Sigmoidal Success Story
Increase in CRR
Improved Net Promoter Score
What was the business problem?
With the multitude of eCommerce businesses nowadays, each company is trying to stand out. The more products you offer, the more extensive the customers’ choices, and a bigger chance they will find something relevant. But searching through thousands of products can be exhausting. For that reason, companies (the most known are e.g. Amazon, Netflix, Spotify) are using recommender systems, which allow customers to find the products relevant to them quicker and easier.
Our client, a fairly new home decor eCommerce business, wanted to create a product recommendations engine to increase the retention rate, achieve better sales, and make the buying process more convenient for its customers.
What savings and profits does the client achieve?
Thanks to the individualized content delivered by Sigmoidal, the client managed to achieve over a 32% increase in CRR in Q1 2020 (compared with Q1 2019). The client regularly gathers data on customer satisfaction. Net Promoter Score has increased by 17% after the implementation of the Sigmoidal solution (as of Q1 2020). Our system allowed visitors to engage with product recommendations on a much larger scale, which in turn generated higher revenue per visitor.
How did we accomplish it?
Most recommender systems take either of two basic approaches: collaborative filtering or content-based filtering. Other methods, such as demographic-based or hybrid approaches, also exist. We chose to mix collaborative and content-based ones.
Collaborative filtering is a recommendation based on a model of prior user behavior. It uses the knowledge of other users to form a recommendation. This is based on the assumption that if user A likes chocolate (or in our example a golden lamp) and user B like chocolate (golden lamp) and popcorn (red curtains), then user A will also like popcorn (red curtains). Essentially, recommendations are based on the collaboration of multiple users, based on which you can filter those who show similar preferences.
Content-based filtering is based on the user’s behavior. This relies on the product’s characteristics. So if a user often buys a specific product (e.g., almond chocolate by company A), we will show him/her the same product by another company (almond chocolate by company B) and also similar products by different companies (hazelnut chocolate).
Therefore, product recommendation runs on predictive analytics, meaning correlating information and making predictions based on it.
We created a recommendation engine that requires a large amount of enterprise data. Sigmoidal utilized, clients big database with multiple useful variables, among the others:
- Site traffic data: customer’s journey through the eCommerce website, including which items they look at
- Transactional info: historical customer data such, as spending habits, past sales, list of items checked out
- Customer profile: demographic data and info about probable interests
But also product specifics, such as:
- Product listings: names, quantity, and a possible target group of a product
- Prices: Pricing information on all products, including past and future sales, and which demographics should see the sale price
- Time-sensitive product data: seasonal product launch dates, etc.
How did we boost the project with ?
Sigmoidal provided Data Privacy and Security measures by scanning clients’ data repositories for the types of data clients marked as sensitive. Our algorithm automatically labeled the important ones based on industry standards and clients’ custom requirements, sorting data into categories and clearly labeling it with a digital signature denoting its classification.
Sigmoidal did not focus on verifying concepts – we verified values delivered. From the beginning of the project, we continuously explored the potential of machine learning in a client’s area, implemented the solution on production, and then iteratively improved it.
Sigmoidal employed its Data bias monitoring technique. We monitored predictions for bias regularly and configured our tools to witness if bias beyond a certain threshold is detected.