# Meesho's PRISM Drives Over 75% of Orders > Source: > Published: 2026-06-04 11:53:25.467687+00:00 # Meesho's PRISM Drives Over 75% of Orders BusinessLine reports that Meesho's proprietary AI discovery engine **PRISM** now powers more than **75%** of orders, according to recent coverage of the company's disclosures. A press release filed with the National Stock Exchange on May 6, 2026, shows Meesho reported NMV of **₹11,371 Cr** in Q4 FY26, up about **43% YoY** (NSE filing). Coverage in Financial Express's CIO section and related interviews attribute multiple FY26 improvements to AI systems: conversion gains of **~15%**, an AI shopping agent that lifted conversion **22%**, a reduction in return-to-origin (RTO) by over **10%**, and operational savings such as a **23%** cut in customer-support costs (Financial Express/CIO). Financial Express also cites a shareholder letter quote from CEO Vidit Aatrey: "This year we have made a deliberate bet on AI as the operating system for how we build." Editorial analysis: These reported figures place Meesho among the more advanced large-scale operational AI deployments in commerce, with clear implications for recommender-system scale, inference engineering, and ML-driven unit economics. ### What happened BusinessLine reports that Meesho's proprietary AI discovery engine **PRISM** - described in public coverage as the "Personalised Ranking & Intent Signal Module" - now powers more than **75%** of orders (BusinessLine). A press release filed with the National Stock Exchange on May 6, 2026, records Meesho's Q4 FY26 results, including **NMV of ₹11,371 Cr**, up **~43% YoY** (NSE filing). Coverage in Financial Express's CIO section summarizes metrics attributed to FY26 AI investments: conversion rates improved by **~15%**; an AI shopping agent lifted conversion by **22%**; return-to-origin (RTO) dropped by over **10%**; the AI integrity layer blocked about **9 million** high-risk transactions and restricted **2 million** consumers and **62,000** sellers; and AI-driven voice/chat agents reduced customer-support costs by **23%** (Financial Express/CIO). The CIO piece also reproduces a shareholder-letter quote from CEO Vidit Aatrey: "This year we have made a deliberate bet on AI as the operating system for how we build." (Financial Express/CIO). ### Technical details Editorial analysis - technical context: Public reporting names components of Meesho's AI ecosystem, including **BharatMLStack** (the in-house infra platform) and tools such as **PRISM**, Geo-India LLM, NIS, Chorus, and TruthMesh (Financial Express/CIO). Industry reporting frames this as a layered stack where shared infrastructure supports horizontal intelligence capabilities and multiple user-facing systems. For practitioners, the key engineering challenge implied by a >75% order share is sustaining high-throughput, low-latency inference across recommendation, ranking, and integrity models while maintaining data freshness and feedback loops at population scale. ### Context and significance Editorial analysis: Companies that report AI powering a majority of transactions are rare at consumer scale. The reported metrics combine improvements across discovery, logistics, integrity, and support, which together affect both top-line conversion and unit economics. From an ML-ops perspective, this suggests investments in productionization: continuous training pipelines, feature stores, online experimentation, and inference-cost optimization. The Financial Express coverage also highlights operational outcomes such as RTO reductions and fraud blocking, indicating that Meesho's AI usage spans both growth and risk-control functions (Financial Express/CIO). ### What to watch Editorial analysis: Observers should track several indicators to validate and contextualize the reported gains: - •How Meesho discloses attribution methodology for the ">75% orders" figure, including A/B test designs or attribution windows - •changes in unit economics reported in subsequent earnings disclosures, which will reveal whether AI-driven conversion gains persist as scale increases - •public details on inference cost and latency, especially for geo/intent models used in last-mile logistics - •measures of model robustness and integrity, given the claim of millions of blocked transactions (Financial Express/CIO; NSE filing) ### Implications for practitioners Editorial analysis: For ML engineers and data scientists, the story highlights mature patterns for operational AI: building a shared infra layer (feature and model infra), reusing horizontal capabilities, and integrating models into product-facing flows such as ranking, routing, and customer service. It also underscores the importance of measurement, clear experiment design and monitoring, when reporting multi-function impact metrics (conversion lift, RTO, fraud prevention). ### Limitations of the public record What happened paragraphs above rely on press filings and media reports. The exact experimental methodology behind the attribution claims and the cost/base-rate assumptions for percentage lifts are not public in the cited coverage. Financial Express and BusinessLine report the figures; the NSE filing provides the financial context but does not publish engineering logs or A/B test data (NSE filing; Financial Express/CIO; BusinessLine). ### Bottom line Editorial analysis: Reported figures portray Meesho as operating a large, cross-functional AI stack that the company and multiple outlets link to measurable business outcomes. For practitioners, the headline is less the 75% number itself than the operational scope implied: ranking and intent models, geo and routing intelligence, integrity systems, and agentic customer support all contributing measurable lifts at scale. ## Scoring Rationale Meesho reporting that an internal AI stack drives the majority of orders is notable for practitioners focused on recommender systems and production ML. The story provides concrete operational metrics and signals mature AI integration, but details on experimental methodology and cost trade-offs remain limited. Practice with real Retail & eCommerce data 90 SQL & Python problems · 15 industry datasets 250 free problems · No credit card [See all Retail & eCommerce problems](/problems/datasets/retail)