E-commerce sites and online shopping have been growing steadily over the last two decades. The growth in numbers has led to diverse development in the new online shopping experience, payment pathways, online merchants, and e-commerce sites. This gives rise to new choices for both merchants and consumers. Next best product recommendation can be very useful in doing this.
Although various choices empower both customers and merchants, often too many options are confusing and can negatively affect the online shopping experience. The latter was a problem one of our clients had faced recently.
The client, an online payment company that enabled digital and mobile payments on behalf of consumers and merchants worldwide, had multiple payment products to fit the merchants’ needs – its customer. The merchants who used more than one product had a broader and deeper product usage and thus were more profitable than single product merchants. As a result, all the products were marketed to all the merchants, causing confusion and subsequent unsubscription.
Our client wanted to narrow down the list of available products and offer only relevant merchants’ relevant choices. The head of development wanted to find the next best product (NBP) recommendation for each merchant optimized by adoption and profitability. Our job was to develop a model that could Offer the next products based on historical adoption by similar merchants and incremental profit.
Creating a solution with BADIR framework
All our projects use the 33-pt proprietary BADIR framework. You can read more about the framework here.
The first step of BADIR was to form the Business Question. We consulted with the product team to try and understand their problem in detail. The Product Team wanted to increase engagement with the merchant base by helping the merchant choose the right product while keeping an eye on profits – net of marketing. So we formed our business question to reflect this motive.
The next step was to form an Analysis Plan. The analysis plan is always hypothesis-driven. In the analysis plan, we decide the kind of data we would like to collect, the time window, and the model that will be built. Once the plan was finalized, appropriate historical Data was extracted to train a machine learning model.
Deriving Insights with Machine learning models
This is one of the most exciting steps within the BADIR framework. In this step, we developed an ML classification model to predict a merchant’s probability to adopt a particular payment product.
We built a machine learning model to identify the merchant segments with different adoption patterns. The model was trained on the historical Data and optimized for expected incremental profit.
The machine learning model was, in fact, a decision tree, and for each of the base product, a separate decision tree was made.
A zoom-in into the merchant segment 39 shows optimization of Expected incremental profit calculated from the decision tree.
Our model provided NBP (Next best product) scores for each merchant. The NBP scores took into account the propensity to adopt a product and the expected incremental profit derived from the possible adoption.
Our recommendation was to use this score to intelligently design plans for outbound marketing, inbound customer service call, and web self-service. The scores could quickly provide a quantitative measure of sorting out products and priorities.
Results of Next Best Product Recommendation
The client accepted our recommendations and the results were quite positive! The model recommendation delivered 6X Conversion and 1.5X Volume. The converters do $500+ incremental volume.
Over time, we have observed the success of the BADIR framework in plenty of our projects. The strategic framework can successfully guide data science projects to give positive results for a company.