CrossSell Recommendation Engine for Ecommerce Site

Project Description

About Us

We are a company that produces marketing materials through mass customization and web-to-print systems. We operate globally with 6 main markets and offer services to help small businesses create an identity for their company.

 

Problem Statement

Today, we have the capability to recommend personalized products through specific merchandising placements on our ecommerce site. We are working to update our strategy and models across our global ecommerce platform. The scope of this project is for two different locations on our ecommerce site, with differing strategies, that we want to build models to solve for.

 

From a business perspective, we want to build revenue-driving CrossSell recommendation models that:

•       Provides our customers with relevant products that satisfy, surprise and delight - at the right time

•       Brings awareness of product offering and inspiration with our depth of designs

•       Is easily, accurately and flexibly managed

 

From an analytical perspective, we want these models to:

•       Ease of Optimization: Current thinking is that the models should be self-learning. That being said, we are open to differing opinions and alternative ways to achieve similar performance results. The main goal is to reduce manual model rebuild and calibration.

•       Scalability: The system should be able to accept new products and start optimizing with current recommendations. Again, we are looking for advice and recommendations on how best to achieve this.

•       Integrate with existing Hadoop-based Decision Engine platform for developing and deploying models

 

Success metrics could include:

•       Take Rate = Items Ordered/Items Offered

•       Engagement Rate = 1+ Items Ordered/Visits to CrossSell page

•       Bookings, Gross Margin

•       Improvements in Net Promoter Scores (long-term)

•       Incrementality; increase in orders or increase in items per order

 

Customer Experience

When a customer visits www.vistaprint.com and selects a product to personalize they go through a flow that has them customize their design (see Appendix A), select options for substrate and quantity and then they are shown a ‘Matching’ page that is our CrossSell experience in the customer journey that showcases their selected design on multiple other products (see Appendix B). This is one of the merchandising locations that we need to build a new recommendation engine for.

 

If you are a customer with an account, or have purchased before, you will have a username and password. Upon entering your credentials you are typically directed to the ‘Returning Customer Homepage’. The top half is dedicated to links for account maintenance, order history and saved or in-progress projects. The bottom half of the page is where we showcase our ‘Recommendations for You’ section that renders your previous designs on additional products (see Appendix C). The desire here is to showcase new or new to the customer products to increase awareness and encourage engagement and purchase. This is the location of the second model that we need to build.

 

To render this type of customer experience there are multiple models and services called. To simplify it;

• There is a model that tells the site what product to show (this is what we are focusing on rebuilding)
• There is a matching (imagery) engine that determines the best ‘match’ to the customers selected design and then renders it on the product. This is what the customer sees (see Appendix B)
• Then a pricing service determines the price to display, taking in to considering the customers referring domain and current promotional activity on the site.

 

Data Available

·         Transactional data: Which includes everything related to orders that customer have placed in the past. We can aggregate this data to any level necessary for modeling.

·         Basket data: Many times our customers add items to their carts that do not end up in an order. The basket data contains product category of last 30 days and current item in carts. The data includes only binary data for whether a product is / was in the cart of not.

·         Site browsing data: Similarly to transactional data, we have all the information grouped by product category and we have aggregations at the following levels: last session (whenever it was), last 7 days and between 7 and 30 days. For each one, we have number of navigations to that product and click information to specific products. We also have access to Tealium data and can build additional variables as needed.

·         Customer segmentation: We have 5 different customer segmentations with 5 levels at most in each. These customers’ segmentations describe visual patterns on what customers are (Consumer vs Business).

·         Product data: We have limited product attribute data currently available but know that there is an opportunity to create data and have plans to explore this as a way to recommend products new to our assortment.

·         Response data: We have access to historical data pertaining to the current CrossSell placements and customer engagement.

 

Expertise Requirements

We are looking to resource this project with someone that can serve as an end-to-end advisor for this project. As a consultant, your job would be to consult on the statistical/modeling approach, architecture of the system and execution. This job is for someone that can develop a good partnership as we will not only implement the system, but we need to understand the pros / cons of any approach as well as future challenge as it related to data management, score computation, site experience impact etc.

 

Knowledge of the latest machine learning algorithms for recommendation engines is desired but not a requirement. As stated above, we think this is the direction we should head but would like to explore all options.

 

Other Contract Details

Timeline: Internal teams at Vistaprint began work in May 2016 to collect, process and prepare the data for modeling exercises. This is expected to be ready for use by mid-June 2016.

Location preference: Majority of the team is located in Waltham, MA with one Decision Scientist in London. We’d prefer that we partner with someone that is able to co-locate with us in our Waltham office.

 

Questions that we have

·         What methodology for modeling would you recommend for our use cases? We’d like to understand the computational complexity and scaling pros/cons.

·         How would you allow new products to start being recommended?

·         What technology or platforms do you have experience with?

·         How do you make sure the recommendation is generating incremental responses and not only targeting people with high likelihood of buying?

·         What could we do to set ourselves up for success in scaling? From a couple thousand SKUs to hundreds of thousands of SKUs?

·         What have we not asked that we should be thinking about?