Quick commerce moves fast. Forecasting demand for it accurately? That requires next-level Data Science. 🛒⚡ The team at Swiggy has just published a deep dive on Swiggy Bytes into how Instamart revamped its demand forecasting by leveraging Google’s TimesFM foundation model and transitioning from flat forecasts to deep, hierarchical modeling. Building a predictive engine for thousands of SKUs and rider fleet planning comes with massive challenges—like censored sales data from stockouts and high variance from product substitutions. Here is how the team solved it: 🔹 𝐀𝐯𝐚𝐢𝐥𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐃𝐞-𝐁𝐢𝐚𝐬𝐢𝐧𝐠: Correcting historical data to account for when items were actually out of stock. 🔹 𝐆𝐫𝐨𝐮𝐩 𝐒𝐊𝐔 𝐀𝐠𝐠𝐫𝐞𝐠𝐚𝐭𝐢𝐨𝐧: Smoothing out demand variance caused by customers substituting similar products. 🔹 𝐃𝐮𝐚𝐥-𝐇𝐢𝐞𝐫𝐚𝐫𝐜𝐡𝐲 𝐅𝐫𝐚𝐦𝐞𝐰𝐨𝐫𝐤: Separating the overarching business objectives from the optimal modeling level. 𝐓𝐡𝐞 𝐑𝐞𝐬𝐮𝐥𝐭𝐬? 🚀 21–38% relative accuracy improvement for warehouse forecasting across categories. 🚀 7.5% WMAPE improvement & a massive 50% reduction in runtime for OPD forecasting compared to their previous TFT-based system. Huge shoutout to the authors SAHIB MAJITHIA, Satpalsingh Ghunia , and Mano Ranjith Kumar M for sharing the architecture behind this! Check out the full read here: https://lnkd.in/eRB9CxiV #GenerativeAI #SupplyChain #TimeSeriesForecasting #QuickCommerce #DataScience #MachineLearning #SwiggyInstamart #TechBlog
Swiggy Instamart Demand Forecasting with Google TimesFM
This title was summarized by AI from the post below.
More Relevant Posts
-
📢 We are thrilled to share our latest article with you! 📄 TITLE: AI Inventory Planning for Lean E‑commerce Stores 🔍 OVERVIEW: Supply Chain Technology AI Inventory Planning for Lean E‑commerce Stores How predictive […]... 👉 Read the full article here: #DigitalTransformation #ProfessionalGrowth #TechTrends https://lnkd.in/dNuA-hqy
-
Selling profitably on Amazon has become too complex for yesterday’s decision-making model. We’re excited to be at 𝗲𝗧𝗮𝗶𝗹 𝗘𝗮𝘀𝘁 𝗕𝗼𝘀𝘁𝗼𝗻 showcasing 𝗖𝗼𝗴𝗲𝗻𝘁𝗶𝗾 𝗲‑𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗲 - our AI-native, agentic profit engine built to help brands stay ahead of growing complexity and margin leakage across marketplaces like Amazon. 🔸 Continuously monitors 70+ profit signals 🔸 Recommends the right actions in minutes 🔸 Orchestrates execution across the digital shelf, supply chain, and retail media But that’s not the breakthrough. The breakthrough is a fundamentally new decision-making model - one that enables brands to continuously sense, decide, and act at the speed modern marketplaces demand. If you’re responsible for Amazon growth, profitability, or e‑commerce performance, join us on the 𝗔𝗜 𝗦𝘁𝗮𝗴𝗲 𝗮𝘁 𝟮:𝟱𝟱 𝗣𝗠 𝗼𝗻 𝗔𝘂𝗴𝘂𝘀𝘁 𝟭𝟬 as Prabal Chaudhri uncovers why winning on Amazon now requires a new way of making decisions. You’ll learn: 🔸 Why traditional decision-making breaks down on Amazon (and where it leaks margin) 🔸 Which capabilities matter most when evaluating AI solutions for digital commerce 🔸 How continuous, cross-signal reasoning can compress detection-to-execution from days to minutes #EcommerceAI #AgenticAI #AmazonMarketplace #RetailAnalytics #eTailBoston [Image Description: Promotional graphic for an eTail Boston session titled “Why winning on Amazon requires a new decision-making model,” scheduled for August 10 at 2:55 PM at Sheraton Boston, featuring Prabal Chaudhri, Head of AI Products, CPG at Fractal.] Matt Gennone | Prabal Chaudhri | Jamie Podhaizer | Jessica Halper, MS | Ganesh Subramanian | Sumith Balagangadharan | Anshu Aggarwal | Ankur Shrimali | Arpit Sodani
-
-
📌 Explore key insights and expert analysis in our new post: 📄 TITLE: AI Inventory Planning for Lean E‑commerce Stores 🔍 OVERVIEW: Supply Chain Technology AI Inventory Planning for Lean E‑commerce Stores How predictive […]... 👉 Read the full article here: #Programming #Development #SoftwareEngineering https://lnkd.in/dNuA-hqy
-
Faster than ever: Walmart is using AI to deliver in under an hour. Retail is being reshaped by speed. Walmart is proving that artificial intelligence isn’t just about smarter predictions — it’s about getting products into customers’ hands faster than ever. By combining predictive insights with real-time logistics, Walmart is setting a new benchmark for delivery excellence. This isn’t just about efficiency — it’s about redefining customer expectations in the age of instant commerce. E-commerce growth: Walmart’s global online sales grew 26% in Q1, with U.S. sales also up 26%. Intelligence layers: AI models predict what customers will order and place products in the right part of the supply chain. AI orchestration: From inventory forecasting to route optimization, Walmart’s algorithms act like a conductor ensuring every product moves at the right time. Breakthrough tech: Advances in large language models (LLMs) enable predictive analytics at scale, helping Walmart anticipate demand and optimize delivery. Customer focus: AI maps external factors like weather and traffic, ensuring deliveries are not just fast but reliable. Vision ahead: Walmart is investing in custom AI models for selection, routing, and scheduling, aiming to make sub-hour delivery the new standard. As Walmart’s CTO Suresh Kumar put it: “AI is now this conductor of this orchestra of product movement.” The future of retail isn’t just digital — it’s intelligent, predictive, and lightning fast. #Walmart #AI #SupplyChainInnovation
-
Really happy with how this piece turned out! It's a quick look at Walmart's speed chapter and how AI models built internally are making it easier to better predict what customers are going to order and when, and put the right product in the right part of its supply chain so it gets to people faster. Comments and insights from Suresh Kumar, Scot Wingo and Andy Szanger. Please read it in your next coffee break! https://lnkd.in/gNe_jrK7
-
Walmart's global e-commerce business grew 26% in the first quarter. Behind that number is a bet on AI that's worth paying attention to. Suresh Kumar described the AI behind it as the conductor of an orchestra of product movement, tying forecasting, replenishment, scheduling, and routing into one system instead of five disconnected ones. Most retailers still run those as separate teams with separate blind spots. Walmart treating them as a single intelligence layer changes the question from what a customer will buy to which store can actually get it to them fastest once you factor in staffing, weather, and traffic. I talked with Vidhi Choudhary about that shift for this piece. This is quick, easy read, that is worth your time. CDW
Really happy with how this piece turned out! It's a quick look at Walmart's speed chapter and how AI models built internally are making it easier to better predict what customers are going to order and when, and put the right product in the right part of its supply chain so it gets to people faster. Comments and insights from Suresh Kumar, Scot Wingo and Andy Szanger. Please read it in your next coffee break! https://lnkd.in/gNe_jrK7
-
For the longest time, I kept asking myself one question: Why do marketplace teams still spend hours jumping between Amazon, Flipkart, Shopify, Blinkit, reviews, inventory sheets, pricing dashboards, and endless Excel files just to understand what's happening? So I decided to build something. Over the past few weeks, I designed an AI-powered marketplace operating system that acts like a team of specialists working together. Instead of showing data, it answers questions like: • Which SKUs need immediate attention? • Where are we losing revenue? • Which listings need SEO improvements? • Which products are about to go out of stock? • Are competitors undercutting our pricing? • What are customers repeatedly complaining about? The idea is simple: Less dashboard watching. More decision making. This isn't just another analytics dashboard. It's an AI command center where Catalog, Pricing, Inventory, Review, Creative, and even a CEO Agent work together to help businesses grow across marketplaces. Building this made me realize something: The future of e-commerce won't belong to companies with the most data. It will belong to companies that can turn data into decisions instantly. This is still a work in progress, and I'd genuinely love feedback from founders, e-commerce leaders, and marketplace professionals. What would you add to an AI operating system like this?
-
-
13/30 — Demand Forecasting & Data Analytics in Quick Commerce How does Quick Commerce predict what customers will need before they place an order? In Quick Commerce, delivering products within 10–30 minutes requires more than just fast delivery. Platforms need to know what products customers are likely to buy, when demand will increase, and where inventory should be placed. This is where Demand Forecasting & Data Analytics play a major role. • Demand Forecasting — Uses historical sales, seasonal trends, festivals, weather, and local demand to predict future orders. • Customer Data Analysis — Helps understand customer preferences, buying frequency, popular products, and peak ordering times. • Real-Time Analytics — Monitors orders and changing demand throughout the day, allowing businesses to respond quickly. • Smart Inventory Planning — Ensures the right products are available at the right dark store, reducing stockouts and excess inventory. • AI & Machine Learning — Continuously analyses large amounts of data and improves demand predictions over time. Why It Matters ? • Reduces stockouts • Minimizes excess inventory and wastage • Improves order fulfilment speed • Optimizes inventory across dark stores • Reduces operational costs • Improves customer satisfaction The key idea: In Quick Commerce, the right data helps predict the right demand, and the right demand helps deliver the right product at the right time. #QuickCommerce #DemandForecasting #DataAnalytics #ArtificialIntelligence #MBA #LearningInPublic #LinkedInLearning
-
-
Ever checked an Amazon product twice and found a different price just a few hours later? 👀 It’s not random. It’s analytics. Amazon uses dynamic pricing—a strategy powered by real-time data and predictive analytics—to adjust prices based on changing market conditions. Some of the factors that influence pricing include: 📈 Customer demand 💰 Competitor pricing 📦 Inventory levels 📅 Seasonal trends 🛒 Shopping behavior ⏰ Time of day Instead of relying on fixed prices, analytics enables businesses to make smarter pricing decisions in real time—balancing customer demand, competition, and inventory. This is one of the best examples of how data isn’t just supporting business decisions—it’s making them. The next time you notice a price change, remember: Behind it is a powerful combination of analytics, AI, and millions of data points. #AnalyticsAroundUs #DataAnalytics #BusinessAnalytics #ArtificialIntelligence #DataScience #MachineLearning #BigData #PricingStrategy #Ecommerce #BusinessIntelligence #TechTrends #Amazon
-
Sunil Rathee Really enjoyed reading this. One of the most practical and honest write-ups I’ve come across on demand forecasting for quick commerce. I had a couple of technical questions, if you’re open to sharing some more details: For availability debiasing, how are you correcting the historical sales and future features? Is it based on an empirical correction using observed availability, or do you have a separate model estimating latent demand? I’m especially curious about SKUs that stay unavailable for long periods (seasonal items), where simply training on available periods doesn’t seem sufficient. Also, for the hierarchical setup using TFT and TimesFM, are you forecasting at a higher aggregation (SKU × warehouse/city, for example) and then allocating to stores? Reason being time complexity of these models for store<>sku level training and scoring would be too bad. If yes, what does the allocation ratio consist of? Is it based on historical fully-instock sales share, OPD, demand share, or some learned allocation model? Would love to understand these aspects a bit better. Great work by the team!