As the retail sector grows increasingly reliant and focused on data and artificial intelligence (AI), it’s essential that retailers understand exactly how first-party data analysis can be crystalized into insights on customer behavior – and, in turn, a tangible competitive advantage.
To that end, consider the chart below, dubbed the “data + AI maturity” curve.
This is a simplified view of how a retailer’s data and AI capabilities (charted on the x-axis) directly correlate with the competitive advantage of its retail media network (charted on the y-axis). A general strategic approach following this curve will see retailers making incremental steps toward sophistication, inching ever closer to the vaunted “predictive analysis” that will allow them to anticipate customer needs and deliver finely tuned, personalized experiences.
This is all far easier said than done, however, and some steps are more important than others when it comes to intelligent targeting. Let’s look at the three most important milestones along the road to predictive analysis in the retail media context.
Clean, accepted data
The “on-ramp” to this curve for any retailer looking to harness the power of data and AI begins with a full view of clean and accepted data across all customer interactions and media placements, whether physical or digital, owned or rented. This data is crucial for understanding the opportunity, managing yield and accurately measuring campaign performance.
As technology formalizes retail media as a category, the chance to lead on metric integrity and data quality is significant. Understanding the unique count of customers along the journey through physical and digital touch points is also crucial, as duplicating customer counts to inflate the value of the media network is a risk to both trust and budget growth in the long term.
Let’s look at the three most important milestones along the road to predictive analysis in the retail media context.
Data is, ideally, streamed to a behavioral data platform (BDP) and stored in a secure, cloud-hosted data lake. Data from SaaS systems updates the BDP via a server-to-server connector. Data is then modeled and enriched by the BDP, where every customer interaction is unified to a single, holistic view of the customer.
This provides a single profile with an event history with thousands of records for each customer. While certainly a critical step, this really is the ground floor when it comes to media targeting – once this foundation is established, maturity can begin to build up.
The first level of true media targeting capability is delivering a message to a surface – a specific platform or device facing a target audience – based on its context. This is the most fundamental form of targeting and a crucial basis for all other targeting capabilities. The role of data at this stage is to forecast the inventory of placements available by placement type and location, which is key for retailers to manage their media network and optimize yield. Message relevance and brand safety are also dependent on this capability.
Implementing contextual targeting requires integration with a campaign booking platform and the media network. The booking platform sends a request to the media network for a list of placements that match the campaign criteria – say, for example, people who have bought Crest toothpaste in the last six months. The media network responds with a list of placements that match the criteria, and the booking platform sends the campaign creative to the media network to make the campaign live.
Validating campaign delivery is important to metric integrity, and the media network regularly updates the BDP when a placement is served with the number of impressions viewed and clicks received. Retailers should track foot traffic, dwell time and transaction value to support campaign performance.
Once contextual targeting is established, retailers can begin capitalizing on the single, holistic view of each customer that, ideally, begins to take shape back in the clean data phase. This is properly known as a “single customer view” (SCV) and serves as a comprehensive record of interactions, transactions and behaviors across all touch points and channels.
The advantage of combining this view with contextual targeting is considerable, as it allows for intricate tailoring of promotions and identification of high-value customers.
The next level of media targeting is rules-based segmentation. This is where retailers can target a message to a customer based on a set of rules informed by previous shopping behavior, demographics, location and other attributes.
Some examples of rules-based targeting include segmenting consumers actively considering purchasing a TV in the near future based on keyword search and category browsing history, while suppressing ads to customers who have recently purchased a TV. Retailers can also target customers who exclusively purchase one brand within a category with a campaign aimed at switching them to a competitor’s brand – an approach known as “conquest advertising.”
Another example is targeting customers who have recently purchased a product to purchase a complementary product, known as “upsell advertising,” which is common in the travel industry with insurance and car rentals, or in tech with accessories and warranties.
The role of data at this stage is precision. It requires a query upon the data lake of customer behavior, across all store channels – and the more recent the data, the better. While many point-of-sale (POS) systems are capable of providing this data, it is often not available in a timely manner. Batch processing of data is not suitable for real-time targeting and can be up to 24 hours behind, negatively affecting the performance of campaigns. Ideally, this data is streamed to the BDP with minimal latency and campaign rules that are evaluated in real time.
From this point on, we enter the realm of the “secret sauce” behind the most successful – and lucrative – utilizations of retail media. Predictive targeting is the pinnacle of the retail media targeting AI maturity curve. It allows retailers to target customers with highly personalized messages based on a predictive model of their future behavior. This capability is a game-changer because it enables retailers to anticipate customer needs and offer highly relevant products or services.
To achieve predictive targeting, retailers need to leverage machine learning and data science techniques to build predictive models based on customer behavior, preferences and purchase history. These models can identify patterns and correlations that human analysts might not be able to detect, allowing retailers to offer highly personalized messages and recommendations.
One of the most famous examples of predictive targeting is Amazon’s “customers who bought this also bought X” feature. This algorithm analyzes the purchase history of millions of customers to identify patterns and suggest related products. By using this feature, Amazon can offer highly personalized recommendations that increase customer loyalty and drive more sales.
Retailers can also use predictive targeting to anticipate customer behavior and offer highly relevant promotions. For example, if someone has recently purchased a new car, a retailer could offer them a promotion for car insurance or car accessories. By anticipating customer needs and offering highly relevant promotions, retailers can improve customer loyalty and drive more sales.
To achieve predictive targeting, retailers need to invest in advanced machine learning and data science capabilities. They also need to ensure that their data is clean, validated and enriched, so that predictive models can be built with confidence. With predictive targeting, retailers can offer personalized messages and recommendations that boost the potency of their campaigns.
While this primer can serve as a playbook for getting intelligent retail media targeting off the ground, it’s important to remember that this is not a one-size-fits-all solution. Each retailer has unique needs and challenges – not to mention unique customers – and their maturity curve may look different from another retailer’s.
No matter the implementation, however, this approach of layered incrementality will be consistent across all successfully established retail media networks. The journey to predictive analysis is a marathon, not a race, and the most important thing for any retailer looking to harness the power of data is to build their capabilities up piece by piece, no matter where they find themselves on the curve today.