Finding the product, setting the price, writing the listing, chasing stock, fixing prices when the exchange rate moves… Commbot does all of it for you. But its real job is this: it never lets a sale happen that won't make a profit — it builds the floor price from commission, shipping and the return allowance, runs 24 separate checks on its own work every 15 minutes, and repairs whatever breaks by itself.
Sign in with Google · set up in five minutes · no code, no install, no extensions
On the left, what you live through today; on the right, how Commbot solves it.
“I don't know which product will actually sell.”
→Engine More than 100,000 products are scanned, the buyer's price is compared with your cost, and only the ones that will turn a profit are put in front of you.
“The product sold out at the source and the order got cancelled.”
→Stock is checked every ten minutes — the listing of a sold-out product deactivates itself and goes back on sale when stock returns.
“I priced it wrong and lost money on a sale I thought was profitable.”
→The price is built with commission, shipping and the return allowance already subtracted. A loss-making price never happens, and it never climbs above your competitors either.
“The exchange rate moved and my prices are stuck in the past.”
→When the exchange rate, the source price or a competitor's price changes, your prices are refreshed every day , on their own.
“Writing listings, translating them and preparing images eats my whole day.”
→Titles, descriptions and tags in the buyer's language are written automatically; and from the product photos vertical video is generated and added to the listing.
“I notice orders too late.”
→The moment an order lands, a notification arrives: which size to buy, the link at the source and the net profit that sale puts in your pocket.
“They put the product in the cart but don't buy it.”
→A product left in the cart unbought gets a time-limited discount ; the marketplace itself notifies the interested buyer when the price drops. The discount ends as soon as the order comes in.
“My listing looks flat while competitors have a discount badge.”
→Every listing gets a strikethrough reference price , so the discount badge appears on it. The price you sell at doesn't change, your profit stays the same — the only thing that changes is visibility.
Six jobs, in order. Every one of them is really running today.
It scans the catalog and compares every product with its price on the marketplace. A product can only become a candidate if its selling price is at least two and a half times its cost. The ones that pass are ranked by demand, competitor count and profit share.
A product whose profit share is below 20% is not listed. Nor is one that doesn't have at least two sizes in stock at the source — putting a product with one size left on sale means accepting the cancellation up front.
The title, description and tags in the buyer's language are written by AI; a second model checks them so that no feature the product doesn't have ends up in the text. From the photos, vertical video is generated and added to the listing.
Commission, shipping by weight and the return allowance are subtracted. The price never rises above the competitors' average and never falls below the profit floor you set. In the end it is rounded the way buyers are used to: 1543 → 1539.
When a product runs out at the source, its listing is deactivated, and goes back on sale when stock returns. Anything that disappears from the catalog entirely is moved to the archive. Listings are never deleted — every step can be undone.
A notification arrives the moment an order lands. When a listing isn't selling, it looks for the problem: if it isn't being seen, it changes the tags; if it's clicked but not bought, the description; if it reaches the cart but never the order, the price. It measures for seven days and rolls back any change that didn't work.
The engine builds every price on this calculation. When the exchange rate, the cost or a competitor's price changes, the calculation refreshes itself the next day.
| Item | Where it comes from | Impact |
|---|---|---|
| Source costThe current discounted price on the source site | catalog · daily | 1.299,90 ₺ |
| Currency conversionInto the currency the buyer sees | daily rate | → ₽ |
| Marketplace commissionCategory commission, from store settings | setting | − %7 |
| ShippingCheap band if under 400 g and $40, otherwise standard | weight + price | − 63 ₽ / 280 ₽ |
| Return allowanceHow the category's return rate feeds into the cost | setting | − share |
| Lower limit — minimum marginIt never goes below it; if it would, the product isn't listed at all | setting | 20% floor |
| Upper limit — competitor medianNever priced above the median of those selling the same product | competitor scan | cap |
| Psychological roundingIn the currency the buyer sees | — | 1543 → 1539 |
| Selling price | daily, from scratch | 4 490 ₽ |
One switch stops the engine completely. The switches below it decide which jobs the engine may do on its own. Any job you turn off detected but not applied: you see it in the dashboard, it never reaches your store.
The whole automation of this store. Turn it off and your listings not deleted — only the automated actions stop, the listings stay exactly as they are.
Deactivates whatever sells out at the source and puts it back on sale when it returns. Turning this off is not recommended: a sold-out product left on sale brings cancellations.
Updates the price when cost, the exchange rate or a competitor's price changes. While it is off, prices are frozen — if your cost rises, you can end up selling at a loss.
Builds a vertical video from the product photos for new listings. Turning it off is safe; only the richness of the listing drops.
Applies a time-limited discount to a product that is added to the cart but not bought. When you turn it off, no new discount is started; the ones already running end the normal way, when they expire or an order comes in.
Shows a discount badge on the listing. You set the percentage yourself; it does not touch the price you sell at or your profit, it only writes the reference price.
While it is off, the engine uploads no product on its own, it only offers a list of opportunities. Turn it on and, out of the candidates that clear the gates, it uploads as many as your daily cap allows — you set the cap yourself.
If a bot writes the wrong price, nobody notices — and at the end of the month the books don't add up. This is where Commbot is different: there is a second layer that audits its own work. Every 15 minutes 24 separate invariants are checked and anything broken is found; most of them are repaired on their own — the rest light up red in the dashboard.
Every listing has a floor price of its own: the lowest profitable price left once source cost, marketplace commission, shipping, the return allowance and sourcing expenses are deducted. The engine can never go below that floor, by any route — not a promotion, not an automatic discount, not the price brake. If a competitor is cheaper, the listing simply won't sell at that price: it would rather not sell than sell at a loss.
Size 38 and size 46 of the same product are priced differently on the market. The engine does not flatten the family to a single price: each size starts from its own current price and moves toward the target, and whatever is already close to the target is not touched at all. That way one cheap size doesn't drag the whole listing down.
What sells out at the source is deactivated, what comes back is put on sale again. But a single reading is not enough: if a mass shutdown is detected, everything is read again over a fresh connection; if the reading was wrong, not a single listing is deactivated. The same holds in reverse: a listing sitting at zero stock on the marketplace while it is still in stock at the source is found and put back on sale.
A single matching error could deactivate hundreds of listings at once. The engine does not deactivate en masse the listings it cannot verify — it applies only as many as the daily cap allows, carries the rest over to the next day and writes it to the dashboard. You can put deactivations entirely under manual approval if you want.
Listings the engine does not delete, orders it does not cancel, and on your behalf it does not shop. Everything it does is a deactivation, an archiving or a price write — all three can be undone, and every one of them is written to the history with a timestamp.
If the watchdog has not taken a measurement for more than an hour, it reports that as a failure of its own. Not realising that a dashboard screen reading "everything is fine" is in fact coming from a check that has gone silent is the most expensive mistake of all.
Several stores are managed from the same dashboard, and every store works by its own rules with its own margin, its own currency, its own daily cap, its own permissions.
A client's store is untouchable. In a store on stock mode the engine cannot write prices, cannot deactivate listings, cannot delete products — these are closed off structurally, not something a wrongly flipped flag could open. Everything the engine finds is shown in the dashboard; to act on it, the store owner's approval is required.
A standard user changes their own commercial settings: profit margin, daily upload count, the priority and unwanted product lists. The engine switches, the deactivation rules and the store mode they cannot touch; if they need it, they ask the admin for permission from the dashboard.
A product uploaded in one store is never uploaded again in another: stores don't cannibalise each other's listings.
Nothing the engine does is specific to a single marketplace: catalogue, stock, pricing and listing preparation are the same everywhere. The marketplace is just a connector plugged in on top.
The marketplace connected right now is Ozon: the integration is live, listings are up, orders are coming in. When a new marketplace is added, everything underneath stays the same — your catalogue, your pricing rules and everything the engine has learned until then come with you.
So opening up a new market is not a setup from scratch; it means plugging one more connector into the same engine.
Five minutes on average. No code, no installation, no browser extension.
No separate password. The same account can hold more than one store — each store has its own engine settings.
You create a key in the seller dashboard and copy two values: Client-Id and Api-Key. The Api-Key is shown only at the moment it is generated; if you lose it, you generate a new one.
Ozon Seller → Settings → Seller API → Create keyYou paste the key and Verify & Addpress. The key is tested live at that moment: if it is invalid, it is not saved and you see the reason on screen. If it is valid, the store goes active.
You choose which jobs to hand over and set the daily cap for automatic listing. You can remove the store at any moment — your listings on the marketplace are not affected.
The dashboard splits into three jobs: running the store, hunting new opportunities, auditing that the data is right.
The goal is an engine that learns from its own results: an engine that knows what sells knows better what to list. Below we separate what is live today from what is on the way.
Opportunity scanning, profit and stock gates, listing copy + video, daily pricing, ten-minute stock sync, order alerts, dosed automatic listing.
Content experiments: diagnosis → change → seven days of measurement → rolling back the loser.
Buyer-side search demand now feeds the selection score: how many people search each keyword, what the add-to-cart rate is on that search — product selection now runs on this data.
Sales → selection feedback: the engine learns from its own funnel data. Orders, add-to-carts and views of the last 90 days are weighted into the candidate score — a product class nobody carts drops off by itself, one that lands rises.
Loss-making sale protection and self-healing audit: 24 invariants, every 15 minutes; a price that drops below the floor, stock that was never written and a lost strikethrough price are repaired on their own.
New marketplace integrations and a pool of winning content templates: a title pattern that works on one listing becomes the default on similar ones.
We don't turn the engine on day one. We connect your store and the engine without touching anything scans your catalog and shows you this: which products make no profit, which listings are priced wrong, how many products sit on sale with no stock. Once you have seen the numbers, you decide whether to turn it on.
About half an hour. Screen share, on your own store's data.
We'll get back to you shortly on the channel you gave. In the meantime you can already connect your store — nothing runs automatically until you turn the engine on.
We watched what sellers on the Russian market actually say and wrote down as much as we could verify. Every post lists the videos it rests on.
From account setup to customs, commission and the payout floor — what Turkish sellers and the Ozon Global Türkiye team say, contradictions included.
From seller videos and Ozon's own webinars: why the first reviews decide a card, how points-for-reviews works, and how sellers answer negatives.
Legal form, tax regime, the VAT threshold and the marketplace report, as sellers describe them. Not tax advice; rates and thresholds change.
All it takes is your API key from the seller dashboard. An invalid key is not saved, and nothing runs automatically until you turn the engine on — on day one you just watch what it could do.