Why yesterday's minimum price is loss-making after the July WB and Ozon commission hikes, where competitor-driven pricing loses money, how to compute the real minimum price from unit economics, and what law 289-FZ brings on 1 October 2026.

In short. A repricer optimizes price relative to competitors. An AI agent optimizes it relative to your profit. These are different jobs, and almost the whole market solves the first one: of roughly fifteen repricers, about nine in ten run on a "−1 ₽ below the competitor" rule and don't know your cost price. While commissions held steady, that was enough. Since July 2026 it isn't: WB and Ozon rebuilt their fee tables, and for many SKUs yesterday's minimum price is already loss-making — but the repricer doesn't know that and keeps holding it. Below: exactly where a repricer loses money, how the real minimum price is calculated with today's commissions, and what an AI agent does differently.

What happened to unit economics in July 2026

Everyone has to recalculate prices at once right now — for two reasons.

Commissions rose on both platforms. On 7 July Wildberries raised its fees: roughly +5 p.p. on FBW, +6 p.p. on FBS and up to +20 p.p. on DBS, affecting more than 90% of products. From 1 July Ozon collapsed its price bands — for the mid price segment the effective commission rose by up to +12 p.p. A product that gave you margin yesterday may sell at zero or a loss today, with the price tag unchanged.

Control over price is returning to the seller. From 1 October 2026 law 289-FZ takes effect: the seller gets a legal right to set a minimum price and to consent to platform discounts. The right exists. But to use it, the minimum price must first be calculated correctly — not taken as "cost minus some percent."

Two things coincided: costs jumped, and the price lever is being handed back to you. A tool that moves price by the competitor and doesn't compute your profit works against you in this situation.

What a repricer actually is

A repricer automates a single action by a rule. You set a corridor (minimum and maximum) and a follow-the-competitor logic: "competitor dropped — you drop too, but not below the minimum." Then it moves the price inside the corridor.

Solution type	Mechanics	Knows your cost price	Computes profit at the new price
Platform's built-in repricer	follow competitor, corridor	no	no
Rule repricer (third-party)	"−N ₽ from competitor", corridor	no	no
ML repricer with elasticity	demand model, but the metric is turnover/rank	partly	partly
AI agent from unit economics	reverse-calculates price from target margin	yes	yes

Nearly every landing page promises "margin protection." But margin protection is only possible where the tool sees the cost price and all per-SKU costs. Confirmed ML with elasticity is effectively a rarity on the market (of the notable ones — Imprice, from 7,500 ₽/mo). Repricer prices range from a few hundred rubles to nearly twenty thousand a month, and price barely correlates with whether the tool computes your profit at all.

Four situations where a repricer loses money

None of the mass-market articles about repricers discuss the limits of their applicability — they all sell "plug it in and earn." Here is where a rule repricer steadily loses money.

The race to the bottom. Two looped repricers on two sellers start dropping the price after each other down to the corridor minimum. Both lose margin; only the buyer wins. The rule doesn't understand it's in a race — it just executes the condition.
The "profitable" loss. The corridor minimum is set without accounting for ad spend, storage and returns. Formally the price is "above cost," in fact after all deductions the deal is negative. The repricer faithfully holds a price that is guaranteed to lose money.
Dumping on low stock. The repricer drops the price to catch a competitor exactly when stock is nearly gone. You accelerate the out-of-stock, sell the last units cheap and lose search positions — recovering them costs more than the discount you gave up.
A product that needs anything but a discount. Sometimes the right move is not to lower the price but to raise it, or to pull the SKU. A repricer won't turn a loss-making product profitable: it has no notion of "this SKU is better dropped," only "catch the competitor."

In all four cases the problem is the same: the tool has no concept of your economics as a primary value. It optimizes position relative to the market, not profit relative to your costs.

How the real minimum price is calculated

A minimum price is not "cost plus some percent." It is a reverse calculation: from the target margin back to the price, through every platform deduction. Step by step, with today's fees:

SKU cost price — from inventory accounting (e.g. from MoySklad via API), not from memory.
Platform commission — at today's actual rate, accounting for the July increase and the scheme (FBW / FBS / DBS), not last year's number.
Acquiring and logistics — intake, delivery to the buyer, handling.
Storage — more expensive the slower the SKU turns over.
Returns — via the real buyout rate: a non-buyout is round-trip logistics at your expense.
Tax and ad spend — the tax rate and promotion costs as a share of price.

Only after gathering all of this can you compute the price below which the deal becomes loss-making — and the target price that yields the required margin. Two clarifications that this content cluster routinely confuses: the Ozon and WB price index affects search visibility and promo eligibility, not the commission size; and there is no formal "dumping penalty" or "penalty for skipping auto-promos" — the punishment is indirect, through a lower platform discount share and search deprioritization.

What an AI agent does differently

An AI agent starts not from the competitor but from your economics. In practice that means:

Cost price and stock come first. The agent pulls them from inventory accounting and computes full unit economics per SKU, instead of comparing price tags.
Price ↔ stock ↔ sales-speed link. It accounts for days-to-zero on stock: you can't dump something that's about to run out.
Recalculation on fee changes. When the platform changes a commission, the minimum price recomputes itself instead of staying yesterday's.
Explainability. "Why 743 ₽" has an answer by cost line item, not "the algorithm decided so."
Price changes only after confirmation. The routine recalculation runs automatically, but the price change itself goes for confirmation. This answers the main fear — "the algorithm will dump the price overnight" — which repricer vendors don't address.

The difference isn't that the agent "moves the price more cleverly." It solves a different problem entirely: it holds profit, not position.

Built-in repricers and the platforms' AI assistants

Ozon has a built-in repricer, and all three platforms launched AI assistants in 2026: Ozon since April on Qwen 3.5, Wildberries since May under the "Jem" subscription, Yandex Market on YandexGPT back since 2025. These are useful analytics and advisory tools. But none of them manages your prices from your profit — and by design cannot: your margin is not their metric. A platform optimizes its own turnover, not your unit economics. So "there's already an assistant in the dashboard" does not close the profit-based pricing job — these are different roles.

289-FZ: the right to a minimum price from 1 October 2026

From 1 October 2026 the seller gains a legal right to set a minimum price and to consent to discounts. This reverses the situation: previously the platform could drive the price down through its promos, now you define the floor. But the right to a minimum price is only as useful as that minimum price is accurately calculated. Set it without July's commissions, storage and returns, and the legal right will lock in a loss. So the law strengthens, rather than removes, the job of computing an honest minimum.

FAQ

What is AI pricing on marketplaces? It is managing price from your unit economics: the agent computes the full per-SKU cost of a deal (commission, logistics, storage, returns, tax, ad spend) and holds the price above break-even with the required margin, instead of merely following the competitor.

How is an AI agent different from a repricer? A repricer optimizes price relative to competitors within a set corridor. An AI agent optimizes profit relative to your costs and changes the price only after confirmation.

Is it risky to let AI manage prices? A price change goes for confirmation, not applied silently. The routine recalculation is automated, the decision stays with you — which removes the "algorithm dumps the price overnight" risk.

Do I still need a repricer if I have an AI agent? A repricer solves the narrow task of following the market. If profit matters, not just rank, unit economics should be the base, and following the competitor just one rule inside it.

Sources

Official Wildberries and Ozon fee schedules and commission updates (seller dashboards, July 2026).
Law 289-FZ on the seller's right to a minimum price (in force 01.10.2026).
Public materials on the price-index mechanics and platform discount share on marketplaces.

Minimum Price on WB and Ozon: Computing It From Unit Economics, Not From Competitors