Selling on ONDC in 2026: The Honest Maths Before You List (Fees, Real Order Volumes, and Who It Works For)

A client who runs four sweet shops called me in September with a number he had heard at a trade event. Three per cent. That was what ONDC would cost him, against the thirty per cent a food delivery app was taking. He had already done the maths on a napkin and it said he was losing about four lakh rupees a year to commission.

The three per cent is real. The four lakh is not, and the gap between those two facts is where most ONDC decisions go wrong in India right now.

ONDC charges no commission on transactions. That part is true and it is written into how the network works. But ONDC is not who you sell through. You sell through a seller app, that app sets its own fee, a buyer app sends you the order and takes its own cut of attention, somebody has to deliver the parcel, and returns land on you. By the time a six hundred rupee order settles in your bank account, the three per cent has company.

This is the post I wish that client had read before the trade event. What ONDC actually is, in plain words. What the real bill looks like, line by line, on one order. The order volumes, including the ones that went down. The three kinds of business it genuinely works for, and the ones that will list, wait, and quietly give up after five months. And if you want to build your own integration rather than rent somebody's, what that really takes.

Save this one. There is a go or no go worksheet near the end that you can copy into a notepad and fill in with your own numbers in about fifteen minutes, and it will tell you more than any vendor demo will.

What ONDC actually is, without the jargon

The single most common misunderstanding: people think ONDC is an app they can download, or a website where buyers come to shop. It is neither. There is no ONDC storefront. No customer has ever opened ONDC and browsed it.

ONDC is a set of rules. It is an agreement about the shape of the messages that shopping software sends to other shopping software. That is genuinely all it is.

Here is the version that makes sense to anyone who has run a shop. Imagine every delivery app in India agreed on one standard order slip. Same boxes, same order, same language. Once that slip exists, any app that takes orders can hand a slip to any shop that fills them, even if the two have never met and have no contract with each other. Nobody owns the slip. That is ONDC.

Which means there are three roles, and you need to know which one you are dealing with every time somebody pitches you.

The buyer app, where the customer actually is

This is the software the shopper uses. Paytm, magicpin, a bank's app, a shopping app you have never heard of. The customer installs that, searches in that, pays in that, and complains to that. In network language this is the BAP, the buyer app. If demand exists anywhere in this system, it exists here.

The seller app, where you actually are

You do not join ONDC directly. You cannot. You sign up with a seller app, which in network language is the BPP, and that app puts your catalogue on the network on your behalf. It answers search requests, confirms orders, reports status. It is your ONDC connection, your dashboard, and the company whose fee you will actually pay. Mystore, Vikra from Zoho, SellerApp and a long tail of others sit here.

The gateway, which you will never think about again

When a buyer app broadcasts a search, something has to fan it out to the seller apps that might have the item. That is the gateway. It matters enormously to the engineers and not at all to you. Mentioned only so that when a slide deck shows you three boxes, you know which box your money comes out of.

“You are not listing on ONDC. You are listing on one seller app, which speaks ONDC on your behalf. Every cost question is really a question about that app.”

The honest numbers, including the ones that went down

I want to be careful here, because ONDC figures are quoted loosely and the loose quoting is usually in the direction of enthusiasm. Here is what the public record supports, with the disagreements left in rather than smoothed over.

On sellers, you will see wildly different counts. One widely repeated figure is more than seven lakh sellers onboarded across twelve hundred cities. A Press Information Bureau release dated 9 March 2026 put it differently: roughly two lakh six thousand merchants had completed at least one transaction as of 31 December 2025. Another account gives about one lakh sixteen thousand active retail sellers across six hundred and thirty cities by December 2025.

Those numbers are not contradicting each other so much as measuring different things. Onboarded means a catalogue exists somewhere. Transacted means money moved at least once. The ratio between the two is the real story, and it is not flattering: on those figures, a large share of sellers who joined have never completed a single order.

On transactions, the network crossed five hundred million cumulative transactions in July 2026, and reported around twelve million orders in a single month in one 2026 release. Growth in the direct to consumer segment has been reported at roughly thirty eight per cent month on month since October 2025.

And now the part that vendor blogs skip. Retail order volume has gone backwards before. Business Standard reported retail transactions peaking at about sixty five lakh orders in October 2024 and falling to about forty six lakh by February 2025, as ONDC capped the financial incentives it had been paying network participants. Incentives were cut by as much as seventy five per cent from one quarter. Orders followed the subsidy down.

That single fact should change how you read every growth chart you are shown. A meaningful part of early ONDC volume was bought. When the discount went, so did a chunk of the demand. Any projection somebody hands you that was drawn in the subsidy era is describing a market that no longer exists in that form.

One more distribution fact worth more than all the totals: hyperlocal food and grocery has accounted for the overwhelming majority of orders, in some periods reported at ninety seven to ninety eight per cent. In the food category the network has been reported at around three per cent of what Swiggy and Zomato do. Three per cent of those two is a real business. It is also not a replacement for them.

The real bill on one order, line by line

Forget percentages for a minute. Take one order and walk it to your bank account. I am using a six hundred rupee order because that is close to the average for a lot of Indian retail and food sellers, and because percentage talk hides how badly fixed costs hurt small baskets.

ONDC itself takes nothing. There is no network commission on the transaction. Hold on to that, because it is the one part of the pitch that survives contact with reality.

Then everybody else arrives.

• Seller app fee. This is your main cost and it varies enormously. Published rates run from about one per cent of order value up to five per cent and beyond, and monthly plans run from zero to the far end of twenty two thousand four hundred and ninety nine rupees a month. Zoho's Vikra, as one concrete example, has published one per cent per order with no monthly fee, plus eighteen per cent GST on that fee.
• GST on the fee itself. Eighteen per cent on top of whatever the seller app charges. People forget this one constantly. A one per cent fee is really one point one eight per cent.
• Payment gateway. Roughly one to two per cent, the same as it would be anywhere else. UPI is cheaper, cards and wallets are not, and we broke the real rates down in how to actually accept payments in India.
• Delivery. Either you do it, which costs you time and fuel, or the network's logistics partner does it, which costs money and on a six hundred rupee basket can easily be the largest single line on this list.
• Packing. Small, real, and never in anybody's slide.
• Returns and cancellations. Not a percentage, a probability. Budget for it as a share of revenue, because it will behave like one.

Worked through on that six hundred rupee order with a two per cent seller app, UPI payment and a shared delivery cost, you land somewhere around five hundred and ten to five hundred and forty rupees actually reaching you, before your own cost of goods. You can put your own figures through the ONDC fee calculator and watch which line does the damage. Call it ten to fifteen per cent of all in friction, not three.

Now put that next to the thing you are comparing it with. A marketplace charging twenty five to thirty per cent on the same order leaves you about four hundred and twenty to four hundred and fifty rupees. So ONDC does win on this line. It wins by roughly sixty to a hundred rupees on a six hundred rupee order, not by a hundred and eighty.

One source puts the all in seller side cost at three to ten per cent against eighteen to forty on the big marketplaces, which is consistent with what I just walked through. Another, less cheerful one, reports some small sellers losing roughly twenty five to thirty five per cent of earnings once every deduction is counted. Both can be true of different sellers in different categories, and which one you turn out to be depends almost entirely on your average order value and who carries the parcel.

“The commission saving is real. It is a margin improvement on orders you were already getting, not a machine that produces new orders.”

The thing nobody tells you: ONDC is plumbing, not footfall

This is the sentence I end up saying on every one of these calls, so here it is in writing.

Being listed on ONDC is like having your shop on a road. The road is new, well built, publicly owned and goes to a lot of places. It does not bring anybody down it. The buyer apps have the customers, and a buyer app will show the customer whatever ranks well, converts well and does not get cancelled. You are competing inside those apps on exactly the things you compete on everywhere else: price, availability, ratings, delivery promise, how clean your catalogue is.

So the honest framing is this. ONDC lowers the toll you pay on each order. It does not generate demand for you, and nothing about joining it changes whether people want what you sell. A seller with no demand problem and a commission problem will be very happy on ONDC. A seller whose actual problem is that not enough people know they exist will list, see a handful of orders, and conclude the network is broken. The network is not broken. It was answering a different question.

If the real problem is that nobody can find you, the lever is your own storefront, your own search presence and your own repeat customers, which is a different piece of work entirely. Taking orders on WhatsApp is usually a faster first win than a new network. We have written up the beginner version of that in our guide on how to start selling online in India, and the reason those two posts are separate is that these are genuinely separate problems.

The three kinds of business it genuinely works for

From the volume distribution, the fee structure and the sellers I have actually watched do this, three profiles do well.

1. Hyperlocal food and grocery with your own delivery

This is where the orders actually are, by a very large margin, and it is where ONDC's economics are at their best. If you already have a boy on a bike, you have removed the single biggest cost line on a small basket, and the commission saving against a food delivery app goes almost straight to your margin. My sweet shop client is in exactly the right category. We have built this shape of operation end to end before, in an online grocery system with a customer app and a driver app. His error was the size of the prize, not the direction.

2. Higher value baskets where fixed costs stop mattering

Delivery and packing are close to fixed per order. At six hundred rupees they are brutal. At three thousand five hundred rupees they are a rounding error, and the percentage saving on commission turns into real money. If your average order value is above roughly fifteen hundred rupees, the maths stops being marginal.

3. Brands that already keep a disciplined catalogue

If your stock counts are accurate, your prices are current and your descriptions are clean, you can plug into another channel cheaply, because the hard part is already done. If your stock lives in a notebook, read the next section before you do anything at all.

And the mirror image, who should not bother this year. Low margin retail commodity goods where you cannot absorb a return. Anyone expecting the network to supply customers. Anyone without real time stock accuracy. Anyone whose category is barely transacting on the network, which you can check before you commit by asking a prospective seller app for order counts in your own category rather than network totals.

The trap that quietly kills seller accounts

Of all the operational problems reported on the network, one shows up again and again, in Business Standard's reporting and in the independent analyses both: returns and rejections ran high, and a lot of it traced back to sellers not tracking stock in real time. Orders were placed and paid for, then cancelled or altered because the item was not actually there.

Understand what that does to you, because it is worse than the lost order. You are not on a platform where one bad week is absorbed by your history. You are on a network where buyer apps decide what to show, and a seller who cancels gets shown less. The punishment for bad stock data is not a complaint, it is silence, and you will read the silence as ONDC not working.

So the prerequisite is not a seller app. It is one honest answer to one question: if somebody buys your tenth unit of something right now, does every place that could sell it know within a minute? If the answer involves anybody checking anything by hand, fix that before you list. In practice that means your stock lives in one system and everything else reads from it, whether that system is Tally, Zoho or your own. We have written separately about wiring Tally into a website and the same principle applies here, with higher stakes, because now a stranger's app is quoting your availability to a paying customer.

Returns and disputes themselves run through the network's Issue and Grievance Management framework, so there is a defined process. A process for handling the dispute is not the same as not having the dispute.

How to actually join, both routes

Route one: through a seller app. This is almost certainly you

Ninety five per cent of sellers should take this route, including every business that is not primarily a software company. You are renting somebody's integration instead of building one, which is the correct call when the integration is not your product.

1. Get your paperwork in order. GSTIN, which is mandatory for e-commerce sellers under section 24 of the CGST Act and not optional the way it is for a small offline shop. If you will be raising e-invoices at volume, our GST e-invoicing guide covers the IRN and IRP side. PAN. A current account for settlements. Aadhaar of the authorised signatory for KYC.
2. Shortlist seller apps and ask each one three questions in writing: the exact fee including GST, whether they are currently accepting new sellers in your category, and how many orders sellers in your category and your city did last month. The third question is the one that tells you something. Vague answers are an answer.
3. Check they are actually open. Onboarding availability moves around. There were reports in mid 2026 of at least one well known seller app pausing new seller onboarding entirely. Do not build a plan on an app that is not taking you.
4. Complete KYC and bank verification, then list your catalogue properly. Real titles, real descriptions, accurate prices, correct categories, honest stock counts.
5. Connect stock. If your seller app can read from your existing system, do that rather than maintaining two numbers. Two numbers become two different numbers.
6. Start with a narrow range and your strongest city, watch cancellation rate more closely than revenue for the first month, then widen.

One warning worth repeating because it costs people money every month. A genuine seller app does not charge a fee to get you approved, to verify you faster or to boost your visibility on the network. There is no such product. Anybody selling it is selling you nothing.

Route two: becoming your own seller network participant

This is the real engineering path, and it is the right one for a narrow set of businesses: marketplaces, aggregators with many merchants underneath them, large brands that want to own the channel, and software companies intending to offer seller app services to others. If that is you, the settlement side is its own project, and we covered it in multi-vendor billing software in India. If you are a single business with one catalogue, this is almost always the wrong spend.

What it actually involves, honestly. You build an application that speaks the Beckn protocol correctly as a BPP. You generate your own signing and encryption key pairs and self signed certificates, noting that buyer node and seller node need separate certificates. You register in the network registry with your subscriber ID, domain and node type. Onboarding is not fully self serve, and getting into the process has involved writing to the ONDC team directly. You then go through certification before you are live on production.

The protocol surface is not enormous but every piece of it has to be right. The transaction actions are search, select, init, confirm, status, track, cancel, update, rating and support. Every message is signed with your private key and verified on the other side. Payloads validate against the published OpenAPI specifications, and a payload that is merely nearly correct is rejected, which is the correct behaviour and also where most first attempts die.

The parts that take longer than people estimate are not the happy path. They are catalogue push and keeping it fresh, real time inventory, the cancellation and return flows, reconciling settlements against your own books, and idempotency so that a retried confirm does not become two orders. That last one is the same discipline as handling payment webhooks properly, and we wrote up how we do that for Razorpay because getting it wrong produces duplicate orders and angry customers in exactly the same way.

We build this kind of integration in NestJS with a Next.js dashboard on top, which is our default for anything where correctness of money and state matters more than novelty. A realistic first production BPP, scoped to one domain and a clean catalogue, is a multi month engagement rather than a sprint, and the honest cost band in India sits well above what a seller app subscription costs for several years. Which is precisely why route one exists and why most people should take it.

The go or no go worksheet

Copy this into a notepad and fill it in with your own numbers. If you would rather it did the arithmetic for you, our free ONDC fee calculator takes the same inputs and shows what lands in your account next to what your current channel leaves you. If you cannot fill in a line, that is the finding. Most people stop at the inventory question, and stopping there has saved them a wasted quarter.

textONDC GO / NO-GO WORKSHEET

A. WHY AM I DOING THIS
[ ] My problem is commission on orders I already get   -> ONDC can help
[ ] My problem is that not enough people find me       -> ONDC will NOT help
    (If you ticked the second one, stop here and fix demand first.)

B. THE NUMBERS I MUST KNOW
   My average order value                    Rs ________
   My gross margin on that order             ________ %
   Commission I pay today, if any            ________ %
   Who delivers today                        self / partner / n.a.
   My cost to deliver one order              Rs ________
   My return + cancellation rate             ________ %

C. THE BILL, ON ONE AVERAGE ORDER
   Order value                               Rs ________
   - Seller app fee        ____% =           Rs ________
   - GST on that fee       18%   =           Rs ________
   - Payment gateway       ____% =           Rs ________
   - Delivery                                Rs ________
   - Packing                                 Rs ________
   = Lands in my account                     Rs ________
   Same order on my current channel          Rs ________
   MONTHLY DIFFERENCE  (gap x my orders)     Rs ________

   If the monthly difference is under Rs 5,000, the admin
   work will eat the gain. Revisit when volume is higher.

D. THE INVENTORY GATE   (all three, or do not list)
[ ] Stock counts live in ONE system, not a notebook
[ ] If my last unit sells, every channel knows in under a minute
[ ] Nobody updates a second stock number by hand

E. THE CATEGORY GATE
[ ] I asked a seller app how many orders sellers in MY
    category and MY city did last month.  Answer: ________
[ ] The answer was a number, not a network total
[ ] That seller app is accepting new sellers right now

F. PAPERWORK
[ ] GSTIN   [ ] PAN   [ ] Current account   [ ] Aadhaar of signatory

G. DECISION
[ ] GO, via seller app: ____________________
[ ] GO, build own BPP  (only if I am an aggregator/marketplace)
[ ] NOT YET. Revisit on ____________  because ____________

What I would tell you on a call

If you run food or grocery, deliver it yourself, and you are currently paying a quarter or more of every order to somebody else, list this quarter. Pick the cheapest credible seller app, start with one city, and judge it on your cancellation rate first and your revenue second. The saving is real and in your category the orders exist.

If your average order value is above about fifteen hundred rupees and your catalogue is already in decent shape, list. It is a cheap extra channel and the fixed costs that ruin the maths for small baskets do not reach you.

If you are a marketplace or an aggregator with merchants underneath you, the integration conversation is a real one and route two is worth costing properly.

Everyone else, and that is a lot of people, should do the worksheet above and most likely wait. Not because ONDC is bad, but because listing on a new channel is the wrong fix for a demand problem, and the subsidy era numbers that made it look like a gold rush are not the numbers you would be joining today.

The sweet shop client, for the record, listed. He is saving a little under two lakh a year rather than four, his cancellations went to almost nothing once his counter staff stopped keeping a separate stock pad, and he is pleased. Half of the number he was promised, arrived at honestly, turned out to be a good outcome.

If you want a second opinion on your own numbers, put them in the comments or send them over. Tell me your category, your average order value and who delivers, and I will tell you straight whether it is worth your quarter, and I will credit you if I use the example in a follow up.