Meesho's AI-Driven approach revolutionizes e-Commerce, empowering small sellers and enhancing user experience

In this interview with Debdoot Mukherjee, Meesho's Chief Data Scientist and Head of AI, we delve into how Meesho's AI strategy has contributed to its rapid growth in the competitive e-commerce market. Their AI-powered solutions have transformed the landscape for sellers, empowering them to navigate challenges and thrive in a dynamic market We'll explore how Meesho's commitment to responsible AI practices ensures a seamless and personalized experience for every user, fostering trust and loyalty within its diverse community of buyers.

DQ: Meesho is a rapidly growing e-commerce platform. How does your AI strategy contribute to the company's ability to stay ahead in a highly competitive market?

Debdoot: Technology is at the heart of everything we do here at Meesho and AI has significantly helped us innovate and creatively solve issues of both our buyers and sellers. Around 50% of our users are either first time internet users or are individuals who are new to the world of e-commerce. As a result, harnessing the power of AI has helped us bridge the gap between an offline and online shopping experience.

For us, AI has undoubtedly shown a major impact on a variety of company operations, favorably impacting user growth, the shopping experience, product discovery, seller growth, and fulfillment. Our team is continuously exploring new ways to leverage Data Science and other cutting-edge technologies to provide the best possible experience for our users.



At Meesho, we have used AI in the following ways:

● Personalisation – We have devotedly worked on creating cutting-edge ML models that have surpassed conventional methodologies. The user experience can now be tailored and personalized using these state-of-the-art algorithms, which detect a user’s shopping intent in real time by tapping into different signals from a session such as clicks and additions to wishlist or cart.

● Cataloging – Taxonomy tagging has made it easier for our new sellers to upload their products, bringing down the overall time taken by ~99.6% - from 2 days to 10 mins. It has also improved the accuracy of the process hence boosting the discovery/search of their products on the platform.

● Fulfillment - A lot of our customers are from tier 2+ areas; as a result, their addresses are not very well defined. We have used AI to help our 3PL partners accurately deliver in the remotest locations using these models.

● Product and Price recommendations: A lot of our sellers on the platform are first-time e-commerce platform users or first-time internet users. By using our AI algorithm, we help our sellers remain competitive by providing them with the right product pricing and strategy tools, which in turn has helped them with better sales on the platform.

● Search optimization: As most of the buyers on the platform are from tier 2+ cities, we have enhanced and enabled voice and text search in vernacular languages for searching products with the help of AI models.

DQ: Could you give us some insights into the key challenges your team faces in implementing AI solutions for a diverse range of sellers and buyers on Meesho? How do you address these challenges?

Debdoot: We work towards democratizing internet commerce for everyone – both buyers and sellers. We understand that the majority of our users belong to tier II and tier III cities or many of them are using the internet for the first time. We recognize the limited experiences these buyers have with the technology such as typing in English or using keywords for doing a simple search on the net. So, to create a definite shopping experience for such users, we use a recommendation-led approach rather than a search-led one. We help our users at every step in their shopping journey with relevant recommendations. We also address challenges like ambiguous search queries or keywords typed in vernacular languages. Our cutting-edge AI models help us decipher the intent of the buyer and provide them with relevant search results or recommendations for shopping.



On the seller side, we cater to small businesses which have limited manpower and constrained budgets. These sellers often lack the expertise to effectively price their products and face cataloging challenges. Our AI powered solutions simplify these processes for them, for instance, a seller can click an image and fill in the essential details to list it on the platform. Furthermore, our team also guides these sellers on how to correctly price their products and marketing strategies which are unique to each seller.

Our AI capabilities also enable us to resolve issues like incomplete addresses and minimize delivery failures.  Additionally, we also use AI to identify and stop fraudulent transactions, providing buyers and sellers with a secure and reliable platform.

In conclusion, our strategy involves utilizing AI-powered solutions to offer customized buying experiences for first-time internet users while empowering small and medium enterprises with streamlined procedures and advice for profitable selling on Meesho.

DQ: As Chief Data Scientist and Head of AI at Meesho, how do you ensure responsible AI practices are integrated into the development and deployment of AI algorithms on the platform?



Debdoot: On responsible AI, I would like to cite examples of work streams at Meesho, which seek to build a fair and equitable e-commerce marketplace that works for everyone. These efforts are central to Meesho’s vision of democratizing internet commerce and how we empower every user, wherever they may be, and enable every seller, irrespective of their size.

Historically, we observed biases favoring large sellers and private brands creep into other marketplace platforms. To a large extent such bias gets injected because ML models used to rank products are prone to give rise to a "rich gets richer" phenomenon. At Meesho, we carefully tune our algorithms so that we can mitigate this effect and establish a level playing field for all our sellers. Irrespective of the volume of listings that a seller has or the categories that they operate in, we want them to succeed on our platform.

On the user side, we are developing an AI-enabled platform that can accommodate users, from even the most remote areas. Our efforts in building effective voice search, understanding ambiguous search queries and addresses, making interfaces more intuitive and personalized are all directed towards this goal of serving the diverse Bharat user base.

DQ: Transparency is a key aspect of responsible AI. How does Meesho ensure transparency in its AI algorithms and decision-making processes to build trust with users and stakeholders?

Debdoot: On the seller side, we focus on offering valuable insights to them for better growth along with strategies which can help them grow on the platform. We recognise that for sellers to be successful, they must be aware of the best practices for placing ads, setting right product prices, and comprehending the variables that affect product visibility and ranking. Hence, we ensure that we provide them with the right guidance, which is needed for their growth.

AI models for product recommendations and ranking decide which products get displayed under what context and for which users. While these models take into cognizance a wide range of signals and user feedback, we make these dimensions clear and available to our sellers so they may concentrate on the components the algorithm prioritizes. Further, we distill actionable insights from user feedback for our sellers to let me know what they can do in order to enhance  visibility and sales for their products. This strategy helps our sellers to make wise choices and optimize their products for more exposure and platform growth. We want to foster a more welcoming and encouraging environment on Meesho where our sellers can succeed by demystifying the algorithm's inner workings.

DQ: Data privacy is a significant concern for users today. How does Meesho prioritize user privacy while leveraging AI technologies to deliver personalized experiences?

Debdoot: Safeguarding user data is our utmost priority. This commitment is demonstrated through various measures we have implemented. We have a dedicated security team focused on ensuring the safe and secure handling of our data. They meticulously scrub and remove personally identifiable information (PII) to protect user privacy. We take steps to anonymize the data, adding an extra layer of security.

DQ: Meesho's AI algorithms generate recommendations for users. How do you ensure that these recommendations are accurate, relevant, and free from any manipulative tactics?

Debdoot: Our AI algorithm provides users with highly personalized feeds by taking into account both their short-term intent based on in-app interactions and their long-term user behavior on the platform, which indicates their product preferences. This strategy enables us to deliver personalized recommendations in real-time while assuring the highest level of accuracy and relevance to each user's particular tastes.

It is important to stress that our AI algorithm is unbiased and free of deceptive practices. We have put in place strict controls to prevent sellers from tricking the algorithm into recommending their goods. When we see such behavior, we take swift action to stop those sellers, guaranteeing a level playing field for everyone and preserving the reliability of our platform.

We are committed towards offering authentic and honest suggestions by continuously avoiding manipulative practices and punishing those who indulge in them, resulting in a fair and reliable marketplace experience for all our users.

DQ: What are your long-term goals and vision for responsible AI at Meesho, and how do you plan to incorporate emerging principles and guidelines in this field to guide the company's AI strategies?

Debdoot: We feel that the ultimate success in our mission will entirely rely on when all small and medium businesses prioritize selling online for growth. To empower such business, we want to radically simplify the selling process on our platform. We are exploring the implementation of a personal assistant for sellers, similar to an experienced agent, who provides valuable insights and guidance on the next beneficial steps. This AI-powered assistant's goal is to increase sellers' productivity and success on our marketplace.

Similar to this, we want to do more for our customers and other user personas than just adopt e-commerce practices evolved for the western world. Our goal is to develop a brand-new, more user-friendly shopping experience focused on voice-first interactions. Our long-term objective is to create a shopping assistant that meets the specific requirements of every customer and helps them make the best decisions possible when shopping.

Our dedication to ensuring a fair playing field for small-time sellers is consistent with emerging guidelines and ethical practices in AI deployment. We are committed to being responsible and thoughtful in our AI implementation, providing fairness and equal opportunities for everyone on the platform.