Teardown · Open networks & digital commerce

Discovery, Reimagined

Searchable Is Not Discoverable

The big question

An open network can make everyone available. How do you make everyone discoverable?

The three-minute version

An open network can make an enormous range of buyers and sellers available to each other. That is not the same as making them discoverable: more supply without better discovery only means more work for the buyer. This series imagines open infrastructure underneath, personal intelligence in the middle and a human way of meeting sellers on top, with the smallest seller able to join by talking. It builds on open protocols, multilingual access and agent payments that already exist. Nine shifts, each with its sources and the result that would end it.

The test that matters most

Do buyers find a better match faster through an agent and short introductions than through a listing grid, and do small sellers they would not otherwise have found win business?

Where things stand The shifts How we’d test it Evidence and sources

02 · where things stand

Where things stand.

The question

The internet made almost everything searchable. Marketplaces then made it discoverable: they brought supply into one place and made it convenient to catalogue, rank, compare and buy. They made digital commerce possible for millions of people.

Open networks now let any buyer app, seller app and service provider connect through shared standards. ONDC delivered its first order in Bengaluru on 29 April 2022^[1], and by July 2026 it had passed 500 million cumulative transactions, according to figures the Department for Promotion of Industry and Internal Trade gave a trade publication^[9]. A network like that creates extraordinary diversity: a multinational and a home baker can sit on the same network.

It also creates a new challenge. Searchability is not the same as discoverability. More supply without better discovery only means more work for the buyer.

Precision narrows the universe. Discovery opens it back up.

The idea

Open infrastructure underneath. Personal intelligence in the middle. Human discovery on top.

People express what they need, in their own words. Their agent narrows an enormous network to what is relevant. And discovery becomes an experience: meeting sellers through short, human introductions, with evidence and protection a tap away.

The smallest seller is the test. If a network is truly open, one person with a phone and a real capability should be able to join and be found when she is genuinely relevant, without first becoming a small version of a large company.

What already exists

Where	What already exists
An open discovery protocol	ONDC is built on the Beckn protocol, described by researchers as a “universal resource discovery and transaction protocol”. A buyer app can see offers from many seller apps, and an optional gateway supports discovery across them^[1].
Registries of participants	Registries list the apps on a network and hold their public keys so that messages can be verified. They are organised hierarchically, a design the researchers compare to the internet’s DNS^[1].
Fair search	ONDC’s 2022 consultation paper proposed that gateways forward a search to all relevant seller apps without filtering, that buyer apps publish their ranking criteria, and that sponsored results be labelled^[2].
Voice and local languages	ONDC’s Saarthi reference buyer app, built with Bhashini, launched with five languages and a plan for all 22^[3]. Sahayak, a WhatsApp chatbot, handles onboarding and service queries in five languages^[4].
Agents that act for people	In October 2025 Razorpay, NPCI and OpenAI announced a pilot in which an agent places an order after one confirmation from the user^[5]. Google’s AP2 protocol uses signed “mandates” that record what a user has authorised an agent to do^[6].
Consent-based sharing	India’s Account Aggregator framework lets people share financial data between institutions only with their consent, which they can revoke^[7]. The new data-protection rules provide for consent managers^[8].

Scope. The consultation paper’s ideas were proposals put out for comment in 2022, and the press coverage of Saarthi and Sahayak comes from announcements. We treat them as foundations to test, not as settled facts.

03 · the shifts

Nine shifts, each one testable.

Each shift moves from today’s model to a possible alternative. Open one to see how it would work, the sources behind it, what could go wrong and the result that would end it.

04 · how we’d test it

The smallest real test.

The set-up

One city and one category. Start with intent-led discovery and the onboarding of micro-sellers by conversation, launch one discovery experience, then add matching, agents and a second discovery provider. Every kill threshold is written down before launch.

What each test would show

Shift	What is tested	The question it answers
Shift 1	Intent against search	Does an agent find a better match, faster, than traditional search?
Shift 2	Capability profiles	Do profiles for 200 participants answer open-ended requests that category search can’t?
Shift 3	Reels against a grid	Do short introductions beat a listing grid on outcomes, and do sellers keep making them?
Shift 4	Why-you profiles	Do buyers choose differently when they can see what makes a seller distinct?
Shift 5	Micro-sellers	Can 100 sellers with no digital presence join by conversation alone, and do they get orders?
Shift 6	Matching	Do matches between buyers and 30 makers turn into orders, and are the agents fair to both sides?
Shift 7	Plural discovery	Can two or three providers share one network without sellers or buyers being locked in to one?
Shift 8	Red team	Does every deliberate attack on the design fail, and do the five health questions hold?
Shift 9	Moving freely	Can participants and buyers move to a second provider without losing identity, history or reputation?

reference

Behind the teardown.

The rules it has to fit, how we would bring it to life, what it would take and how to read it.

Four principles throughout

AI should reduce the search space without reducing the possibility space.

Let stories earn attention, evidence earn confidence, and outcomes build reputation.

You can pay to communicate. You can’t pay to become more relevant.

Agents negotiate the complexity. People own the commitment.

Trusted discovery

Discovery and trust work together. Reputation, protected payments and evidence help decide who is worth discovering, so a buyer can ask not just “who can do this?” but “who can do this, under conditions I trust?” That half of the question is explored in the Trust teardown, whose first shift is its counterpart to this series.

How we’d bring it to life

Insight Visualization. Agree what must be true: that intent-led, reel-based discovery helps buyers find better matches and helps small sellers get found.

Situational Awareness. Map how people discover sellers in one city and one category today, and how small sellers reach their customers.

Evolution Cycles. Onboard micro-sellers by conversation, launch one discovery experience, then add matching, agents and a second discovery provider.

Grow what works, category by category, keeping the five-question test of Shift 8 at the centre.

What it would take

Business Building, Product & Technology Engineering, Systems Architecture, Data & AI, Market Presence & Distribution, Business Model Innovation and Operating Model Design, with open-network bodies, sellers and consumer groups involved from day one.

How to read this teardown

This is a Level 1 · Concept teardown: the idea, how it would work and how we would test it. Every figure is sourced and every policy claim is cited, and the sources are listed at the foot of each page. Examples marked illustrative are constructed to explain an idea, and the numbers inside them are not data. No separate evidence review has yet been commissioned for this series, so it does not carry the claim-by-claim evidence ratings that Level 2 teardowns do.

The sellers and buyers in the examples are invented.

evidence and sources

What the evidence supports.

Every claim sits in one of four tiers. Sources carry the status of our last check.

How strong it is

Established: ONDC and the Beckn protocol separate buyer apps from seller apps, so a buyer app can see offers from many; multilingual voice and chat access for sellers has begun; agent-assisted payments are in pilot; short video is a major discovery surface in India (Meta-commissioned studies, directional).
Evidence-supported: Open networks need governance as well as protocols: researchers who studied Beckn and ONDC found that governance, not the protocol alone, is central to adoption.
Hypothesis: That intent-led, agent-assisted discovery finds better matches faster than search; that capability profiles surface sellers that category search misses; that a plural discovery layer stays contestable.
Unknown: How much discoverable capability a city holds; whether sellers will keep profiles current; whether providers will adopt a common capability schema; whether people will delegate to agents, and within what limits.

Before a pilot we would confirm

ONDC scale figures: confirm the 500 million cumulative total with DPIIT or ONDC Open Data.
Beckn specification: confirm how a search request represents a buyer’s intent (the paper reviewed does not discuss intent-based search).
Primary FSSAI regulation text for registration thresholds not read (tax-advice summary used).
Whether ONDC has adopted the 2022 consultation proposals on search fairness and sponsored-result labelling.

Notes on the evidence

The ONDC scale figures come from a trade-press report that attributes them to DPIIT. We have not confirmed them with DPIIT, and the report’s monthly and annual figures do not sit easily together, so we use only the cumulative total.

The Meta studies cited in Shift 3 were commissioned by Meta and are directional.

Saarthi and Sahayak are described in coverage of ONDC’s own announcements. We have not tested either.

ONDC’s consultation paper is from September 2022 and sets out proposals for comment.

Sources

[1]Dittrich, Jørgensen, Prakash, Rafnsson and Hinrichsen, Beyond Platforms: Growing Distributed Transaction Networks for Digital Commerce, Information and Software Technology (in press)Checked; a case study of Beckn and ONDC based on interviews
[2]ONDC, Building Trust in the ONDC Network: Consultation Paper (29 September 2022), section 4.1Checked; proposals for comment, not adopted policy
[3]CXOToday, ONDC introduces Saarthi, a multilingual reference application powered by Bhashini (11 September 2024)Checked; coverage of an announcement
[4]Elets eGov, ONDC launches Sahayak WhatsApp bot to support sellers in Indian languages (23 July 2025)Checked; coverage of an announcement
[5]Razorpay, Razorpay, NPCI and OpenAI come together to launch agentic payments (9 October 2025)Checked; a company announcement of a pilot
[6]Google Cloud, Announcing Agent Payments Protocol (AP2) (16 September 2025)Checked; a company announcement
[7]Sahamati, What is an Account Aggregator?Checked
[8]SCC Online, Digital Personal Data Protection Rules, 2025: key highlights (26 December 2025)Checked via a legal summary; the Rules’ own text not read
[9]Cross-Border Magazine, ONDC passes 500 million transactions as India’s open commerce network scales (7 August 2026)Checked; a trade report attributing the figures to DPIIT, not confirmed with DPIIT

Spotted a source we missed or a rule we misread? Tell us what we got wrong.

Looking for a pilot partner

We would most like to test shifts 1, 3 and 5 with a real partner. If one of them fits a pilot you could host, fund or take part in, we would like to hear from you.

Shift 1 Shift 3 Shift 5

Begin a private conversation

The network should be open. The discovery should be intelligent. The experience should be human. The trust should be portable. The market should remain contestable.

Build with us

How we work

Studio

More

In confidence