Updated November 2026.
For a decade, “AI in retail” meant recommendation widgets and chatbots that answered FAQs. That’s no longer the interesting part. Shoppers in India now pay for groceries inside a chat with an AI assistant, a UAE cinema chain has sold tickets through an AI agent, and the brands that win the next few years will be the ones whose catalogues those agents can read.
This guide covers 12 AI use cases across the whole retail business, says which are mature in India and the UAE today, and ends with a 90-day plan. We build AI agents for consumer brands, so we’ve been specific about where our own experience is deep and where it isn’t.
What is AI in retail?
AI in retail is the use of artificial intelligence across the retail value chain. It’s easiest to think about in three layers:
| Layer | What AI does | Example use cases |
|---|---|---|
| Discover and buy | Helps shoppers find, compare and purchase | Conversational commerce, AI shopping agents, personalization, visual search |
| Sell and serve | Handles the work of selling and supporting customers | WhatsApp agents, generative product content, order status and returns, store operations |
| Plan and supply | Decides what to stock, where, and at what price | Demand forecasting, inventory and replenishment, dynamic pricing, supply chain |
Most retail AI programmes start in the middle layer, because that’s where the cost and the customer complaints are. The money increasingly sits in the other two.
What is agentic commerce, and why should retailers care now?
Agentic commerce is shopping where an AI agent, rather than the customer, does the searching, comparing and paying, within limits the customer has approved. The customer says “order my usual groceries for Saturday”; the agent finds the products, checks prices and completes the payment.
It stopped being a concept in both of our markets over the past year.
India. In October 2025, Razorpay, NPCI and OpenAI launched a pilot of agentic payments on ChatGPT, built on UPI (Razorpay). In February 2026, Razorpay and NPCI brought the same capability to Anthropic’s Claude, with Zomato, Swiggy and Zepto among the merchants, so users could complete food and quick-commerce orders inside a conversation (Razorpay). The same month, Mastercard demonstrated India’s first authenticated agentic card transaction, and by September 2026 NPCI was reported to be preparing a dedicated agent protocol for UPI (Forkast).
UAE. In November 2025, Mastercard made its first Agent Pay transaction outside the US in the UAE, with Majid Al Futtaim, where an AI agent booked cinema tickets (WAM). In May 2026, Visa launched its Agentic Ready programme in the UAE, and research it commissioned found nearly 60% of UAE businesses either strongly interested in or actively exploring agentic commerce (Visa).
Here’s the part most brands miss: an AI agent doesn’t see your banner ads or your homepage. It reads structured product data. If your catalogue, stock and prices aren’t available to it, your products simply don’t exist in that conversation.
Is your catalogue ready for AI shopping agents?
- Every product has clean, structured attributes: size, colour, material, price, variants
- Live stock and price are available through an API, not just on a web page
- Return, exchange and delivery policies are written as clear rules, not marketing copy
- Product descriptions exist in the languages your customers search in
- Your store can be reached by AI agents through a standard interface such as the Model Context Protocol (MCP)
We build MCP-first and contribute to open-source MCP tooling; we explained the approach in Building AI-Ready Applications with MCP-First Development.
12 AI use cases in retail, and how ready each one is in India and the UAE
| # | Use case | What the AI does | Maturity for mid-market brands (2026) | What it needs first |
|---|---|---|---|---|
| 1 | Conversational commerce on WhatsApp | Answers product questions and takes orders in chat | Mature, widely deployed in India | Live catalogue and order system |
| 2 | Agentic commerce | Lets AI assistants find and buy your products for shoppers | Early: live pilots in India and the UAE | Agent-ready catalogue, payments partner |
| 3 | Hyper-personalization | Tailors products, offers and messages to each shopper | Mature in large e-commerce; growing in D2C | Unified customer and order data |
| 4 | Visual and multimodal search | Finds products from a photo or video | Mature at platform level; brands need good images and data | High-quality product images and attributes |
| 5 | Generative AI product content | Writes descriptions, translations and marketing copy at scale | Mature | Brand guidelines, product attributes |
| 6 | Order status and returns automation | Answers “where’s my order?”, starts returns and exchanges | Mature | Order system and courier tracking |
| 7 | Voice agents | Handles order, delivery and cash-on-delivery confirmation calls | Growing fast in India | Phone lines, order system |
| 8 | Store operations platform | Tracks audits, stock requests and tasks across stores | Mature as software; often still on WhatsApp groups | Store staff on mobile |
| 9 | AI demand forecasting | Predicts demand by product, store and day | Mature at large retailers; underused in the mid-market | Two or more years of clean sales data |
| 10 | Predictive inventory and replenishment | Decides what to reorder and where to send it | Growing | Forecasts plus live stock data |
| 11 | Dynamic pricing and markdown optimization | Adjusts prices and end-of-season discounts | Growing, with care | Competitor, stock and sell-through data |
| 12 | Smart store tech and supply chain resilience | Computer vision on shelves, frictionless checkout, disruption alerts | Mostly large chains; high capital cost | In-store hardware, supplier data |
The maturity ratings are our view, based on published data and the projects we see. The pattern is consistent: customer-facing use cases (1, 5, 6, 7) can go live in weeks; planning use cases (9 to 11) pay more but need clean data first.
How is generative AI transforming the retail customer experience?
Conversational commerce: India shops on WhatsApp
In India, WhatsApp is where shopping starts. A retail whitepaper from Meta and the Retailers Association of India found that 72% of product discovery in India happens through WhatsApp, and that retailers using it for commerce saw a 61% average improvement in return on ad spend (via TechBuzz). These are platform figures, so treat them as direction, not a forecast for your brand.
A useful WhatsApp commerce agent does four jobs:
- 1. Sells from the live catalogue. “Show me cotton kurtas under ₹2,000” returns three in-stock options with prices and a link to buy.
- 2. Knows what’s on which shelf. If store stock is readable, it can say whether a size is at the customer’s nearest store and hold it.
- 3. Tracks every order from the store and courier systems.
- 4. Starts returns and exchanges within policy, offering an exchange first, and hands complaints to a person with a written summary.
It runs on the brand’s existing WhatsApp Business number through an official provider, so customers keep the number they know.
Hyper-personalization: expected, not a bonus
McKinsey’s research found 71% of consumers expect personalized interactions and 76% get frustrated when they don’t get them, and that personalization most often lifts revenue by 10 to 15% (McKinsey). The research dates from 2021, but generative AI has made the same personalization affordable for mid-market brands: a WhatsApp agent that remembers a shopper’s size and last order is personalization, delivered in the channel they already use.
Visual search and generative product content
Google says Lens now handles more than 25 billion visual searches a month, and roughly one in five has commercial intent (Scalevise, citing Google). Shoppers photograph a product and search for it. Whether your product appears depends on image quality and structured product data, the same foundations agentic commerce needs.
Generative AI closes the content gap behind both: product descriptions written at scale, translated into Hindi, Arabic or regional languages, and kept consistent with brand guidelines. It’s one of the cheapest, fastest wins in retail AI, provided someone reviews the output before it goes live.
How do retailers use AI for demand forecasting, inventory and pricing?
This is where the larger money is, and where mid-market brands in India and the UAE are furthest behind.
Demand forecasting. McKinsey’s research found AI-driven forecasting can cut forecasting errors by 20 to 50% and reduce lost sales from product unavailability by up to 65% (Oracle, citing McKinsey). For Indian brands, the hard part is the calendar: Diwali, wedding season and sale events swing demand far more than an average month suggests. In the UAE, Ramadan, Eid and the Dubai Shopping Festival do the same.
Predictive inventory and replenishment. A forecast is only useful if it changes what gets ordered and where it goes. The same agent pattern we use for procurement, where an agent runs sourcing and quotes inside SAP and halved the PR-to-PO cycle for a pharma manufacturer, fits store replenishment.
Dynamic pricing and markdown optimization. AI can adjust prices to stock, demand and competitor moves, and decide when and how deeply to discount at the end of a season. Use it carefully: frequent visible price changes can damage trust, and marketplace price-parity rules limit what you can do. For most mid-market brands, markdown optimization is the safer first step.
Supply chain resilience. AI agents can watch supplier lead times, flag delays and suggest alternative suppliers before a stockout. It works only when supplier data is digital, which brings us to the unglamorous prerequisite for everything in this section: one clean view of stock and sales across stores, website and marketplaces. That’s what a store operations and automation platform provides.
What we’ve built for retail and consumer brands
We’ll be precise here. Our deepest experience is in commerce and customer conversations, not store chains.
- Interakt. Our team took Haptik’s idea for a WhatsApp sales platform to a production version one in seven months, designed from day one for a billion messages a year. Haptik has run and grown it since; it’s now used by more than 50,000 businesses.
- Boondi for Bombay Sweet Shop. A WhatsApp and voice agent that answers from live Shopify orders in Hindi, English and Hinglish, and captured more than 1,200 sales leads in its first two months. Read the Boondi case study.
- Haptik’s enterprise bots. More than 100 enterprise bots over seven years, including for retail and FMCG brands, on a platform with 15 billion-plus conversations.
- Operon. A contact centre that decides call by call which conversations an AI voice agent handles, the model we use for order-status and returns calls.
We haven’t yet published a case study with a retail chain, and we haven’t shipped a retail demand-forecasting system. We scope a first project with each new retail client so that it can become one. If you’re a food brand, the same Boondi work is covered in more depth on our food and beverage page.
AI in retail in India vs the UAE
| India | UAE | |
|---|---|---|
| Where shopping starts | WhatsApp, marketplaces and quick commerce | Malls, delivery apps and brand apps, with very high AI usage: 73.3% of working-age people used generative AI in Q2 2026 (Microsoft, via Middle East AI News) |
| Agentic payments | UPI-based pilots via Razorpay and NPCI on ChatGPT and Claude | Mastercard Agent Pay, Visa Agentic Ready with UAE banks |
| Languages | Hindi, English, Hinglish and regional languages, by chat and voice | Arabic and English, often in the same conversation |
| Demand peaks | Diwali, wedding season, sale events | Ramadan, Eid, Dubai Shopping Festival |
| Data law | Digital Personal Data Protection Act, 2023 | Federal data protection law; DIFC and ADGM have their own |
The first two rows of the India column and the language and peak rows reflect our field experience as much as published data.
Where should a retail brand start with AI? A 90-day plan
| Weeks | What happens | Use cases covered |
|---|---|---|
| 1–4 | Launch a WhatsApp agent on your live catalogue and order system; prototype in week one, production in about six | 1, 6 |
| 3–8 | Clean and structure product data; generate missing descriptions and translations; make the catalogue agent-ready | 2, 4, 5 |
| 6–10 | Put stock, audits and sales across stores and channels on one platform | 8 |
| 10–13 | Use that clean data to baseline forecast accuracy and pick the first forecasting or markdown pilot | 9, 11 |
One channel first, then the next. It’s the same approach we recommend everywhere, because most AI pilots stall when they try to do everything at once. If you’d rather start with a ranked plan than a build, our AI strategy and roadmap work does exactly that.
What to measure
| Metric | Use case | Why it matters |
|---|---|---|
| Median first-reply time on WhatsApp | Conversational commerce | Slow replies lose the sale |
| Chat-to-order conversion | Conversational commerce | Shows whether the agent sells or only answers |
| Share of order-status queries resolved without a person | Returns and service | Cleanest measure of time saved |
| Exchanges as a share of return requests | Returns | Revenue kept instead of refunded |
| Forecast error (MAPE) by product and store | Demand forecasting | The baseline every forecasting claim should be measured against |
| Stockout rate and sell-through | Inventory and pricing | Shows whether forecasts changed what was on the shelf |
Seven questions to ask a retail AI vendor
- 1. Does the agent read our live catalogue and stock, or a copy that goes stale?
- 2. Can we keep our existing WhatsApp Business number?
- 3. How will AI shopping agents, on ChatGPT, Claude or others, be able to find and buy our products?
- 4. What happens when a customer complains or asks for something outside policy?
- 5. Which of our systems are already connected in your live agents, and which need building?
- 6. Which languages are live today, and how do you test new ones?
- 7. Who owns the conversation data, the product data and the configuration?
Make sure AI agents can sell your products: see what we’d build first on the AI for retail page. For how we build agents in general, see Agentic AI in Production.
Frequently asked questions
What is AI in retail?
AI in retail is the use of artificial intelligence across the retail value chain: helping shoppers discover and buy, selling and serving customers, and planning stock, prices and supply. Common uses include conversational commerce, personalization, demand forecasting, inventory optimization and dynamic pricing.
What is agentic commerce?
Agentic commerce is shopping where an AI agent searches, compares and pays on a customer’s behalf, within limits the customer has approved. In India it is piloting on UPI through Razorpay and NPCI on ChatGPT and Claude; in the UAE, Mastercard completed its first Agent Pay transaction outside the US in 2025.
Can customers in India shop and pay through ChatGPT?
Yes, in pilots. Razorpay, NPCI and OpenAI launched agentic payments on ChatGPT using UPI in October 2025, and Razorpay and NPCI extended agentic payments to Claude in February 2026 with merchants including Zomato, Swiggy and Zepto. Availability depends on the merchant taking part.
How is generative AI transforming the retail customer experience?
Generative AI lets retailers answer product questions instantly in chat, personalize recommendations and messages, write product descriptions in several languages, and support visual search. In India much of this runs on WhatsApp; increasingly, AI shopping agents also complete purchases directly.
How do retailers use AI for demand forecasting?
Retailers use AI to predict demand by product, store and day using sales history, seasonality, promotions and external signals. McKinsey’s research found AI-driven forecasting can reduce forecasting errors by 20 to 50% and cut lost sales from unavailability by up to 65%.
What are the best AI use cases in retail and consumer packaged goods (CPG)?
The fastest wins are conversational commerce, order status and returns automation, and generative product content. The largest long-term value comes from demand forecasting, inventory optimization and pricing, which need clean sales and stock data first.
Can an AI agent sell from our Shopify catalogue on WhatsApp?
Yes. An agent can read products, sizes, prices and stock from Shopify or a POS, share them in the chat and send a link to buy, on your existing WhatsApp Business number.
Do you build retail AI agents for brands in the UAE?
Yes. KnackLabs builds AI agents for mid-market companies in India, the UAE and Israel, with a senior engineer embedded in the business. Hindi, English and Hinglish are live today; other languages, including Arabic, are added and tested with the brand before launch.

Book a 30-minute call and we’ll show you what we’d build first for your brand.
Sources
- Razorpay, Razorpay, NPCI and OpenAI launch agentic payments (October 2025)
- Razorpay, Razorpay and NPCI launch agentic payments on Claude (20 February 2026)
- Forkast, India eyes AI-driven UPI payments as NPCI weighs agent protocol (September 2026)
- WAM, Mastercard launches agentic payments in UAE (19 November 2025)
- Visa, Visa launches Agentic Ready programme in UAE (20 May 2026)
- TechBuzz, From Scroll to Chat to Cart (on the Meta and Retailers Association of India whitepaper)
- McKinsey, The value of getting personalization right, or wrong, is multiplying (2021)
- Oracle, AI in demand forecasting (citing McKinsey)
- Scalevise, Google Lens tops 25 billion monthly searches (citing Google, June 2026)
- Middle East AI News, on Microsoft’s Global AI Diffusion report for Q2 2026
- KnackLabs, AI for Retail, Boondi, Operon and procurement agent
Note: This article is for general information only and is not business, legal, payments-regulatory or investment advice. Agentic payment programmes described here were pilots or early launches at the time of writing and may change. Third-party figures belong to their publishers, are summarised in our own words and reflect the sources as of November 2026. Maturity ratings are KnackLabs’ opinion. Results from KnackLabs projects are specific to those deployments and do not guarantee similar results elsewhere. KnackLabs accepts no liability for any loss arising from reliance on this article. WhatsApp, Shopify, ChatGPT, Claude, Mastercard, Visa, UPI and other third-party names and trademarks belong to their respective owners.
- AI Transformation
- Retail
- Consumer Brands
- D2C
- Agentic Commerce
- Generative AI
- Demand Forecasting
- India
- UAE



