Automated price management on marketplaces: an AI repricer

Three classes of "automated pricing" — and why the results differ

Very different things are sold under the label "marketplace repricer". The difference decides the main question: will the price defend your margin, or just chase a competitor?

1. A classic repricer. It tracks a competitor's price and repeats it with an offset: "stay 2% below SKU X". Fast, but blind: it does not know your cost price, the category commission, or your tax. If the competitor starts liquidating stock below their own cost, the repricer will faithfully drag you into the red after them.
2. Seller-account tools: promo calendars and minimum prices. Wildberries and Ozon propose promotions themselves and can add products to them automatically. In the account you can set a minimum price or leave a promotion manually — but deciding "profitable or not" for every SKU, every day, remains your job.
3. An AI agent. It computes the price in reverse — from the money that must stay with you: retail price = fixed costs / (1 − commission% − acquiring% − tax% − target margin%). For every product it keeps a price band: a floor below which the sale is a loss, a target price with the desired margin, and a ceiling above which sales drop. It watches competitors too, but through common-sense filters, and accompanies every price change with an explanation: what moved and why.

From here on, "agent" means only the third class. The first two are tools that still need a human at the wheel.

What a classic repricer cannot do

Compute from unit economics. The formula "competitor price minus offset" contains no cost price. The agent builds the full cost map first: purchase cost, packaging, marketplace commission, delivery logistics, return logistics weighted by the buyout rate, storage, acquiring, tax. Only then does it look at competitors — inside the floor-to-ceiling band. If the competitive price is below the floor, the agent stops at the floor, not in the red. How the base price itself is derived is covered in our article on AI pricing.

Filter competitors. A blind repricer reacts to any price in its watchlist. The agent screens out the ones that are dangerous to follow: sellers whose stock is below a threshold (they will soon disappear from search), whose price is anomalously far below the median (a likely liquidation), or who have had no sales for a while.

Defend against WB auto-promos. Wildberries can enroll a product into a promotion automatically — and if nobody reacts, the price slides down without you. The agent polls the promo list over the API on a schedule, recomputes the margin at the promo price, and pulls the product out of an unprofitable promotion. There is also preemptive protection: if a promotion is announced in advance, the "before-discount" price is raised so that the promo price never breaks the floor.

Tell SPP apart from your own discount. SPP — the loyalty discount on Wildberries — is paid by WB, not by the seller, and should not be baked into your margin math. A blind repricer sees a rival's storefront price with SPP applied and "answers" it with a real discount of yours — a classic mistake that eats margin.

Understand Ozon points. On Ozon the platform may discount a product itself and compensate the seller with points. Points are not cash: they can only be spent on promotion within the platform. A blind Ozon repricer never sees that difference; the agent computes margin without the points and suggests leaving a promo when the discount is large and the real margin falls below your minimum.

Price from inventory. If a week of stock is left and the next supply arrives in a month, cutting the price is the worst move: stock runs out, the listing goes out of stock, and recovering search positions takes a week or two. The agent compares the sales velocity with the supply date: low stock — the price goes up and sales slow down until the delivery; excess stock — the price eases down so you stop paying for storage.

Setup in plain language, not development

A classic repricer makes you assemble rules in a builder: thresholds, offsets, competitor lists — per product. With Samreshuuu you explain the strategy the way you would to an employee:

From then on the agent works on a schedule: it checks rates and competitor prices, recomputes the bands, and either changes the price itself or sends you a draft for approval — the mode is your choice. Automated pricing turns from a black box into a policy you dictated in words.

How to set up automated price management: step by step

1. Connect the marketplaces over the official APIs. Wildberries — with an API token from the seller account, Ozon — with a Client ID and a Seller API key. The walkthroughs are in how to connect AI to Wildberries and how to connect AI to Ozon. Your account login and password are never handed over, and access can be revoked.
2. Provide the unit economics. Cost price (purchase plus delivery to the marketplace warehouse), packaging, tax regime, target margin, and the buyout rate for your category. Commissions and logistics rates the agent fetches fresh — they change, which is one more reason the price must be recomputed regularly.
3. Describe the strategy in words. The repricing goal: hold the margin, undercut competitors, defend against promotions, or manage inventory. For each goal the agent builds its own logic — from a simple "hold the margin" to a schedule around a big sale.
4. Pick the control mode. Start with draft-then-approve: the agent computes and proposes, you confirm. Once a rule has proven itself, move the routine to full automation and keep only sensitive steps — like leaving a promotion — on approval.
5. Turn on scheduled checks. Competitor and promo monitoring runs on a schedule; during hot sale periods it is worth checking more often, down to every 30–60 minutes.

Comparison: classic repricer, seller account, AI agent

The median and the "discount spiral": where promo money is lost

The promo discount is computed not from your current price but from the median sale price over the recent weeks. That is where the main trap hides: you join a promotion at a low price → the median drops → the next promotion demands an even lower price → the median drops further. This is the discount spiral, and climbing out of it is hard.

The agent manages the median in advance: several weeks before a major sale it raises the price so the median has time to grow and the promo price stays above the floor; after the sale it returns the price gradually, over a few days, not in one jump. Before every promotion there is an economic filter: minimum promo price = cost per unit / (1 − target margin). Anything below that is a loss, and the agent will say so plainly. Sales and promo analytics are best viewed together with marketplace analytics.

Honest about the downsides

If you have a handful of SKUs and change prices once a month, an agent is overkill: a spreadsheet and a unit-economics calculator will do. If your only task is "stay 10 rubles below the competitor" with a comfortably large margin, a classic repricer copes too. A loss-making promo entry is sometimes justified — a new product with no reviews, stock about to incur storage penalties — but that is an owner's decision, not an algorithm's: the agent will price the trade-off, and accepting the loss is on you. And a caveat about data: repricing quality depends on a correct cost price — if you do not know yours, start there, not with automation.

Checklist: how to choose a repricer for marketplaces

1. Does the tool compute from your unit economics or only from a competitor's price? If the settings have no cost price, commission, or tax — it is a blind repricer.
2. Is there a price floor, and what happens at it? The right answer: the price stops at the floor instead of following the competitor further down.
3. Does it protect against WB auto-promos and Ozon promos? Ask directly: what happens when the platform itself enrolls a product into a promotion at a bad price.
4. Does it account for stock and supply dates? Cutting the price on low stock is a straight road to out-of-stock and lost positions.
5. Does it explain the changes? You should see not just "price updated" but why: which factor moved and how it affected the math.
6. Can you start in approval mode? Trust an automation with money only after it has shown correct drafts for a few weeks.

Frequently asked questions

What is the best repricer for Wildberries and Ozon? The best one is the tool that computes the price from your unit economics rather than merely copying a competitor. The practical test: it must know your cost price, commission, logistics, and tax, hold a price floor, defend against WB auto-promos, and tell Ozon points apart from real money. Classic repricers only cover competitor tracking; an AI agent of the Samreshuuu class covers the whole loop — from computing the band to explaining every change — and is configured in ordinary language, with no rule builder.

How does automated price management on marketplaces work? The scheme is always the same: the tool regularly checks the inputs (competitor prices, rates, promotions, stock), recomputes the price under a given strategy, and applies it through the marketplace API. The difference is the strategy. A blind repricer knows one rule — "competitor minus offset". An AI agent derives the price from the target margin with the reverse formula, keeps a price band, manages the median ahead of promotions, and can slow down or speed up sales with the price depending on stock. Changes are applied either automatically or in draft-then-approve mode.

Is there an automated pricing service for Ozon? Yes, and when choosing one, mind two Ozon specifics. First: the commission depends on the price bracket, so the calculation is iterative — change the price, land in another bracket, the commission changes, recompute. Second: Ozon promo points are not cash on your account but a promotion currency, and they must not be counted into margin. Samreshuuu connects to Ozon over the Seller API, prices with commission brackets in mind, and proposes leaving a promo when the real margin without points falls below your minimum.

Is it safe to let AI change prices? Yes, under two conditions. First — hard limits: a price floor the agent will never cross no matter what competitors do, and a ceiling above. Second — a control mode: start with draft-then-approve, where every change arrives for your confirmation with the math explained, and automate only the rules that have proven themselves. The connection uses official APIs with revocable access; your account login and password are never shared.

What should I do about Wildberries auto-promos? Monitor and calculate rather than ignore. WB can add a product to a promotion on its own; the way out is regular polling of the promo list over the API, margin math at the promo price, and automatic exit from promotions where the margin is below your minimum. When a promotion is known in advance, preemptive protection works: the before-discount price is raised early so the discounted price never breaks the floor. And remember the median: aggressive promos today worsen your promo terms tomorrow.

---

Last updated: July 2026.

Sources: Wildberries seller API documentation (prices and discounts, promo calendar, commission and logistics rates); Ozon Seller API documentation (price management, promotions and promo points, the commission matrix by category and price bracket); WB and Ozon seller-account help on promotions, SPP, and minimum prices; the unit-economics and reverse-price-formula methodology from Samreshuuu skills.