Successful AI & ML Use Cases for Ecommerce - Indian Context

Having worked in the e-commerce sector for a few years, I have been witness to its evolution and innovation in small & large organizations in India.

E-commerce in India had its own set of interesting problems and the kind of innovation that e-commerce product managers brought about was path-breaking.

This has included innovation in: Shipping and Delivery, Checkout & Payments, Search & Display, Sales & Deals - such as: Open Box Inspection, Cash on Delivery, OTP for delivery verification, Wallets with Cashbacks, Store Brand Deals, Regional content etc.

And so much more ...

Recently however, I had the chance to discuss with a colleague on AI specific e-commerce Innovations and how things were changing. We could agree on these Eight at-least:

1. Personalized Recommendations & Search Results : This is perhaps one of the earliest uses of Machine Learning and Data Science and is now in use from the biggest brands ranging from Amazon to Netflix but is still evolving with innovations.

- The factors , or ML 'Features' which are inputs to the recommendations that you get, can include aspects such as Past Purchases, Browsed items, Customer profile info. etc. A great example here is Flipkart where as a user, I've often found the search results and recommendations more accurate than in other sites. Flipkart claims to have experienced a 10% increase in click-through rates and 3% increase in conversion through the use of Machine Learning based recommendation module.

The next generation recommendation engines - not only depend on user driven features but use collaborative inputs, seller related signals and a lot more to tweak their offerings. These models are also becoming self correcting and can internally tweak model parameters in real-time changing recommendations based on resulting performance metrics.

2. Price Optimization & Demand Forecasting : I have personally worked on creating SaaS products which provided 'What-If' analysis with price to revenue forecasts allowing user to optimize pricing in the Life Sciences sector. At that time , ML tools and libraries were not that widely known and available. Today when I look at the same problem from the point of ML models, libraries including the commonly available scikit-learn , I find that forecasting problems can be solved quicker and more accurately. In this cool article, Nicholas shows how Machine Learning reduced forecasting errors from 20-60% as seen in benchmarking competitions.

Despite drop-shipping etc. , forecasting using accurate ML models is very important for the suppliers and e-commerce marketplaces to ensure there are no stock-outs and sellers are able to keep inventory at optimal levels. It also helps in ware-house acquisition planning.

3. Returns Management : According to Instamojo, eCommerce returns in India was between 25-40% in 2022 during festive season. Imagine the savings and profitability impact if this no. could be halved !

Reducing Malicious returns and figuring out ways to reduce product mismatch related returns by looping that into search, recommendations and other modules is another area that AI is getting good at. AI can also flag outlier return data and help customer service teams assess return requests.

Apart from that, AI and ML are used in the routine tasks of Returns Management from Return address labelling to optimally deciding pick-up carriers based on proximity and a host of other factors.

Another use case has been to record and analyze return causes, crunching and classifying them using AI to get the best solution to minimize returns. These solutions could be something as simple as improving product information, better quality images, adding crowdsourced FAQs sections for the product.

4. Content Structuring & Enrichment: This wasn't a very common use-case earlier but is slowly gaining ground. I implemented this as part of the Data Catalog product we were building for a large US ecommerce retailer. One of the problems we needed to solve as Product managers was to ensure that the Product content which generally is in Short and Long Description format is properly structured, enriched and contains all the details. This is achievable through complex classification and parsing model in machine learning where the model is trained how to identify what the content is related to and structure it accordingly. For example - lets say your product description is "red polka dotted snazzy 12 size laced 2023 model sneakers" . Yes, I know those must be very cool sneakers :) . But coming back to AI , there is something called as Content Structuring and Parsing which'd manually take ages, but which AI can do easily with a high level of accuracy converting your Product description to a set of crisp Product attributes which can even become search facets. So, in this case, the attributes could be the following:

Color: Red

Design : Polka dotted

Size: 12

Model Year : 2023

Shoe Type : Sneakers

In this way Machine Learning models can help structure content, flag incomplete or low quality product descriptions and a lot more.

5. Product Discovery & Seller Selection : In large e-commerce marketplaces such as Flipkart and Amazon, there are multiple sellers selling the same product. Have you noticed how certain sellers comes as a default for you and other sellers available a click away, often in an ordered list. This may not be the same for two people and is not just on the basis of price. There may be other factors which the ML model is using to figure out which seller to suggest for you based on the kind of delivery speed, price range, delivery costs, seller reviews, customer category and many other aspects that you're having. In ML, these are called as 'Features' of the model.

An even more compelling use case is the custom ordering of the carousels, product list in the screens that you browse - right from the home page to even a search screen. This is really critical because if you get un-related search results in the so called 'Recommended / Relevant' filtered results - you are sure to get put off. Along with this, users should be given the flexibility to re-order the lists as per their usual criteria such as Bestsellers, New, Price etc. This is another area where A-B testing is required to see the impact of these sorting orders on the browse and purchase behavior.

6. Anti-Fraud & Credit Scoring : Earlier, BNPL or Buy Now Pay Later was rarely seen in the Indian market but from Covid onwards, BNPL has become a must-have feature which the large e-commerce players are using. As a Product Manager, you may need to figure out how to plan for this feature, which player to integrate with, and how to create the customer flow. However in the background , these BNPL providers may be running credit scoring models based on many factors to figure out whether they can offer you credit on that order or not. Assessing the fitment of these models, parameters to expose and to what level is something that you may need to decide about with your technical team.

The use of ML in Anti-fraud has been there for some times and this is not specifically implemented by the e-commerce provided but more so by the Payment Gateway , Wallets etc. For a Product Manager it is important that their Epics / User Stories cover the implementation of Anti-Fraud features which their payment providers are making available.

7. Chatbots & Customer Service : This is an easy one. Chatbots have proliferated everywhere - whether large and small e-commerce sites. E-commerce platforms such as Big Commerce, Shopify etc. provide these add-ons. It seems a simple enough decision then to use Chatbots powered by ML in your e-commerce portal / marketplace, doesn't it ?

True, but the challenge is integrating it in the customer journey, training it well, and doing A/B Testing to understand how it is performing in different scenarios.

A Product manager needs to make sure that the ChatBot is helpful to the customer while switching immediately to human interaction when it appears that the Bot is not solving the query properly. Using a Chatbot as just a cost saving measure in customer servicing is something I've witnessed both as a Manager and a User, and I can say for the latter that it's not a very pleasant experience at times. I have also found that handling edge and negative cases for the Chatbot input and provisioning for how it'd handle that is pretty important.

8. Personalized Communication & Emails: Last but a surprisingly very powerful use case - is using AI backed marketing and e-mail automation platforms to send personalized offers, trigger buying intent at the right time and drive purchase. I came across an article which'd mentioned that Nearbuy (from Ankur Warikoo ) had used to this method to get great click-to-open ratios and boost revenue. Maybe Ankur can confirm on that :)

*************

On a closing note, while I have used the term Machine Learning and AI quite a lot in this article, neither of them would obviously be possible without data or rather Big Data as it is called.

Without easily accessible, governable, understandable data - no product manager or marketer can use Machine Learning or AI.

So have you implemented or come across other scenario where you used ML for success in your e-commerce journey ? Do write in with your experiences !