AI and Automation for 1C

Recognizing waybills and documents from photos

Our own DocVision service, connected to 1C.

Suppliers send waybills, invoices and acts in every form: a scan, a phone photo, a PDF from their own software. A warehouse worker or accountant retypes them into 1C line by line, checks the totals and looks up each item in the catalog. On a large delivery that is an hour of work and inevitable typos.

We have our own document recognition service, DocVision. It reads the document, extracts the header details (supplier, BIN, number, date, amounts) and the line items, then matches the lines to your catalog by supplier code, SKU, barcode or name. A draft receiving document is created in 1C.

1. An employee photographs the waybill or forwards the file to a bot.
2. The service recognizes the details and lines.
3. Lines are matched to your catalog, and uncertain ones are flagged.
4. A draft receiving document appears in 1C.
5. The employee checks the flagged lines and posts the document.

Totals are checked separately: if the sum of the lines does not match the document total, or a price differs sharply from the last purchase, the line is flagged. That way recognition catches not only its own errors but also the supplier's mistakes on the waybill.

Matching learns from confirmations: once an employee links a supplier's item to your card, the link is filled in automatically next time. More in recognizing invoices and waybills with input into 1C (in Russian).

Smart product search

Find an item the way the client names it.

Standard 1C search looks for a substring. A client writes "wood screw 4 by 50", the catalog says "Screw ShSGD 4.0x50", and search finds nothing. A manager spends minutes on every order line, and a newcomer cannot find the item at all without help from an experienced colleague.

Smart product search accounts for typos, synonyms, different ways of writing sizes and units, transliteration and the meaning of the query. To do this, the catalog is indexed in a separate search service with vector search, and 1C queries it through an HTTP service. Results come back as a list of product cards with stock and prices, like a normal product selection.

The search understands Kazakh, Russian and mixed queries, as well as the terms builders, mechanics or medical staff use, once their vocabulary is added. The tidier the catalog, the more accurate the result, so we check its quality at the start.

The same search can be connected to the website, a sales rep mobile app or a bot. Then the client finds the item themselves and the manager receives a correct order. Technically this is part of the 1C API, which opens database data to other systems securely and with restricted rights.

Dashboards and analytics on 1C data

Numbers on the executive's phone, not in a report on the accountant's computer.

1C reports are good for accounting but inconvenient for a daily glance from management: they have to be opened, configured and generated, which is almost impossible on a phone. A dashboard on 1C data shows the essentials in a browser or on a phone: revenue, margin, payments, stock, plan vs actual by manager, daily and weekly dynamics.

Data is exported from 1C on a schedule into a separate store, and the dashboard works from that. This way heavy analytics queries do not slow down the database users work in. Updates run hourly or daily, depending on the task.

Dashboard access is role-based: the owner sees everything, a branch manager their branch, a sales manager their own sales and plan. Dashboard numbers match 1C because they come from the same registers, not from separate spreadsheets.

If what you need is not a dashboard but a report inside 1C with the right breakdowns, that is a task for 1C reports on SKD. The difference and how to choose are described in dashboards on 1C data (in Russian).

Timelines and cost

Each scenario is estimated separately.

Neural network and server costs are accounted for separately: they depend on the volume of documents and queries and are calculated upfront during the pilot. We name them before launch so there are no surprises on the bill a month later.

Post-launch maintenance can be included in a 1C support subscription: we monitor recognition quality and update notification rules and thresholds.

Who this is for, and how we differ

AI on top of accounting that already works.

AI for 1C pays off where there is a lot of repetitive manual work or a lot of data that nobody has time to look at: distributors with hundreds of deliveries a month, retail with thousands of items, manufacturers with constant purchasing, chains with several warehouses and branches.

The difference is that we are both 1C developers and developers of AI services, mobile apps and bots. So we see the whole chain: how data is stored in 1C, how to expose it safely, how to build a convenient interface for staff. There is no need to coordinate three contractors. If 1C itself needs to be put in order first, start with the 1C developer page.