The combination of data-driven machine learning and in-house merchandising expertise to drive product recommendations and personalize the customer experience across the various customer touch points has major benefits to your bottom line.
Recommendations are everywhere, from eCommerce sites to Netflix to Spotify to even news sites. They seek to predict the "rating" or "preference" a user would give to an item and help the users discover other products they would like to buy. You can think of them as product pitches to engage the user by offering a personalized customer experience.
Amazon estimates that about 35% of its sales come from recommendations. Netflix estimated that their recommendations influence about 80% of hours streamed.
When product recommendations inspire even minimal customer engagement, they lead to a 70% increase in purchase rates, both in that initial session and in return sessions.
Customers who add recommended products to their carts and/or purchase in the initial session tend to purchase significantly more items and have 33% higher average order values.
According to Kibo, “The brands that offer ‘personalized commerce’ not only see higher ROI, they are in the best position to quickly pivot their business to meet changing customer behaviors, and deliver personalized messaging to accompany it.”
This ability to pivot was vital in 2020 when the COVID-19 pandemic changed customer preferences drastically. In the scramble, brands have inadvertently offered out-of-stock products, dramatically increased delivery times, sent out tone-deaf messages, and run unfortunate promotions that point to untimely or sold-out items. Companies that could adapt quickly to the shopper’s COVID-19 behavior thrived while others were forced to layoff and cut costs.
What Kind of Recommendations Do Well?
Use the power of both data-driven machine learning and in-house merchandising expertise to drive product recommendations and personalize the customer experience across the various customer touch points.
Let’s explore some types of recommendations, ranging from the easiest to set up to the toughest.
Simple recommendations like bestsellers, most popular
Curated merchandiser recommendations
AI-powered product recommendations
It’s important to think of AI as Augmented Intelligence, not Artificial Intelligence. Gartner defines Augmented Intelligence as “a design pattern for a human-centered partnership model of people and artificial intelligence (AI) working together to enhance cognitive performance, including learning, decision making, and new experiences.”
1) Simple Recommendations Perform Well
The most straightforward recommendations use the "wisdom of the crowds" or “social proof.”
They have titles like “bestseller,” “trending,” and “most-viewed,” but the taglines change based on the creativity of the design team. These recommendations use basic analytics data like purchases, views, and clicks to rank the products in the catalog. The taglines change based on the creativity of the design team.
These have provided the highest uplift across important metrics such as CTRs and conversions in repeated studies. If your catalog and purchase data is small, this type of recommendation may be all you need. However, if you have a larger store, then you will want to leverage AI-powered product recommendations.
2) Merchandiser Curation Works Well by Itself in Certain Cases
Merchandiser curation works well in cases when there is not enough historical data:
New products launched
Fashion products where you can use designer insights to create “complete the look”
Campaign-driven pages for a specific campaign
3) Using Machine Learning for Product Recommendations
Machine learning can uncover unknown relationships and deliver a unique, personalized experience for each user. It uses inputs such as:
The current interactions (Views, Clicks, Add To Cart, Wishlist, Like, Share, Purchase, and others)
Interactions from other customers with similar tastes
Content (Product info, Media/Text)
They can also consider other sources of information such as local time of day, local weather, device, location. You can augment this with data further with information from other channels like physical stores - purchases of products, products bought together, store traffic patterns etc. Machine learning allows you to scale up the personalized experience for the users as users and catalog grow.
An extreme example of this is the Discover Weekly playlist from Spotify. Spotify delivers a playlist called Discover Weekly to over 200 million Spotify users every Monday. The playlist delivers a unique set of 30 songs they've never heard on Spotify before. This allows Spotify to delight the listener every week, but it also reduces Spotify’s licensing costs by highlighting non-popular songs (which are typically cheaper). Retailers can leverage a similar approach to reduce long-tail inventory and non-seasonal inventory.
Hybrid is Best
Of course, the best approach of all is a hybrid one that blends recommendations from each of these three types based on the customer context.
A Case Study
One of our clients had plenty of insightful customer data. They wanted to use that to deliver personalization for even anonymous users in a way that increased conversation rates.
Our team of experts created an analytics engine that processed both past data and real-time data. It predicted purchases for both anonymous and new customers. It then identified top products and the categories from which customers were likely to purchase.
Set up analytics to track against your business KPIs. Some popular ones are revenue, profitability, average order value (AOV), revenue per user, CTR and eCommerce conversion rate
A/B test multiple experiences to see what works for each type of customer
Create self-learning algorithms that align the recommendations with your business objectives such as profitability, revenue, engagement, costs, etc.
Recommendations Across the Marketing Funnel
Place different types of recommendations as the customer moves along the marketing funnel from awareness to purchase.
For example, bestsellers are the most successful on the home page and category pages, while the product page should have related or complementary products.
Here are some specific suggestions for each type of page:
Home page - Broad recommendations such as most popular, bestsellers
Category page - Category bestsellers
Brand page - Brand bestsellers
Product page - Complementary and related products.This is a good place to showcase any curated looks.
Cart page - Complementary products based on the cart. You can also show their wishlist items here. (Note: Allow ‘Add To Cart’ without navigating away from the page. A/B test this to validate that the recommendations do not interrupt the purchase flow.)
Out of stock page - Replacement items. Visual recommendations can be used here to show similar products.
Order Confirmation - Show complementary products, wishlist items
Recently viewed items - This can follow the user across the site
Leverage Recommendations Across Different Channels
Leverage recommendations across all your channels. Email recommendations are one of the best performing product recommendations channels. The product recommendations can be leveraged across different types of emails - welcome, abandoned cart, order confirmation and more.
Narendra Ramachandra is a Digital Product Manager with a diverse experience with eCommerce, financial services, insurance, and healthcare engagements. He has successfully led several of Object Edge’s product teams while building digital products for external clients and internal initiatives, including an AI-based recommendation engine.