I find where a shop loses time and automate what pays off fastest
A shop selling on its own site and on Wildberries and Ozon, Russia's two biggest marketplaces, piles up more manual work every month. Stock levels drift apart across the platforms, support answers the same questions over and over, and the accountant pulls documents from three places. In a two-week audit I work out where key staff spend their time and draw up a list of tasks to automate, with a payback estimate for each. In my estimate, three fixes come first: stock sync, AI for first-line support and closing documents generated by the 1C accounting system.
Headcount grows every month, and revenue can't keep up
For the estimate, take a home-goods shop with 280 orders a day and 12 product categories, selling on its own site, on Wildberries and on Ozon. The team is 14 people, and a new hire joins almost every month.
People in key roles are always busy with something urgent. Stock levels on the site, on Wildberries, on Ozon and in 1C keep drifting apart. A manager reconciles them by hand three times a day, but the shop still sells items it doesn't have and runs into cash gaps.
Support answers the same questions 80–120 times a week: "where's my order", "what size is the bed", "do you have this color". The accountant needs 9 working days to close each month, pulling documents from three places: 1C, the marketplace dashboards and email.
The audit produces a task list, with the quickest payback first
Every task on the list comes with its own payback estimate. For a shop like this one, three fixes sit at the top.
The site, WB, Ozon and 1C are reconciled with each other. The workflow automation tool n8n polls all the systems and matches stock by SKU.
GigaChat, Sber's language model, answers the 40 most common questions. It finds the answers in the product cards and the FAQ table (a technique called RAG).
When a company places an order, 1C generates the UPD (Russia's universal transfer document), and the customer receives it by email.
Stock is synced every 15 minutes
Each platform has its own ID for the same product, its own units and its own way of storing stock. In n8n I build a single SKU mapping table and set every source to be polled every 15 minutes. 1C is the master system. After each sync, n8n recalculates stock on the site and updates the WB and Ozon listings through their seller APIs.
AI takes the simple questions, a person takes the hard ones
The bot answers questions like "where's my order", "what size", "is it in stock" and "when will it arrive" on its own; the estimate assumes that covers 78% of messages. A manager takes over in four cases: the customer is irritated; the message contains words like "return", "defect" or "lawyer"; the model's confidence in its answer is below 0.7; or the customer asks the same question a second time.
Closing documents assemble themselves
When a company's order ships, 1C generates the UPD. Then n8n picks it up through 1C's REST and OData interfaces, attaches a PDF guide and sends it to the customer with the tracking number. The accountant only checks that everything adds up.
- → An AI sales chat on a shop's website runs on Russian language models and answers shoppers from a base of 562 product facts
- → Revenue reconciliation across three sources runs every day with a 3% threshold
- → A converter for Federal Treasury statements turns them into a file ready to import into 1C
Tools that work with 1C and the marketplaces
Runs all the syncs and workflows
The master record for stock. The closing documents are generated here too
All orders from the site and the marketplaces in one window. The AI support reads from here
Reads and writes stock, updates listings, tracks order statuses
The same as for WB: stock and orders synced in both directions
Answers customers on first-line support
The FAQ base and the team's written procedures
For n8n and the supporting services. The shop chooses the hosting provider
What to expect after launch
The figures are for the same example shop. Where a card shows two numbers, the first is today's figure and the second is the target. For your shop, I'll recalculate everything during the audit from your system logs and timings.
almost one full-time role on a 40-hour week, or ₽89k a month in payroll
25 times lower: the shop almost stops selling items it doesn't have
investment: ₽380k for the in-depth audit + ₽220k to build = ₽600k
including questions passed to a manager; the estimate assumes the bot resolves 78% on its own
the accountant only checks that the documents add up
The main thing the estimate shows is how much manual work can come off the team before the next hire. The time freed up can go to growth work: stock turnover, purchasing and new categories.
The payback counts payroll savings only: ₽600k ÷ ₽89k a month is about 7 months. I left out the gains from fewer penalties for selling out-of-stock items and from faster support. They depend on the shop, and I estimate them during the audit.
When an audit is the right place to start
The typical situation: "manual work piles up faster than we can hire". An audit is for an owner who already senses that "something needs automating" and wants to know what to tackle first. It suits:
- → Retailers and online shops that sell on their own site and on 2–3 marketplaces
- → Companies of 10–50 people with a lot of manual work: wholesale, B2B services, agencies
- → Companies that make and sell their own goods, with a warehouse, suppliers and a website
- → Service chains (salons, clinics, car repair shops) with several locations and shared accounting
- → Any team where the owner sees everyone putting out fires and nobody getting to planned work
- I spend 90 minutes watching each key person at work to see how things are really done
- I export 90 days of logs from 1C, the CRM and the marketplace dashboards to see which operations happen most
- I measure how many hours the 20 most frequent operations take
- I hand over a list of tasks to automate, with a payback estimate and the assumptions behind it. We decide the order of work together
Blog article on this topic: Inventory sync between 1C, Wildberries and Ozon: why the numbers drift apart
Related cases and ready-made solutions
I stress-test AI chatbots and find the vulnerabilities before anyone exploits them
I attack a copy of the system following the OWASP LLM Top 10, a public list of typical threats to AI…
A Telegram bot for foremen: stock, budgets and payments from 1C right on site
A foreman opens the Telegram bot, sees data from 1C UNF in two taps and confirms deliveries with a photo of…
Converter for Federal Treasury statements into 1C
Reads XML statements from the Federal Treasury's GIS system in both format versions, V3 and V4, and builds a…
If your team grows faster than revenue, it’s time to look at your processes
You can start with the ₽5,000 audit: a 60–90 minute call and a written report. If there's a lot to cover, I run a detailed two-week audit. Implementation is a separate conversation once you have the report, and the decision is yours.
Or just ask a question. During business hours I reply within 2 hours