Most e-commerce platforms are built around a simple assumption: the customer knows what they want. You type "blue kurti size M", the search bar obliges. That model works well for cities where people have grown up on Google. It works less well for the next hundred million Indians coming online — people who browse the way they'd walk through a bazaar, open to being surprised.
Meesho has quietly built its entire product discovery infrastructure around this second kind of shopper. And the numbers now make the bet look prescient.
The Search Bar Was Never the Point
Meesho's recommendation engine, PRISM — short for Personalised Ranking & Intent Signal Module — was developed to support a shopping experience driven by browsing and discovery rather than keyword-based searches. More than 75 per cent of orders on the platform now originate from AI-driven personalised feeds powered by PRISM.
That's not a minor feature update. It means the majority of Meesho's commerce isn't triggered by anything the user consciously typed or asked for — it's surfaced by a system reading their behaviour in real time.
What's Actually Running Under the Hood
PRISM operates through a network of more than 100 AI ranking models trained on over 400 trillion input signals, and executes more than 6 trillion inferences daily within milliseconds. During peak sale events, the system processes nearly 100 million inferences per second.
The infrastructure powering all of this is BharatMLStack, Meesho's in-house machine learning platform built to support high-throughput AI workloads at significantly lower inference costs than conventional cloud infrastructure.
Then there's a component that addresses a distinctly Indian problem — regional demand fragmentation. Trendpulse, an LLM-powered discovery engine within PRISM, identifies emerging demand patterns across regions, cities, and local consumer segments, enabling the platform to surface products that align with evolving regional shopping trends. The system supports more than 10 multilingual experiences, including Hindi, Bengali, Marathi, Tamil, Telugu, Kannada, Malayalam, Gujarati, Punjabi, and Odia.
The Scale That Makes It Work
Recommendation engines are only as good as the data feeding them. Meesho's flywheel here is formidable. The platform counts 263 million monthly active users, approximately 17 billion product views daily, 1.592 billion ratings, and 505 million reviews. It served 264 million annual transacting users and recorded 717 million placed orders in Q4 FY26 alone.
In FY25, Meesho accounted for 29–31% of total e-commerce shipments in India, the highest among e-commerce players in the country.
The strategic logic is clean: more users generate more behavioural signals, which sharpen PRISM's recommendations, which drive more orders, which attract more sellers. Beyond serving buyers, PRISM enables sellers to reach high-intent audiences more effectively by surfacing their products to users who are most likely to engage and purchase.
Discovery as Infrastructure
As Meesho's Chief Data Scientist Debdoot Mukherjee put it, "The next hundred million Indians coming online will not search, they will discover. They will not type, they will speak, browse, and expect technology to meet them where they are."
The implications extend beyond Meesho. If the discovery-first model proves durable at this scale, it challenges a core assumption baked into most e-commerce design — that intent must be declared before it can be served. In Bharat's next chapter of commerce, the platform that can infer intent before a user articulates it may have the most durable edge of all.
Sources
- Meesho says over 75% of orders now come from AI-powered discovery engine PRISM - Storyboard18
- Over 75% of Meesho orders now come from AI-powered recommendations | Company News - Business Standard
- Meesho Says Over 75% of Orders Now Come From AI-Powered Product Discovery
- Meesho IPO Review 2025 – Financials, Strengths & Risk Factors