scores and machine learning models. This will allow retail brands to accurately segment their customers in highly specialized ways for better targeting on new products or extensions of existing ones.

With increasing media costs and the proliferation of content, retailers need to achieve a much higher level of accuracy in finding the right consumers for the right product to achieve the highest possible return on a limited marketing budget. And flexible machine learning models are the most reliable way to achieve this.

To learn more about how Treasure Data CDP helps retailers exceed their goals, visit TreasureData.com.