Hey guys, Casey Barwell here. As a part of the Data Science team at SmarterHQ, I’ve been working closely on our most recent product launch— Product Recommendations. While this has been a popular product in the Martech space, there can be a lack of clarity surrounding what goes on in the background to power recommendations, and what those capabilities can do for marketers.
In an effort to provide more insight, I’m going to reveal the data science behind product recommendations that delivers the power solution sets to help guide your customers down their purchase path. Let’s dive in!
So what are product recommendations and why should you use them?
Product recommendations are simply product suggestions that help customers discover, compare, and purchase products. As customers engage with products on your website, Recommendations can be used as a strong personalization mechanism to increased conversions, increase purchase frequency, as well as increase average order value by serving up complimentary products beyond a customer’s initial browsing and purchase behavior.
What are the main ways to make product recommendations?
There are two main categories of methods for making product recommendations — using a human-driven curated process; or a machine learning, data-driven approach, which generates relationships between products to other products, and/or people to products.
How does a human-driven curated approach for product recommendations work?
A human-driven process for Recommendations relate products through the curation of business rules — using attributes about the products. For example, SmarterHQ’s Merchandiser allows you to relate products around basic rules surrounding top selling products, sales items, or products with low inventory. Additionally you can layer rules around additional attributes such as product brands and categories. The curation really revolves around configuring how you want to relate a product to another set of products, or a person to a set of products based on rules you build.
How does a machine learning approach to product recommendations work?
The most critical part of any machine learning or predictive modeling approach always starts with the collection of data, in this case behavioral data, which includes things like products viewed, purchased, or carted together. We can further enhance clients’ data collection process by including multi-channel data (for example clients using StoreFront combine in store sales with web behavioral data.) Using these transactional datasets, combined with client specific parameters (using data driven values), we can generate product relationships that score the significance of the relationship between products, as well as the lift (a key measure which tells us how much more likely a product will be purchased given another product that has been viewed, carted, or purchased.) Finally these product-to-product relationships can be scored and personalized for each customer based on their unique behavior with the products they have viewed or purchased.
What are the benefits and challenges of each approach to recommendations?
There are advantages and disadvantages to each style of recommendations – each method does a pretty good job of complementing the other. Machine learning by its very definition is meant to allow a computer system not to be explicitly programmed, but to be driven by the naturally occurring patterns occurring in the data. So, it responds rapidly to all sorts of changes in taste, style, season by leveraging that information directly from the data.
Timeliness is critical and our approach allows for multiple models to be built daily at scale. However, due to the culling of rules through machine learning, the byproduct of machine learning leverages statistical significance that ultimately filters out certain individuals or products which limits the reach of this approach.
The human-curated approach allows for the ability to easily set rules that are flexible and allow for the broadest possible reach. The challenge, of course, comes with the dynamic nature of consumer purchasing trends, and ever-changing inventories. Some mix of the machine learning and human-curated recommendations complement with the shortcomings of one approach and the strengths of the other.
For marketers, the most difficult part of using a machine learning approach to product recommendations is hiring an excellent Data Science Team to put in the long hours – but when you choose SmarterHQ’s behavioral marketing platform, the hard work has already been done for you! And with our self-adapting models, you never have to worry about your consumers receiving the most relevant recommendations that ultimately help them continue to shop with your brand.
What’s so special about SmarterHQ’s approach to machine learning based product recommendations?
The lifeblood and competitive advantage of SmarterHQ’s data science process is all about the data we collect and generate, and our backend Big Data infrastructure that allows us to engineer efficient processes which all our machine learning processes depend. Our initial release is only the beginning, and we are poised to deliver even more value through our behavioral marketing platform!
That being said, why don’t you see it for yourself?