Ramgopal Prajapat:

Learnings and Views

AI/ML Use-Cases in Ecommerce

By: Ram on Dec 16, 2022


Ecommerce is at the forefront in leveraging AI/ML Innovations in transforming customer experience and delivering value to their stakeholders. There are multiple business models and associated challenges; hence the use-cases may be slightly different for the different ecommerce players. For example, for closed marketplace like Tata CLiQ fake or counterfeit products are NOT the challenge but for open marketplaces like Meesho or Amazon this is a big area of focus.

On Platform AI Use-Cases

  1. Search: AI and Machine learning algorithms are It involves multiple set of algorithms – natural language processing (NLP) is used for spell corrections, entity recognition (whether searched term is a brand or price range) and indent identification. ML based supervised methods are used for category prediction to reduce search space etc.  Visual search requires different approach - such as embedding of the image first, classification of the image to different categories and then matching with the data to pull the relevant products.
    1. Text search
    2. Visual Search
    3. Audio Search
  2. Relevance & Ranking: Product relevance refers to how well the products or services being offered meet the needs and interests of the target audience but it also important to consider relevance from the business metrics perspective. Whether relevance ranked are resulting to improved total business or conversion rates.  Relevance ranking is used along with recommendations and search for ranking products or items.  
    1. Ranking of the products based on relevance
  3. Recommendations – Improving customer engagement on the platform and Average order value for the customers, recommending right set of products is very important and AI/ML is pivotal for this.
    1. Recommending products using Market Basket Models
    2. Recommendations using Collaborative Filtering, Item Based Filtering, Hybrid Models or Deep Learning Models
  4. Personalization: Personalization in ecommerce involves using AI and Machine Learning for improving shopping experience for customers. The personalization is based on a customer's previous purchases or browsing history. It is used for:
    1. Product personalization
    2. Category Personalization
    3. Bank and offer personalization
    4. Email and Campaign Personalization
  5. Ad Relevance – Some of the ecommerce platform make money by showing ads from their sellers. For example, Amazon and Meesho does this and they leverage ML Ad relevance machine learning models to find right set of ads of each of the customers. These models use data on user demographics, browsing history, and other factors to determine whether the user will click to the ads if shown.
    1. Matching relevant ads
    2. Ranking ads

Off Platform AI Use-cases

  1. Catalog Augmentation - Auto Tagging: From the images uploaded by the sellers, identifying the attributes /tags using computer vision and NLP algorithms to augment the catalogue so that search and other functionalities work effectively.
    1. Identifying information from the images
  2. Supply Chain Analytics for Market Places involves allocating orders to the right stores and estimating delivering time based on various factors including source and destination.
    1. Delivery Time Estimation
    2. Order Allocation
  3. Market Mix Modeling: In the ecommerce industry, market mix modeling can be particularly useful for identifying the most effective marketing strategies and channels for driving sales and maximizing profitability. For example, a business may use market mix modeling to determine which social media platforms are driving the most traffic and sales to their website, or to identify the most effective email marketing strategies for converting leads into customers
    1. Spend Allocation across marketing channels to achieve the scale but maximize ROAS
  4. Fraud Detection: Fraud is common in e-commerce, and it impacts the financial health of a business, customer & seller experience and damage its reputation. Return Fraud, Account takeover and counterfeit product fraud are common type of fraud in ecommerce.
    1. Fake Returns: Some of the customers returns fake returns intentionally that have not been purchased or were not delivered to the customers. This is probably done to take advantage of liberal return policies.
    2. Counterfeit Products: Using computer vision models to identify if wrong product/ replica products of genuine products are used by the seller. This is very common in the fashion and luxury goods industry. The counterfeit products fraud can have serious consequences for marketplace ecommerce- may lose revenue and damage their reputation if consumers discover that there are fake products on the platform.
    3. Repeat Offenders
  5. Forecasting: Machine learning can be used to forecast and predict demand for certain products, allowing ecommerce platforms to better optimize their inventory and search results.
    1. Forecasting Demand at various level and time horizon

Leave a comment