Meesho’s expanding use of artificial intelligence is changing more than product recommendations on an e-commerce platform. The company says its AI-powered systems now influence nearly 75 per cent of orders, handle 62 per cent of customer queries and support a seller network of approximately 960,000 transacting businesses, many of them operating from Tier-II and Tier-III cities. The shift shows how digital commerce is increasingly being organised through automated decisions made at several points: what a customer sees, how a seller prepares inventory, how a query is answered and how a transaction is completed.
The figures were disclosed by Pratik Kumar, Meesho’s head of engineering, in an interaction with Business Standard. Kumar said close to 75 per cent of the platform’s orders come through personalised ranking via Meesho’s Prism system. He described AI as a technology that had helped address problems that were otherwise difficult to solve, while outlining its use across product discovery, demand forecasting, customer support, voice-led shopping, fraud prevention, address prediction, logistics and seller enablement.
That breadth is significant because it places AI inside the operating system of commerce rather than limiting it to a search or recommendation feature. On Meesho, the technology is being used to decide which products are shown to customers, anticipate demand, assist sellers with routine processes and reduce the need for human intervention in support functions. The company’s stated objective is not only to improve platform efficiency but also to make online shopping easier for users who may be less comfortable with typing or text-heavy interfaces.
### From search to prediction
Prism uses signals including demographics, geography, device usage, browsing behaviour, product clicks and ignored listings to personalise product rankings. According to Kumar, the system is designed to identify user preferences even when a shopper arrives without a clearly defined purchase intention. This makes the platform’s interface an active part of the purchasing process: it does not merely respond to a fully formed search but attempts to infer what a user may want to see next.
The use of geography as one of the ranking signals is particularly relevant to a platform that serves customers and sellers across different parts of India. Meesho’s account of regional demand suggests that shopping behaviour is not uniform across the country. Kumar cited Navratri demand as an example, saying that preferences for particular clothing colours can vary by region and day because of local customs. The platform’s AI systems therefore operate within a market where cultural and geographic differences influence inventory and purchase decisions.
Meesho also uses TrendPulse to anticipate demand and help sellers prepare inventory before emerging trends become fully visible. The supplied account does not establish how accurate these forecasts are or how sellers respond when predictions are wrong. It does show, however, that demand forecasting is being positioned as a platform function rather than a task handled only by individual merchants.
For small sellers, this can alter the relationship between local businesses and the wider market. Instead of relying only on their own knowledge of customer demand, sellers may use platform-generated signals to decide what to stock or promote. At the same time, the available information does not establish whether access to these tools is equal across sellers, whether the systems favour particular categories or how much control merchants retain over decisions influenced by automated recommendations.
### The interface is becoming a service layer
Meesho’s Chorus customer support agent handles 62 per cent of user queries, according to the company. Its Vaani AI agent extends automation into voice-led shopping, enabling users to discover products, understand product information, add items to carts and complete transactions through voice interactions.
This is an important shift for digital access because it addresses the friction created by text-based interfaces. The company says it is deploying AI to help customers who may be less comfortable typing search queries or navigating text-heavy applications. The report does not provide user numbers for voice shopping or evidence on whether Vaani has increased completed transactions. It does establish that voice is being treated as a route through which users can complete the full shopping journey, rather than simply ask a preliminary question.
The same service-layer approach is visible on the seller side. AI agents respond to questions about financial journeys, payouts, monetisation and campaign creation. These functions are closely connected to the everyday administration of online selling. Faster answers may reduce dependence on human support, but the supplied evidence does not quantify the effect on seller income, transaction completion or dispute resolution.
The urban significance lies in the geography of participation. Meesho’s annual transacting seller base has grown 87 per cent year on year, and the company says AI is being used to simplify seller onboarding, particularly for businesses in Tier-II and Tier-III cities. The available report does not identify the cities or states contributing to that growth, nor does it provide a breakdown of sellers by location. Even so, the stated focus places platform automation within a broader pattern of commerce extending beyond the largest metropolitan markets.
### The cost of making AI routine
AI adoption at this scale creates a cost challenge for the platform. Every recommendation, query response, prediction or automated interaction requires computing capacity. Meesho says its in-house Bharat-MLStack has reduced AI inference and model-related costs by around 60 to 70 per cent compared with external providers, according to Kumar.
That claim points to an important institutional issue: the expansion of AI in commerce depends not only on whether a model can perform a task, but also on whether it can do so economically across a large user and seller base. Personalised ranking and automated support become more valuable when they can operate at scale without making each interaction prohibitively expensive. However, the report does not provide the baseline costs, the period over which the reduction was measured or the total expenditure involved in developing and operating Bharat-MLStack.
The distinction matters because efficiency claims can describe a reduction in one component of technology spending without showing the overall cost of an AI programme. Meesho’s statement establishes that cost containment is a stated priority and that the company believes its internal technology has delivered savings against external providers. It does not establish the platform’s total AI budget or the financial return from each system.
### What the evidence confirms—and what it does not
The evidence supplied by Meesho points to an integrated model of AI adoption. Prism works on product discovery and ranking. TrendPulse supports demand anticipation. Chorus handles a majority of reported customer queries. Vaani enables voice-led shopping. Other systems are used for fraud prevention, address prediction, logistics and seller support. These are not isolated experiments; they cover customer interaction, merchant operations and internal platform processes.
The evidence also shows that the company is linking AI expansion to inclusion and reach. Voice interaction is intended to reduce the dependence on typing, while automated seller support is intended to make onboarding and routine platform tasks easier. The focus on Tier-II and Tier-III sellers connects these tools to the growth of digital commerce outside India’s largest cities.
Several questions remain unanswered in the supplied material. There is no independent assessment of the accuracy of Prism’s recommendations, TrendPulse’s forecasts or the responses produced by Chorus and Vaani. There is no information on how automated systems handle complaints, incorrect product information, fraud disputes or users who need human assistance. The report also does not describe the safeguards governing the use of demographic, geographic, device and behavioural signals in ranking products.
These gaps do not invalidate Meesho’s reported figures, but they define the limits of what can be concluded from them. The numbers demonstrate the scale at which the company says it is using AI. They do not, by themselves, demonstrate whether the technology has improved affordability, seller earnings, customer satisfaction or access to commerce.
For India’s urban economy, the larger question is how platform infrastructure will shape participation in markets as more decisions become automated. Meesho’s account presents AI as a tool for matching demand, assisting users and reducing operational costs. Its next significance will depend on how these systems affect the people and businesses using them, especially where platform-generated rankings, forecasts and support become central to everyday commercial decisions. For now, the company’s disclosures establish a rapid expansion of AI across the e-commerce chain, while leaving its wider social and economic outcomes to be measured.
