How Zepto Became India's Most AI-Driven Quick Commerce Company

Zepto is four years old. It operates 1,139 dark stores across 66 Indian cities. It processed 64 crore orders in FY26, averaging 1.75 million orders per day, with Q4 FY26 alone hitting 2.33 million daily orders. Its operating revenue more than doubled to 22,623 crore rupees in FY26 from 11,110 crore rupees in FY25, representing 103.63 percent year-over-year growth. It has 47.97 million annual transacting users. And it is now in active IPO proceedings, having filed its Updated Draft Red Herring Prospectus with SEBI on June 8, 2026, targeting a fresh issue of 8,010 crore rupees.

None of that happened because Zepto was faster than its competitors. It happened because Zepto built a fundamentally different operating architecture, one where artificial intelligence is not a supporting function but the core infrastructure around which every commercial decision, every inventory movement, and every consumer interaction is orchestrated. This case study examines how Zepto constructed that architecture, what it produced in measurable outcomes, and what the strategic model reveals for leaders navigating AI-first transformation in high-velocity consumer markets.

The Market Zepto Entered and the Problem It Was Solving

India's grocery retail market is estimated at approximately 50 trillion rupees, equivalent to roughly 617 billion dollars, making it one of the largest and most fragmented consumer markets in the world. As of 2024, kirana stores, the neighbourhood retail units that have served Indian households for generations, still accounted for 92.2 percent of that market. Online grocery penetration remained below one percent of total grocery spend.

The category economics were brutal. Consumers were accustomed to zero-cost grocery acquisition. Delivery expectations were either relaxed, with scheduled next-day slots from incumbents like BigBasket, or entirely absent. The logistics infrastructure required to guarantee ten-minute delivery at unit economics viable for commercial sustainability did not exist. No Indian logistics operator, technology firm, or retailer had solved the last-mile density problem at scale.

Zepto entered this environment in 2021, founded by Aadit Palicha and Kaivalya Vohra, both Stanford dropouts then aged nineteen. The founding thesis was not that India needed faster grocery delivery. The founding thesis was that speed, at ten minutes or below, would function as a category-defining constraint that forced a completely different operating model, and that the data generated by that operating model would become the most valuable asset the company could build.

That thesis has proven directionally correct. The speed promise was the entry point. The AI architecture was the competitive moat. And the data is now the emerging business within the business.

‍

The Architecture That Made Ten Minutes Possible

Dark Stores as the Physical Foundation

Zepto's operational model is built on a network of dark stores, compact, no-frills micro-fulfilment centres positioned within one and a half to two kilometres of high-density residential areas. Unlike conventional warehouses that optimise for storage capacity, Zepto's dark stores are engineered purely for picking speed and delivery throughput. The layout, SKU selection, and inventory positioning within each store are determined not by human planners but by AI systems running continuous optimisation against local demand signals.

By March 2025, Zepto had crossed 700 dark stores, ahead of its own internal target. By September 2025, the network had expanded to over 900. As of March 31, 2026, per the UDRHP filed with SEBI, Zepto operated exactly 1,139 dark stores across 66 cities, supported by 75 warehouses. The company plans to deploy a further 1,900 new dark stores using a portion of the fresh IPO issue proceeds, allocating 1,629 crore rupees specifically for network expansion.

The density of this network is not incidental. It is the physical expression of an AI-driven location strategy. Zepto uses data heatmaps that analyse population density, traffic patterns, purchasing power distribution, and consumer behaviour clustering to identify optimal store locations before a single square foot of space is leased. The store is placed where the algorithm says demand will be most dense, not where real estate happens to be available.

Demand Forecasting as the Operating Intelligence Layer

The ten-minute promise is only commercially viable if inventory at each dark store is precisely matched to local demand at every hour of every day. Zepto's demand forecasting infrastructure is the system that makes this possible.

The company deploys a multi-model machine learning architecture that processes historical transaction data alongside real-time inputs including weather patterns, local events, time of day, day of the week, festival calendars, and individual user behaviour signals. The models deployed include ARIMA and Facebook's Prophet for identifying seasonal shifts from historical patterns, and Random Forest, Gradient Boosting, and LSTM networks for capturing complex non-linear demand behaviours across sequential data. Power BI dashboards surface outputs to store managers and supply chain planners, translating machine intelligence into actionable inventory decisions at the store level.

The platform currently offers 46,623 SKUs on average across categories ranging from fresh produce and daily groceries to household electronics and cosmetics. Managing inventory precision across that SKU breadth, at 1,139 locations, processing 1.75 million orders daily, is not a human planning problem. It is a machine intelligence problem. Zepto's forecasting infrastructure is the system that keeps stockouts minimised, wastage in perishables contained, and the unit economics per order moving in the right direction even as scale compounds.

Route Optimisation and Last-Mile Control

Once an order is picked and packed, the clock is still running. Zepto's last-mile delivery operates on proprietary route-optimisation algorithms that calculate the fastest path for each delivery in real time, accounting for live traffic conditions, rider location, order volume, and proximity of concurrent deliveries. Unlike marketplace models that rely on third-party logistics networks, Zepto maintains full control over its delivery fleet. This is not accidental. Last-mile control is a prerequisite for the ten-minute guarantee, and it is the control point that allows the AI routing system to function without the variability introduced by third-party operators.

The integration of dark store density, demand forecasting accuracy, and real-time route optimisation is the operational triad that converts the ten-minute promise from a consumer aspiration into a repeatable, scalable, and increasingly efficient system. The adjusted EBITDA loss per order improved from 142.68 rupees in Q4 FY25 to 59.40 rupees in Q4 FY26, a more than 58 percent improvement in order-level economics in a single year, driven directly by this operational architecture compounding at scale.

‍

From Logistics Company to Data Intelligence Business

The Strategic Pivot That Competitors Missed

The most significant strategic move Zepto made in the 2024 to 2025 period was not network expansion. It was the recognition that the data generated by its operating model was a commercially monetisable asset in its own right, and that the company's long-term margin structure depended on building revenue streams beyond delivery fees and product sales.

India's Q-commerce gross merchandise value grew from 1.5 billion dollars in 2022 to an estimated 6 to 7 billion dollars in 2024, reaching approximately 963 billion rupees in CY2025. As the category matured, the competitive logic shifted. Every serious player, Blinkit under Eternal, Swiggy Instamart, and BigBasket's BB Now, was investing aggressively in dark store density and delivery speed. Logistics, the original differentiator, was becoming table stakes. The next layer of competitive separation would come from data.

Zepto had a structurally unique data asset. Millions of daily transactions, hyperlocal purchasing behaviour at the PIN-code level, real-time consumer demand signals across hundreds of urban micromarkets, and full-funnel visibility from search to cart to purchase to repeat. No traditional market research firm, no FMCG brand, and no advertising platform had access to this quality of hyperlocal consumer intelligence in real time. The company was generating this data as an operational byproduct. The strategic decision was whether to treat it as internal infrastructure or as a product.

Zepto Atom and the Creation of a New Revenue Category

In May 2025, Zepto launched Zepto Atom, a subscription-based analytics platform built for FMCG and D2C brand partners. Zepto Atom gives paying brands access to granular, real-time performance data including market share breakdowns at the PIN-code level, minute-by-minute sales tracking, customer impression and conversion data, repeat purchase patterns, and full purchase journey analytics.

The platform is positioned as a direct competitor to traditional market intelligence firms like Nielsen and Kantar, whose data is typically delayed, aggregated at the city level, and priced for multinational budgets. Zepto Atom offers neighbourhood-level precision, real-time reporting, and a pricing model accessible to new-generation D2C brands that operate on fast feedback loops and lean planning cycles. CEO Aadit Palicha stated at launch that the platform was built to give brands insights at a more competitive price point than conventional research, with the granularity that city-level aggregates have never provided.

The platform includes Zepto GPT, a natural language processing assistant trained on Zepto's proprietary transaction data. Brand managers can query the system in plain language, asking questions such as how to grow market share for a specific product category in Bengaluru, and receive data-backed recommendations and automated reports without requiring internal data science capability. This is the agentic AI model operating within a commercial B2B product: autonomous systems interpreting, synthesising, and surfacing intelligence without requiring human intermediation at every step.

The financial outcome of this strategic pivot is visible in the FY26 numbers. Advertising revenue, which includes Zepto Atom and the broader retail media stack, surged 151 percent year on year to 1,636 crore rupees in FY26. Platform services contributed an additional 564 crore rupees. These are the fastest-growing revenue lines in Zepto's P&L, and they scale in direct proportion to platform GMV without proportionate increases in operational cost. Zepto entered a consumer analytics market in India estimated at 1,000 crore rupees. Its advertising revenue line alone has already approached that number in a single fiscal year.

The Business Outcomes

The numbers Zepto has produced between FY24 and FY26 are the output of an operating model where AI is embedded at every stage of the value chain rather than applied as a point solution to isolated problems.

Operating revenue grew from 4,454 crore rupees in FY24 to 11,110 crore rupees in FY25 to 22,623 crore rupees in FY26, more than doubling in each successive year. Order volumes expanded at a compound annual growth rate of approximately 119.5 percent between FY24 and FY26, making Zepto the fastest-growing quick commerce platform in India by order volume among scaled players, per the Redseer Report cited in the UDRHP. Annual transacting users reached 47.97 million as of March 31, 2026, up 25 percent year on year. The company processed 64 crore total orders in FY26, averaging over 17 lakh orders daily across the full year and reaching 23.3 lakh daily orders in Q4 FY26 alone.

The adjusted EBITDA loss per order improvement from 142.68 rupees in Q4 FY25 to 59.40 rupees in Q4 FY26 is the most strategically significant number in the entire FY26 disclosure. It confirms that the unit economics of the model are improving at meaningful pace as AI-driven operational efficiency compounds with scale. Revenue from warehousing, packaging, and last-mile services more than doubled to 2,780 crore rupees. Advertising and platform services revenue together crossed 2,200 crore rupees, representing the early but accelerating monetisation of Zepto's data and media assets.

The IPO, with a fresh issue of 8,010 crore rupees managed by Goldman Sachs, Morgan Stanley, Motilal Oswal, JM Financial, HSBC, IIFL, and Axis Capital, marks Zepto's transition from a venture-backed growth story to a public market accountability structure. A pre-IPO placement of up to 1,602 crore rupees is also under consideration.

Where the Model Faces Structural Pressure

No case study of Zepto is complete without an honest account of the tensions in the model.

Profitability remains structurally distant. The net loss widened 26 percent year on year to 5,905 crore rupees in FY26, from 4,699 crore rupees in FY25. Total expenditure rose 79 percent to 29,026 crore rupees, with delivery and handling costs surging over 90 percent and procurement of goods accounting for 63 percent of all costs. The company has disclosed in its UDRHP that it may continue to incur losses and may not be able to sustain its historical growth rates. That is a standard IPO disclosure, but it reflects a genuine structural question: whether the economics of ten-minute delivery at national scale can reach profitability before the capital markets demand it.

Regulatory risk has emerged as a new variable. The Enforcement Directorate issued summons to co-founders Aadit Palicha and Kaivalya Vohra on April 8, 2026, under FEMA for documents related to foreign investments and shareholding patterns. The matter has been disclosed in the UDRHP and does not currently represent a legal bar to the IPO, but it introduces a governance dimension that institutional investors will scrutinise during the roadshow.

The competitive intensity of the Q-commerce category remains extreme. Blinkit, operating under Eternal (Zomato's listed parent), reported approximately 20 million monthly active users at peak versus Zepto's approximately 13 million MAU. Swiggy Instamart operates within a multi-product consumer platform with bundling advantages Zepto cannot replicate as a standalone operator. BigBasket's BB Now is pursuing a 1,000 plus dark store target. The dark store density advantage Zepto built is being pursued aggressively by well-capitalised incumbents, narrowing the window in which Zepto can claim a structural operational lead.

User retention showed its first strain in March 2026, with monthly active users recording a sequential decline, the first month-on-month dip since Zepto's rapid expansion phase. The company spent 1,389 crore rupees on advertising in FY26 to sustain user acquisition and retention, and plans to allocate a further 520 crore rupees from IPO proceeds to marketing through its subsidiary Zepto Marketplace Private Limited. The cost of holding user attention in a category where all competitors are spending aggressively is a compounding pressure on the path to profitability.

Zepto Atom, despite its strategic clarity, faces an adoption trajectory that is not purely a product challenge. Legacy FMCG organisations with established relationships with Nielsen and Kantar operate on long procurement cycles and internal data governance frameworks that slow platform analytics adoption even where the product is demonstrably superior. Building the partner ecosystem for Atom at the pace required to make it a material standalone revenue contributor is a genuine execution priority.

Strategic Implications for Leaders

1. Speed as a strategic constraint is a legitimate design principle for AI-first operating models. Zepto did not deploy AI to improve an existing logistics model. It designed an operating commitment, ten-minute delivery, that made AI the only viable path to execution. Leaders in high-velocity markets should interrogate whether their own performance commitments are demanding enough to force genuine AI integration rather than incremental AI enhancement. A constraint that cannot be met without AI is a more powerful forcing function than a mandate that AI be adopted.

2. First-party data is a balance sheet asset that most organisations are not valuing or monetising at the level the market will eventually demand. Zepto's construction of Zepto Atom is a case study in converting operational data into a standalone commercial product line, one that grew 151 percent in a single year and is now approaching the size of India's entire traditional consumer analytics market. Every organisation generating high-frequency consumer transaction data should be asking whether that data has a direct monetisable form beyond its internal optimisation application.

3. The gap between AI strategy and agentic execution is where most enterprises lose value. Zepto's architecture demonstrates that the compounding outcomes of AI deployment come not from individual tools applied to isolated problems, but from orchestrated, agentic systems operating continuously across the full value chain: location selection, demand forecasting, inventory management, route optimisation, consumer personalisation, and brand intelligence. The organisations that close this gap, between Consulting 4.0 strategy and agentic execution at scale, will define the competitive hierarchy in their categories. Those that do not will find themselves optimising within a category whose economics have already been structurally redefined by those who did.

Final Thoughts 

Zepto's rise from a 2021 Mumbai startup to an IPO-bound enterprise processing 2.33 million orders daily by Q4 FY26 is regularly characterised as a story of speed. That framing is accurate but incomplete. Speed was the market promise. The AI architecture was the delivery mechanism. And the data generated by that architecture is now the long-term business within the business.

What Zepto has built is not a delivery company that uses AI. It is an AI-orchestrated commerce platform that is visible to consumers as a delivery experience and to brand partners as a data intelligence platform. The distinction matters enormously for anyone studying this model for strategic lessons, because the investment required in machine learning infrastructure, dark store density, proprietary analytics capability, and agentic decision-making systems is not the investment of a logistics operator. It is the investment of a technology company competing for category ownership in a 617 billion dollar market.

The leaders who will extract the most from this case study are not those who see in Zepto a model to replicate. They are those who recognise in it the strategic logic that applies to any high-frequency, data-rich, consumer-facing business: the organisation that controls the data, orchestrates the intelligence, and embeds agentic AI across the full value chain will not merely outperform competitors in the near term. It will structurally redefine the category economics that all competitors must operate within.

Read our Other Case Studies

Blinkit’s Dark Store Network & AI Logistics