Learn how Google Recommendations AI can drive new e-commerce success—using your own first-party data
Why to read:
- Understand how a product recommendation engine is a powerful first-party data strategy to drive better personalization—without cookies or third-party data
- See how Google Recommendations AI provides an easy-to-deploy, cost-effective way to get started with personalized product recommendations
- Learn how established e-commerce organizations can use Recommendations AI to drive new growth
- See how Recommendations AI can support and enhance custom recommendations engine development
With the deprecation of third-party cookies looming, product recommendation engines like Google’s Recommendations AI have come into the spotlight as powerful and proven solutions for delivering the personalized, predictive and assistive experience that consumers expect— fueled entirely by your first-party data.
If you’ve ever added a recommended product to your e-commerce cart—and today, that’s likely most of us—you know that good product recommendations are the perfect kind of marketing: relevant and personalized.
In fact, great recommendations don’t feel like marketing at all. They feel like a natural element of an exceptional customer experience—helping us more easily and quickly satisfy our needs: Of course I’d like some soy milk to go with this coffee. These blue shoes are perfect with the top in my cart. I would like to get a wall mount for this new TV.
From a business perspective, effective product recommendations power e-commerce success in two key ways—often simultaneously:
- Increasing Average Order Value: A customer who already decided to buy something may put one more thing in their basket, thus increasing the Average Order Value.
- Increasing Conversion Rate: A customer who otherwise wouldn’t buy anything may be tempted into buying a recommended product, thus increasing the Conversion Rate.
While the second-order effects might be a little hard to measure, consistently delivering these kinds of highly relevant, helpful recommendations can deepen customer loyalty—and likely create some new brand ambassadors.
Better personalization—without cookies
Some experts said the launch of Recommendations AI was Google’s way of getting ahead of cookie deprecation. But it’s more aptly viewed as Google learning from the good, bad and ugly of the past two decades of e-commerce businesses using third-party data-driven marketing strategies.
Where that departing paradigm had businesses obsessed with gobbling up as much personal information as possible about the customer (age, gender, location, socioeconomic status, etc.) the emerging revolution in Recommendations AI shows that all of that third-party data is far less important than what a user is actually doing on your site right now. Their first few clicks might reveal more about what they’re up to today with more accuracy than lifetime third-party (cookie) data about ads they clicked and viewed in the past.
Engines like Recommendations AI look across tens or hundreds of thousands of user sessions, profiles and transactions—using AI to connect the dots, find the patterns, and anticipate what a given user is looking for based on the successful outcomes of previous users. Because this is inherently anonymized, it just so happens to align perfectly with the shift toward privacy: better personalization and better CX—without cookies.
Versatile recommendations across the customer journey
You don’t have a recommendation engine? It’s high time to get one. You have one? Recommendations AI from Google can still yield substantial benefits to your bottom line. While there are no products which fit all businesses, Recommendations AI is a rather versatile one. It comes in the form of three different types and each type is designed to work best at the certain moment of Customer Journey:
- Recommended for You, which works best at the home page
- Others You May Like, that activates during browsing through the products
- Frequently Bought Together, dedicated for cart expansion at the last step of the conversion funnel
As you can see, these models are complementary to each other and able to cooperate in order to cover the entire Customer Journey on your website.
How organizations use Recommendations AI: From launching e-commerce to advancing a successful e-commerce business
You’re right to be skeptical of any product that promises one-size-fits all value. Recommendations AI, however, was designed for versatility from the ground up. It can enhance CX and drive business value on every e-commerce maturity level—whether this is your first foray into product recommendations, or you already have a strong recommendations engine in place. Here’s how can you benefit from using Recommendations AI:
Kickstart your initial e-commerce expansion with Recommendations AI
If you’ve only recently launched your e-commerce platform, your options for recommendation engines are probably limited. That’s because historical transactions are the much needed fuel that powers recommendation engines—and without these precedents, it’s difficult for any AI model to learn.
But Google’s Recommendations AI comes pre-trained before you even use it, using Google’s vast number of transactions from different e-commerce retailers. So as long as you exceed 10,000 product detail page view events (or add to cart events, depending on the optimization objective) per month, you can deploy Recommendations AI on your website even before gathering critical mass of transactions needed for learning. Moreover, this type of model should continue to improve over time, provided that the scale of your operations also continues to improve. The more transactions from your shop, the more single models can rely on your specific transactions, users, and product catalog and less on the pre-trained knowledge, making it more precise in the long run.
Driving new and sustained growth in established e-commerce businesses
If you happen to have well functioning e-commerce with some type of recommendations solution, first of all, congratulations. (Most e-commerce businesses fail in their infancy; if you managed to stay afloat, you definitely did something right.) Now how do you sustain that sweet growth rate over the long term? Chances are that the recommendations engine that you’re using is not fully utilizing the power of the first-party data you have gathered—and it might be worth testing an existing solution against Recommendations AI to see which one fares better. Maybe you can unlock an additional source of incremental profit?
And it doesn’t end with recommendations. In due time, you will probably need other AI models in order to nurture your best customers (Lifetime Value model) or prevent churn (Churn/Cancellation model). These solutions work best if you have a Data Lake in place. Good news: the implementation of Recommendations AI is a wonderful excuse to build one!
To serve the right recommendations to the right customers at the right time, you will need to prepare at least tables with Transactions, Customers and Products and keep them up to date, since Recommendations AI automatically recalibrates every three months (which is a default setting, subject to change). Those three tables are the key pieces of data for your business and gathering them in one place (in, say, BigQuery) could be considered as an MVP for said Data Lake. Having all of the data in one place shortens time from ideation to implementation for next AI models and enables reliable dashboarding for more business oriented reporting.
Advancing a mature e-commerce platform with custom recommendations development
Recommendations AI can also help enhance and advance the performance of a well-oiled e-commerce machine that’s already successfully helping millions of customers find what they’re looking for.
First of all, implementation of Recommendations AI will be very quick, as you probably have a well working Data Lake already and the process is well documented step-by-step, so that every data professional will surely know what to do.
In addition, both model training and hitting the API to receive online recommendations will be inexpensive vs. the profit generated. Google’s pricing policy rewards businesses by lowering the cost of each API hit after reaching a certain threshold—a feature that only the largest e-commerce businesses can take advantage of.
Perhaps most importantly, Recommendations AI can help mature e-commerce organizations with talented in-house data teams to build a custom recommendations engine. Recommendations AI can serve as a benchmark for this project—a target for the internal data team to beat, if you will. And because development of a custom solution may take some time, Recommendations AI can also serve as a reliable and effective provisional solution. A small team of skillful data scientists and engineers can have a Recommendations AI model up and running in a matter of a couple of weeks. With this small investment, you’ll be able to reap the incremental revenue from effective recommendations very early—potentially even using that revenue to fund your custom reco engine development project as your team works to beat the new Google Recommendations AI baseline.
Of course, there’s always a possibility that Recommendations AI will ultimately win this fight, making you 100% sure it’s the best possible solution. And what if the custom solution wins? Then its implementation will be much faster, since most of the rails and pipes from Recommendations AI could be reused for your custom solution. It’s the definition of a win-win scenario.
Different flavors of the models
Recommendations AI is a versatile solution, but it’s far from being one size fits all. There are many ways in which the model might be customized, to better serve its purpose and adjust for the organization.
The single model:
- can be optimized using different metrics (maximize Conversion Rate or Click Through Rate, whichever drives the most value)
- allows various degrees of product differentiation, thus preventing serving products only from one category if that’s the goal
- can display more high-value products using its proprietary price re-ranking feature, driving higher Average Order Value
- utilizes filters to support exclusion of products—very important in dealing with out of stock or promotion-only items
The CX challenge is urgent: Act now to build advantage—or quickly fall behind
Every e-commerce organization today knows they need to make big investments in enhancing their customer experience in order to compete with experience leaders—Amazon, Netflix, Uber, StitchFix and more—who are redefining expectations around hyper-personalized, hyper-relevant, seamless and assistive experiences. But e-commerce companies are also facing third-party cookie deprecation and the broader shift toward consumer privacy that’s taking away the conventional “fuel” of their customer-centric experiences.
The e-commerce world is increasingly recognizing what those experience leaders have always known: First-party data is not just the only way forward in the cookieless future, but a much richer source of fuel for hyper-personalized, predictive and assistive customer experiences.
Whether you’re still stuck spinning your wheels with where and how to begin a first-party data-driven marketing transformation, or you already have a well-established first-party data strategy in place, Google Recommendations AI provides an easy-to-deploy, cost-effective and highly versatile tool that harnesses first-party data to drive a next-generation customer experience. DELVE has already helped several organizations deploy and optimize Recommendations AI. In most cases, Recommendations AI pays for itself within months—and paves the way for future projects using this rich first-party data.
Ready to get started? Check out Google’s official documentation if you’re heading the DIY route. Or, read our case study to see how DELVE can help you achieve a fast, hassle-free and successful implementation.
Michał Wrąbel
Director of Data Science