🕗 5 min read ・ Retail Media ad revenue isn’t just defined by the budget available from your advertisers. It’s also the rate at which it can be spent. That’s where Data Science and Automation can help.
How data science is powering the new wave of Retail Media platforms
Retail Media ad revenue isn’t just defined by the budget available from your advertisers. It’s also the rate at which it can be spent. That’s where Data Science and Automation can help.
5 min read
Amazon Advertising has built one of the most profitable businesses in the world. Its profitability is due, in part, to a highly intelligent technology platform. Amazon has successfully used AI algorithms to combine automation and campaign optimization to deliver high conversion rates.
Some retailers and marketplaces have tried to emulate that success with off-the-shelf, white-label Retail Media platforms. But these technologies do not feature automation. Profitability is lost because of the need for high-cost human expertise in campaign management and optimization.
Fortunately for other retailers and marketplaces, cutting-edge AI is not limited to Amazon. With platforms such as relevanC, powered by cutting-edge Machine Learning algorithms, access to the $100BN Retail Media opportunity is being democratized. Let’s take a look behind the scenes to understand how Data Science and Machine Learning are used to tackle the Retail Media challenge.
The Retail Media profit equation
Retail Media ad revenue isn’t just defined by the budget available from your advertisers. It’s also the rate at which it can be spent. If an ad is unable to attract the necessary clicks then the retailer is wasting its advertising real estate and leaving profit on the table. Ultimately what is unspent here will be spent advertising on a competitor’s website.
A Retail Media platform must be able to place ads on the retailer’s website in a way that maximizes Spend Rate. That’s where Data Science can help.
Increasing Spend Rate
From a data science perspective, increasing Spend Rate means tackling problems, the answers to which are not necessarily compatible with each other. Mathematicians call this “multi-objective optimization”.
To achieve our objective of achieving the highest possible Spend Rate we need an algorithm that can maximize both Conversion Rate and Fill Rate while keeping computation time as low as possible – under 50 milliseconds. More about computation time later.
Let’s take an example in which we are serving ads for Apple iPhones. Tying the ad to the keyword “smartphone” would result in a high Conversion Rate for searches containing that keyword. But some of the potential sales to shoppers using the keyword “iPhone” would be missed. On the other hand, leaving the interpretation too loose could result in iPhone ads appearing in search results for apple juice!
An ad that balances Conversion Rate with Fill Rate is “relevant”. Relevant to the shopper as it corresponds to their search, and relevant for the retailer because it provides the right financial returns too.
Irrelevant ads annoy customers and diminish loyalty. This drives down organic revenue and reduces Spend Rate by consuming valuable web real estate with ads that don’t convert. In layman’s terms, a relevant sponsored product ad is one that performs better financially than the organic result it is replacing. And in slightly less layman’s terms:
SalesProfitsSponsored + AdRevenuesSponsored > SalesProfitsOrganic
Defining relevancy for a sponsored product ad that has already been served is easy. The Machine Learning challenge, however, is to do the same in real-time by computing the likelihood of clicks and the likelihood of purchase at the moment a shopper presses the search button. And that problem looks something like the following:
LPSponsored * MarginSponsored + LCSponsored * CPCSponsored > LPOrganic * MarginOrganic
Use Machine Learning to predict relevancy
Multi-unknown parameter equations, like the one above can only be solved efficiently using Machine Learning algorithms. Training them requires huge volumes of first-party transactional and behavioral data that exist at scale on e-commerce websites and marketplaces.
Shopper expectations demand that search results – paid or organic – appear almost instantaneously. So a Retail Media platform must be robust enough to ingest organic data in real-time and score product relevance without impacting the shopper experience. relevanC’s Sponsored Product platform does this for the Cdiscount marketplace, handling over 10 billion page views per year.
Natural Language Processing levels the playing field
If calculating relevance in real time using historical data isn’t enough of a challenge, in marketplaces retail media algorithms also need to support fair competition for the top display positions. A marketplace must help all its sellers scale their businesses fast to keep them engaged and of course, to encourage more advertising spend. To do this, advanced Retail Media platforms will integrate Natural Language Processing algorithms that use product catalog information to attribute relevance to look-a-like products where sales history data is limited or non-existent.
The ultimate Goal is Automation
A marketplace seller cares only about growing a business and getting a high return on ad spend (ROAS) with a minimum of effort. A Retail Media platform, when done right, like all well-made software, hides its complexity and power behind an intuitive user experience.
By leveraging the techniques discussed above, relevanC has created an advanced advertiser interface that automates campaign creation. The integrated AI-powered campaign tool automatically chooses the best combination of products, keywords and bidding strategies in real-time, consistently outperforming an experienced advertising professional.