The Blooming Love Triangle in Tier 2/3 India: The Shopkeeper, The AI Agent and Meta

A retail journey and its actors are looking across a new beautifully dressed-up Partner, the Agents powered by AI.

Part 1: The Phone Call

Last week I got a call from a shop owner in Sivakasi, my hometown, a Tier 3 city, known more for firecrackers than digital commerce. He runs one clothing store and his question was whether I could build an AI agent or automation for his sales or CRM operations?

Before answering, I asked how he sells today. His reply was faster : Instagram and WhatsApp. No website, no product catalogue, no customer database, and he admitted that he has no time left to reply to every message himself. He had almost none of the raw material we assume an AI system needs, but he was ready to hand the whole thing to an AI after the call.

I said I would put together a proposal. Then I sat with the call, and a different picture formed. He thinks the AI is a new relationship. He has not quite noticed that he is already married. His shop lives inside someone else’s house, and has for years. Now he is looking across this beautifully dressed-up AI partner. This post talks about the love triangle among the shopkeeper, the AI agent he wants, and the partner he already has.

Part 2: Growth, The Dependency and The Threat

Digitization Growth Maturity

Two truths sit in parallel in Tier 2 and Tier 3 India, and they do not agree.

The first is that these towns are now the center of gravity. Roughly 63% of India’s e-commerce orders already come from Tier 2 and Tier 3 cities. India’s Digital Maturity Index rose to 60.8 in 2026, up from 55.9 in 2023. And AI is arriving here faster than in the metros: adoption is reported to be growing around three times year on year, led by smaller cities, and India is now the single largest market in the world for Meta AI.

The second truth is that the depth is thin. India has about 65 million small businesses, but only around 5 million of them sell online at all. By one industry report, only about 5% of MSMEs are fully digitized, though more than half have adopted at least some digital tool. My Sivakasi shopkeeper acts as another evidence - present online, not built online. Selling, but with no system underneath.

So the honest reading is not “Tier 3 is behind” but “Tier 3 is online without being digitized, and AI is landing on that thin layer first”. That is a strange, combustible starting point.

Dependency: Meta already owns five of the seven steps on the customer journey

Here is the map that changed how I saw the call. Trace one customer through a full retail journey and mark who owns each step.

Discovery and inspiration: Instagram, Reels, Facebook. Meta.

Research and questions: ChatGPT, Gemini, Google Search, WhatsApp chat, and increasingly Meta AI (I’m assuming here because that RayBan Ad could be a flash point), where 63% of usage already happens inside WhatsApp. Partially Meta.

Trust and consideration: Now the WhatsApp group, the forwarded recommendation, the local influencer. Meta. (Could move from group friend to Meta AI Word - Refer my previous article)

Decision and catalogue: WhatsApp Business catalogue sits here, though Meesho and Flipkart share this step. Split, tilting to Meta.

Payment: UPI, where PhonePe and Google Pay hold nearly 80% and WhatsApp Pay sits at about 0.65%. Not Meta.

Fulfilment and delivery: sellers and logistics. Not Meta.

Repeat and support: back to WhatsApp. Meta.

Five of seven, and the two it does not hold are payment and delivery. The reach behind this is not subtle. Instagram has over 480 million users in India, its largest audience anywhere. Facebook has over 400 million, also the largest. WhatsApp has more than 550 million monthly users and reaches around 85% of the country’s internet users. Meta’s three apps together touch the entire internet population in India. For a Tier 3 shop, “going digital” and “being on Meta” have quietly become the same sentence.

Threat: Are there anyone naming it

The reason to care is the one step Meta does not yet own. In January 2025, the payments regulator removed the cap that had limited WhatsApp Pay, letting it offer UPI to all 500 million plus WhatsApp users in India. The loop was there to be closed. And yet WhatsApp Pay is still stuck near 0.65% of UPI. Meta owns the shop window, the conversation, and now the AI assistant, and is reaching, so far without success, for the wallet as well.

Others have started to name the risk plainly. As one Indian tech newsletter put it, WhatsApp is not a product Indians use, it is infrastructure Indians depend on. The traders’ own body captures the contradiction perfectly: the Confederation of All India Traders has campaigned loudly against the dominance of large digital platforms, and also signed up with Meta to train ten lakh traders to sell on WhatsApp Business. They are fighting the landlord and moving in at the same time.

The government engine has tried but mostly missed. In November 2024 the Competition Commission of India fined Meta 213 crore rupees and barred WhatsApp from sharing user data with other Meta apps for advertising for five years. A year later, the appellate tribunal kept the fine but struck down the data-sharing ban. India’s real bet is not punishment, it is building open roads alongside the private one. As Nandan Nilekani, the architect behind Aadhaar and UPI, said of the government’s open commerce network: “We are going to democratize e-commerce in India. It won’t be limited to a few players.” UPI stayed interoperable by law. ONDC is trying to do for commerce what UPI did for payments. Bhashini is opening up the languages. Whether any of it reaches a shopkeeper who just wants more sales this month is the open question.

Part 3: The “would be” path

Two scenarios, and what each does to research

I do not know which way this goes. So let me lay out two honest scenarios and, since research is my trade, ask what each would do to the study of buyers.

Scenario one: the agent moves in with Meta. The easiest thing to build is an agent inside WhatsApp, because that is where the billion users and the shopkeeper’s existing relationship already live. Business AI on WhatsApp, launched for Indian sellers in May 2026, points straight here. Discovery, advice, order, and eventually payment all close inside one blue-ticked thread. The behavioral questions this raises are uncomfortable. If a buyer never leaves one app, can we even observe a choice, or only a path? How do you separate a real preference from a nudged one inside a closed loop? And what happens to independent market research for retail when the seller, the buyer, and the panel all live inside one company’s walls, visible only on its terms?

Scenario two: the AI agent arrives as an outsider. The thing that broke open web search could break this too. An agent that sits above the apps and treats Meta as just one source, pulling offers across Meesho, ONDC sellers, and the local store, in Tamil or Bhojpuri. Here the disruptor is whoever pairs a neutral agent with the wallet Meta lacks, a payments giant like PhonePe or Paytm, or an ONDC-native buyer app. The research questions flip entirely. When a Tier 3 buyer is shown real choice for the first time, in their own language, what do they optimize for, price, trust, habit, or the voice that feels most like home? Does choice overload arrive for people who never had choice? Who do they trust when the recommendation is language-native but brand-agnostic? Nobody has measured any of this, which is precisely why it interests me.

Notice the asymmetry. Scenario one mostly needs old research, the study of lock-in and habit, which we understand. Scenario two needs research that does not exist yet, because no one has watched a first-time buyer choose across ecosystems through an agent. The more open the future, the less we actually know about the people living in it.

The questions I am sitting with

If Meta already owns five of the seven steps, does an AI agent close the last two for it, or pry the first five back open?

When the shopkeeper finally gets his agent, whose interests has it quietly been built to serve, his, the buyer’s, or the house they both live in?

And the one closest to my work: if the whole journey moves inside a single trusted thread, who is left outside it to study what is actually happening?

I write about AI, behavioral economics, and user research. Mostly questions, occasionally answers. If you are thinking about any of this, I would like to compare notes.