Give Your Business a Boost with a Recommender System
Generate New Business with Accurate and Timely Product Recommendations
This morning, I woke up with an urge to have my daily dose of cereal. But all I came across was an empty box, which reminded me I had to stock up! I opened my online grocery app and while trying to find my favorite cereal, a pop-up grabbed my attention to something that intrigued me. The pop-up prompted me to visit some offers, recommended just for me. As a result, I ended up not only buying my favorite cereal, but loads of other things too. The app was able to bring my attention precisely to the items I needed. It seemed like the app was better at gauging my needs than I was!
Though I was late for my meeting because of my extended shopping span, I made up my mind to check the app’s recommendations section the next time too. I wanted to load up on household needs, which is a weekly affair. Basically, the online grocery app knew what I needed and presented it to me, converting my needs into more business for them (and probably creating some stickiness for me).
Given the limited amount of time we all have, being able to get our attention at the right moment and converting it into a transaction — that is what business is all about! If you can tell me what I need to buy, even before I know I need it, you have your sale and I am a happy customer, and likely to return.
This is where recommenders come into play.
Recommender systems have been in the world for a long time and the credit goes to companies like Netflix and Amazon. They gave us the “people who bought this also bought” view at the bottom of their web pages. I know for certain that a lot of my purchases have been from this section.
Collaborative filtering is one of the most popular methods of building a recommender. There are two types of collaborative filtering, namely:
- Item based filtering: Identifies similar items based on your purchasing behavior and then recommends those items to you
- User-based collaborative filtering: Identifies other users with buying behaviors similar to yours and then recommends items the other users have purchased
Building a recommender system is easier said than done and requires a bit of statistical knowledge and model-building skills; but the most difficult part is the scaling. Building it for 10,000 customers and then, building it for 10 million customers are two completely different things. While your model could easily churn out accurate results for the 10,000 set, the moment you reach a million users, the model’s accuracy and its run time begin to show variations.
At Mindtree, we have used both the filtering approaches and have seen the challenges with each. As a result, we worked on a proprietary approach - a graph-based recommender system. This system is not just able to reduce the common recommender issues like serendipity (new items to be recommended, avoiding the previous recommendations) and recency (no new recommendations to be added because temporal dynamics can’t be added), but it is also able to scale as needed to accommodate an increase in customer base, even to 10’s of millions.
In our test environment, the performance parameters are 10 times faster than a relative collaborative filtering model running on the same system. The best part is that you don’t have to build two models for items and users. You can use the same model to derive both item-based and user-based recommendations.
With Mindtree’s proprietary approach to derive recommendations, cross-selling and up-selling become easier and provide you with more selling opportunities and value for your customers.
While we can create the retailer’s online recommender system (we built one for the largest online grocer in India), the same approach can be extended to many other areas like visual products, ads, content and more. The result will be a higher probability of generating new business.
Reach out to firstname.lastname@example.org to understand how we built a top-notch recommender system for one of the largest online grocers in India, and how the system is helping them stay ahead of the curve.