One of the biggest advantages of online presence is that it is easier to get customer to speak up and offer feedback. Online Review Statistics suggest that 94% of consumers refuse to patronize a business if they come across a negative review. And over 97% of consumers report that customer reviews influence their purchasing decision. As a consequence it puts tremendous pressure on the companies to assess what customers are thinking and what ratings they are leaving behind. Other than the Scores provided by customers, there is a mine of information available in the ‘Comments’ box of Feedback section. But when, there are thousands of visitors who leave behind Comments on an e-commerce website, it becomes almost impossible to manually read and decipher the nature and sentiment of feedback. That is where ML-based sentiment analysis can help.
It is possible to train ML based models to identify the ‘topic’ on which customers have commented and additionally assess sentiment of every comment. To assess the topic, NLP technique can be used. For example, in case of sale of shoes, customers would generally comment about a few fixed set of topics like shoe’s fitting, style, comfort, price, color, brand authenticity, etc. A trained NLP model can peruse through the thousands of comments in the feedback section and highlight which topics customer have generally commented about. Further sentiment of feedback comments can be assed using ML modeling. Historical feedback review comments can be first categorized as positive, negative and neutral. The ML model can then be trained on such historical (training) data and then tested using fresh comments for the model’s accuracy in assessing a sentiment of the comments. Once the model offers good accuracy, ML-based model can offer great efficiency by perusing through thousands of comments, assessing their sentiments and providing a summary of percent positive, negative and neutral comments.
ML and NLP can thus offer great insights on the topic (subject) on which customers are commenting and related sentiment. This can enable e-commerce players to undertake improvements on product, distribution or servicing areas as needed.

E-commerce players, especially the aggregators who offer multiple products by onboarding thousands of small and big suppliers, have a well-defined process for registering new suppliers. The seller onboarding process involves collection of documents, validating those and making entries in their IT applications, setting credit limits, listing products the vendor will sell and generating new vendor code. This requires huge manual effort and is a cumbersome process.
Intelligent automation which includes RPA, OCR and NLP technologies can be used to collate information coming from new sellers through various sources like scanned images, PDF, MS Word, MS Excel and validate those and enter into the IT applications. This can speed up the onboarding process and enable quick follow-up with sellers for missing information. The extent of data quality check can improve with pre-defined business rules in RPA and OCR. Further, with intelligent automation the overall cost of the onboarding process can be reduced drastically and scaling up business can be easily achieved without the need to build larger teams.

E-Commerce companies receive the trail of customers’ experience in their websites. This includes number of times customers login, time spent on websites, types of searches made, preferences for brands, price elasticity, average purchase ticket size and much more. This enables E-commerce companies to offer a ‘Suggested Section’ or a ‘Recommended for you’ section. While most e-commerce companies have such a feature on their website, the accuracy of such recommendation varies. Using Machine learning based models that leverage historical data can greatly enhance the accuracy or ‘hit rate’ of recommendations made by the websites.
ML models can be trained to recognize patterns of search, surfing and purchase behavior of the millions of online buyers using unsupervised learning techniques. Once, the model recognizes dozens or hundreds of such user patterns, the websites can monitor behavior of buyers online and match them with past patterns to identify what could be the best product, brand, pricing that the buyer would prefer. This can greatly assist the buyers in easily coming across what they had like to purchase and e-commerce players can improve their sales prospects. ML-model based Product recommendations can enable business growth by 10-25% in a very cost effective manner.