
Meesho Counts Its AI Gains in Engineering Pods. Its Last 1.4 Points of Margin Came From Delivery.
Meesho's contribution margin reached 6.0% of merchandise value in July–September, and its own ratios put the 1.4-point quarter-on-quarter gain on cheaper order handling, a saving it says it will partly reinvest. The AI figures it published beside the margin measure engineering output and compute, and when an analyst asked for shopper metrics, the CFO gave no conversion figure.
Neritus Vale
Meesho, the Indian value marketplace, lifted its contribution margin chiefly because parcels got cheaper to deliver, and it says it will partly reinvest that saving. The AI figures it published beside the margin count engineering output and compute; none measures whether shoppers buy more. Contribution margin reached 6.0% of net merchandise value (NMV) in July–September, up from 4.6% the quarter before, its shareholder letter shows. The letter defines it as revenue less the costs of handling placed orders.
Meesho’s own press release invites a broader reading. Vidit Aatrey, the chief executive, credited “logistics efficiencies and improved monetisation”, then turned in his next sentence to AI. The margin bullets sat under “Margin Expansion Driven by Structural Efficiencies”, directly above a section titled “AI Across the Platform”, and they credit logistics, the Geo-India address model and stronger advertising monetisation together. Taken at its word, the release shares the credit among delivery, AI and advertising.
The company’s ratios put the quarter-on-quarter gain on the cost side. Marketplace revenue held at 31.9% of NMV across April–June and July–September, by our calculation from the first-quarter and second-quarter letters. Whatever advertising added, revenue per rupee of merchandise did not rise. The 1.4-point gain came from lower order-handling costs.
Dhiresh Bansal, the chief financial officer, put that down largely to delivery. The quarter-on-quarter movement was “largely led by logistics”, he said, in the call transcript published by Investing.com. He named fewer parcels returned to origin, denser orders, a redesigned network and Valmo, Meesho’s logistics arm. Advertising he would not break out; he expects its benefit “as percentage of NMV” from April 2027.
Year on year, the same arithmetic puts the rise on the revenue side. Order-handling costs were 25.9% of NMV in July–September 2025 and again in July–September 2026, by our calculation from the letter and JM Financial’s tables. Revenue rose from 29.2% to 31.9% of NMV over the same twelve months, which is how 3.3% became 6.0%.
Meesho’s fourth-quarter letter blamed consolidation among third-party couriers for logistics disruptions that “hit margins temporarily”. Bansal’s “outer guidance” for this year, he said, is to restore some of the “lost logistics-related spread” back to its level in the year to March 2025. He was describing the logistics spread, not contribution margin, which was 4.9% in that year.
Meesho has warned that the margin may not stay at 6%. “We may not hold Contribution Margin at this level in the near term,” the letter says. Bansal described a lag: savings are banked first and passed “back onto our sellers and hence consumers in the form of lower pricing” later. For the full year he expects contribution margin “substantially better than FY26”, while Inc42 reports Nomura flagging a possible reversal of the gain in the third quarter.
Meesho held back part of the saving this quarter, and the CFO blamed GPUs. Bansal said “we had incurred certain additional costs in relation to servers, especially GPUs, and hence we did not pass it along back”. Server costs rose to 2.3% of NMV from 1.9%, which the letter ties to GPU capacity reserved ahead of the festive sale “for our growing AI workloads”. Advertising and promotion spending rose from 2.1% to 2.9% of NMV, twice the server-cost increase. Once the server cost normalises, which Bansal expects by March 2027, Meesho “will reduce pricing”.
Server costs sit below the contribution margin line, so they limited what Meesho could pass back and did not touch the 6.0%.
Meesho has handed savings back before. Revenue per order has already slipped; Bansal agreed with CLSA’s Aditya Soman that it “reduced a little bit” and said Meesho passes back “a significant proportion” of the logistics benefit. Price cuts, Aatrey told analysts, make “products and categories that were unviable last year or last month” viable. In company data tabulated by JM Financial, the average charge to sellers per placed order fell from ₹56.7 in the year to March 2024 to ₹43.9 in July–September 2025.
The letter’s engineering figures describe how Meesho builds software. Output “measured by our internal productivity score” rose 171% between March and September, the letter says. It does not say how the score is built. In 65% of engineering pods, one engineer now directs the AI agents.
Its other AI figures count compute, not sales: the GPU footprint grew 150%. Agents also reach “how we serve shoppers”, the letter adds, yet all four in its table work on code, data science and security.
A telling figure measures how fast Meesho tests models. The research agent takes a model to a live test in three days instead of 17, a step Inc42 called deployment. What those tests did for sales, the letter does not say.
Asked on the call for a shopper number, the CFO offered a share of orders. Citi’s Vijit Jain wanted AI metrics from the discovery feed, such as “improved conversions, more click-throughs”. Bansal pointed to “the percentage of orders that come from Discovery Feed” and rising time in the app. He also said a better feed helps new users “convert faster”, but gave no conversion figure. On an app built around its feed, that share records where people buy and says nothing of what the model added.
The last conversion figure dates from May. Meesho’s fourth-quarter letter said improvements to PRISM, its recommendation system, “improved conversion by c.15%” over the year. It gave no method or baseline.
The strongest objection is that the delivery saving is itself AI’s work. The letter lists improvements to Geo-India, Meesho’s address-reading language model, among the reasons delivery got cheaper, and credits TrustMesh, a fraud model, with fewer returned orders. Bansal said “using GenAI and LLM for improving accuracy of address” cuts delivery attempts, and so cost. If those models made most of the saving, AI is inside the 6% and this argument fails.
The objection moves the credit and leaves the conclusion largely intact. Bansal put no number on the models, and his middle-mile changes were physical: “bigger trucks in some places”, he said. Grant the address model every rupee anyway, and AI has cut Meesho’s own cost of delivering a parcel. Meesho says it will reinvest “part of the gains”, and Bansal wants to “maintain some flexibility around how much is passed back”; neither names a split by source of saving.
Apparel is the category where a pass-back would matter most, though no source says where the saving will go. Apparel made up 33.8% of placed orders in July–September 2025, a year-old split in company data reproduced by JM Financial. Aatrey ties viability to price cuts, so if the saving reaches sellers as lower prices, clothing is where it would show first. That is our inference, not Meesho’s statement.
The businesses that might move the margin later have no disclosed share of NMV. Meesho Mall’s NMV grew 81%, but Mall is commission-free, the letter says, so the growth adds merchandise without adding commission. Content Commerce NMV grew 160%. Neither has a disclosed share of NMV, a gap Inc42 also flagged.
July–September was a delivery story with AI figures printed beside it. Those figures count engineering output and compute. The advertising margin Meesho expects from April 2027 depends on shoppers converting, and on conversion it has published no figure since May.