What users say about working with the agents
I run a small business — goods for farming and outdoor recreation, four stores on Ozon and Wildberries, two sole proprietorships. We started from scratch in June 2026: no automation at all, everything by hand.
Today 86 scheduled tasks are running, plus 4 Telegram bots, 9 services and 40 scripts for the marketplaces’ APIs. Instead of manually bringing the seller accounts together, there is a ready-made breakdown of yesterday across all four stores (the day’s summary, a 14-day forecast, stock, product processes, buyer discount analysis, platform news) and 3–5 actions a day.
202 shipments have been handled through the portal: distribution across clusters, request drafts, cargo spaces, labels, time slots, acceptance acts. Out of 208 tasks, the 5 failed ones show up in the journal instead of surfacing a week later.
I used to open four seller accounts every morning and spend half an hour bringing the numbers together. Now one message arrives: what happened yesterday across all four stores, what is going on with prices and promotions, where stock is below 7 days, what is happening with reviews — and three to five concrete things to do today. Since October 1 the report has been short — one A4 sheet: what happened / where we are heading / what to do. It can be read on a phone in two minutes.
The most painful spot was prices: the commission was 20 % here, a custom rate there, and I kept estimating profitability by eye. I thought I had raised the price and earned more, when in fact I was giving the platform more. They explained it to me on a live example: at a price of 285 ₽ the commission was −57 ₽ per unit, at 600 ₽ it was already −299 ₽. Now such things are visible in advance, and there is a unit economics floor — the price simply never drops below it.
Once I noticed: the product was in a promotion but had no promo price. We sorted it out — I had set the promo quantity limit myself, Ozon does not refresh it, the limit ran out, and the product stayed on the participants list but without a discount. The rule now is simple: the limit is always 0, no restriction. The price now changes directly, and the selling price for the buyer takes effect instantly. No more dancing around promotions — everything goes through one mechanism, and the whole project’s automation is switched to this channel.
Shipments used to be a separate all-day adventure. Out of 208 tasks, 202 have been completed — and the five that failed are visible in the journal with the reason, instead of surfacing a week later. I am particularly pleased with the acceptance act analysis: if the warehouse accepted less than we shipped, or mixed up the goods, the system finds the discrepancy itself and warns two days before the approval deadline. Short acceptances like these used to simply eat the money.
This was a dark zone for me: why does a competitor have the same price while the buyer pays less. Now there is a storefront snapshot for Ozon and Wildberries, and it shows what discount the platform gives. The most useful discovery: on October 15 it turned out that on items under 300 ₽ the discount had dropped to zero for most sellers, while above 300 ₽ it held at 25–50 %. We were looking for the cause in advertising, and it was the price threshold. Without the system you do not find this out — you only guess.
Bookkeeping was a separate headache: stock in the seller account, stock in MoySklad, stock in the marketplaces’ warehouses — three different versions of the truth. Writing shipments off to MoySklad is done with a “Write off to MoySklad” button right in the bot: it first shows the plan, then transfers on confirmation. We once ran 30 days of history — 17 moves, zero errors. The whole database is 707 MB, and it is entirely mine, not “somewhere in the cloud”.
I used to learn that a print had finished only when I walked into the workshop. Now a “print finished” or “print stopped” notification arrives. I calculate an item’s cost simply by writing the request to the bot in my own words: “3d-ML-5g-GR red MAKO plastic, printing on the A1, 13 pieces on the plate, packaging 10 ₽”. The output is a breakdown by plastic, time and packaging.
814 reviews from Wildberries are in the database — with ratings, texts and products. There is no need to page through seller accounts store by store to understand what customers complain about. This turned out to be unexpectedly useful: you can see that the problem is not the product but, say, the packaging — and that is fixed not with a discount but with a different box.
A separate word about the agent’s character. It argues with me when I am wrong and does not agree out of politeness. Once I said “make the report prettier” — I got back an explanation of why that format breaks in Telegram, and a working alternative. It does not invent numbers: if something is not in the database, it writes “I don’t know, I need to check” and goes to check.