Ensuring High Quality Retailers and Brands for Online Businesses in the Age of AI

Author: Ivaylo Bahtchevanov

Ecommerce has revolutionized the way we shop and interact with brands today. Within a few clicks, consumers can access a vast array of highly personalized products and services. While convenience is an integral part of the customer experience, ensuring the quality of the brands and retailers on your platform is a major competitive advantage. A good retailer will bring to your platform trust and credibility, a reliable product experience, enhanced customer experience, and ultimately long-term platform growth.

By partnering with high-quality brands and retailers, an e-commerce platform can guarantee the authenticity and reliability of the products offered, building trust and credibility that attracts a loyal customer base. Positive customer experiences translate to increased sales per user, high retention, and a very engaged user base. As the platform’s reputation grows, it becomes an attractive destination for new brands and retailers seeking to reach a wider audience. This virtuous cycle of growth contributes to the sustained success and expansion of the e-commerce platform.

Graph Learning – Capturing the Long-Tail

While it’s easy to identify the quality of well-known brands with established reputations, there are only so many of them and the long-tail of newer, less well-known brands will ultimately unlock the greatest revenue potential. The challenge is to identify which of these newer brands will become high quality partners for your platform. Since you have limited historical data and context, traditional machine learning struggles to make robust predictions about future retailer performance until it’s too late.

Graph learning, and more specifically, deep graph neural networks (GNNs), presents a new approach to machine learning that effectively solves this issue. Your tables that store information about your existing retailers and users can best be represented as a graph of interactions and relationships. Graph learning brings the concepts of representational learning directly to relational data and allows these interactions to be modeled in their natural state. 

When a new retailer is looking to onboard your platform, GNNs leverage signals from the connectivity of the broader graph to make predictions on a new entity. Contextual aspects of the retailer might resemble similar characteristics of existing partners and can learn from their past behaviors to identify behaviors for the new entity, overcoming the cold start problem that merchants and retailers face. While traditional learning requires a fixed input (which is impossible when you have asymmetric information across different retailers), graph learning maintains the input in its relational form and learns the relevant information without imposing assumptions and constraints on the data.

You can read more about GNN efficacy for relational data in B2C businesses here. That said, building a solution or platform that can perform graph learning in your enterprise is very difficult. We explain in detail what it takes to build a production system for GNNs in our graph webinar, and more about scaling GNNs to meet enterprise commercial needs in our scale webinar.

Kumo – Bringing Graph ML to Ecommerce Data

Kumo brings state-of-the-art graph learning directly to where your data lives and automates the end-to-end machine learning process, enabling you to run state-of-the-art GNNs directly on top of your raw data. With a unique abstraction called the Predictive Query, Kumo makes it easy to define the unique ML problem specific to your business in a SQL-like no-code language which will run and deploy the end-to-end model. By running ML models quickly with no additional overhead, you can run many experiments to identify the best brands that will move your business forward.

For example, you can determine the expected GMV a new retailer can bring to your platform by predicting their lifetime value over a relevant period of time (for example, predict the total purchases made over the next month or year) once they’re onboarded. Read our blog on LTV prediction to learn more about how this works with Kumo. Similarly, when you provide credit to vendors on your platform, you can predict the expected payback period, along with the probability of default. This ensures you’re taking into account the correct financial allocations and allows you to focus on the right set of retailers that will enable you to meet your goals.

Once they’re onboarded, you can use Kumo to empower the retailers in different ways. For example, you can ensure they’re able to engage with the right users by making intelligent recommendations (see our blog on how personalization with Kumo) connecting users to relevant products and brands. You can increase the transaction activity by sending users relevant notifications at the right time. 

You can read more about different ways of using Kumo in our ecommerce blog.