Selling on marketplaces: what it takes to connect eMAG or Amazon to your stock

A marketplace does not add a sales channel. It adds a second business that competes with your own shop for the same units in the same warehouse, and it does not know your shop exists.

Everything difficult about marketplace integration follows from that one fact. This article is about what it actually takes, based on delivered work. Timelines and costs are our own 2026 ranges.

The listing is not your product

The first surprise is always the same. In your system, a product is one record. On a marketplace, that record splits in two, and they behave differently:

1. The product
   • What the item is: title, description, images, category, brand, EAN.
   • Owned by the marketplace’s catalogue, often shared between every seller of the same item.
   • Changes slowly. May need approval.
2. The offer
   • What you sell it for: your price, your stock, your handling time, your warranty.
   • Yours alone, and the only thing you are free to change whenever you like.

Integrations that treat these as one thing fail on the first price change, because they try to push a whole product update through a catalogue approval queue in order to move a number by two lei. Model them separately from the start, and price and stock updates become a small, frequent call rather than a large, rejected one.

The four flows, and their tolerances

Like an ERP integration, “connect the marketplace” means four separate jobs with different rules:

Catalogue, out. Products and their attributes. Slow, batched, and the one most likely to be rejected for a missing field. Every marketplace has mandatory attributes per category, and they are not the same as yours.

Stock and price, out. The one that runs constantly. This is where money is lost, in both directions.

Orders, in. Must never lose a record, and must arrive with enough information to fulfil and invoice it.

Status, out. Confirmed, packed, shipped, tracking number, cancelled. Marketplaces measure you on how fast this flow moves, and your seller rating depends on it.

Where the money leaks

Overselling, because stock is shared. Your shop and the marketplace both sell the last unit within the same sync window. On a marketplace this is not just an apology — cancellations count against your seller metrics, and enough of them suppress your listings.

There are only two honest fixes, and both are business decisions, not technical ones: reserve the unit at checkout, or hold back a buffer per channel. Buffers are the pragmatic answer for most sellers — publish 8 of the 10 you hold, and the last two absorb the race.

Underselling, quietly. The mirror image, and the one nobody notices. A failed stock push leaves the marketplace showing zero while your warehouse is full. There is no error, no cancellation, no complaint — just an absence of orders. Alert on the absence: if a channel has sold nothing in X hours during business hours, something is wrong.

Commission, returns and fees. The payout you receive is not the sum of your orders. It is the orders minus commission, minus fulfilment fees, plus and minus returns that happened weeks ago. If your accounting expects the first number, someone reconciles the difference by hand every month, forever. Bring the settlement file into the integration and post it against the orders it belongs to.

What actually breaks, in order

Category attributes. Every category has mandatory fields you do not hold: energy class, package weight, warranty months, a brand code from their list. Discover this before quoting, not during. On a catalogue of any size, filling those gaps is a data project of its own.

Rate limits. Marketplace APIs are strict and they enforce it. A full catalogue push at 10am will get you throttled or blocked. Batch, back off on 429s, and schedule the heavy work when nobody is buying.

The same product, twice. You list an item the marketplace already has, and now there are two catalogue entries competing — sometimes yours against yours. Match on EAN before creating anything, and treat “create product” as the exception, not the default.

Returns. A return is not a cancelled order. Stock comes back, sometimes not in sellable condition, sometimes weeks later, sometimes to a different location. Model it explicitly or the stock figure drifts every month and nobody knows why.

What it costs and how long it takes

For a company with 10–200 employees, per marketplace:

• Offers only — stock and price on an existing catalogue, orders pulled in: 6,000–14,000 EUR, 3–5 weeks.
• Full catalogue — products created and maintained from your system, with attribute mapping: 12,000–28,000 EUR, 5–9 weeks.
• Two or more marketplaces plus your own shop, sharing one stock pool with buffers and reconciliation: from 25,000 EUR, delivered in stages.

The second marketplace is not a second full price. Roughly 40–60% of the first, because the mapping layer, the buffer logic and the error queue already exist — provided the first was built with a second in mind. Say so up front if a second is likely, because it changes how the first is structured.

Before you start, get these three answers

Who owns the price? If a marketing person changes prices in the marketplace’s own dashboard and the integration pushes over it every fifteen minutes, you will have that argument once a week. Decide which system is the source of truth and enforce it.

What is the buffer? Per channel, in units or percent. This is a commercial decision about how much stock you are willing to leave unsold in order to avoid cancellations. The developer cannot make it for you.

What happens on a partial fulfilment? Five lines ordered, three shipped. If your model assumes one order equals one shipment, discovering this after launch is a rewrite.

How to know it worked

Measure before you connect anything: hours a week spent updating listings by hand, cancellations for stock reasons, days between an order and its invoice.

• Realistic target: cancellations for stock under 1% of marketplace orders;
• Watch: the nightly reconciliation between the settlement file and your own order totals — the drift should trend to zero and stay there;
• Success signal: adding a third marketplace is a configuration task, not a project.