AI in 1C: what you can automate and how to implement it

What you can automate with AI in 1C

1C remains the accounting system. It stores data, calculates and posts documents by its rules. AI works alongside it and takes the jobs that involve reading, matching and sorting: parsing a scanned invoice, working out which contract a payment belongs to, finding a document from a description. Anything that needs a strict calculation by formula is still done by 1C itself.

Below are the tasks teams usually take on first. They involve a lot of repetitive manual work and have a clear test for "done right."

• Entering source documents: invoices, acts, delivery notes, contracts. AI extracts the details and line items and prepares a draft document in 1C.
• Reconciliations: matching payments, acts and invoices, and finding discrepancies with counterparties.
• Classification and allocation: assigning payments and documents to counterparties, contracts, accounts and departments.
• Search in the accounting database: ask a question in plain words and get an answer with links to documents.
• Prompts for employees: a reminder about an empty field, a possible error, or a mismatch with the contract.
• Exchange with CRM and other systems through an API, with no manual data transfer.

AI in 1C: document flow and accounting

A typical picture: invoices and acts arrive by email, in messengers and through 1C. An employee opens each file, retypes the details and line items, looks up the counterparty and contract in the reference lists, and only then creates the document. AI takes over the first part. It identifies the document type, extracts the details and lines, matches the counterparty and contract to the 1C reference lists, and prepares a draft. The accountant sees the original file next to the filled-in fields and checks them instead of typing.

Reconciliations and allocation work in a similar way. Allocation means assigning a payment or document to the right counterparty, contract and accounting item. AI compares the payment purpose, amounts and dates with open invoices and acts, suggests an option, and shows its basis: which feature led it to pick this contract. If the match is complete and the case falls under a clear rule, it moves on by itself. If not, it goes to review. For reconciliations with counterparties, AI collects the discrepancies into a list of "what doesn't match and why," instead of someone comparing statements by eye.

One important caveat. If documents always arrive in the same format, for example a bank export or electronic exchange, you don't need AI. A standard rule-based import is enough. AI makes sense where the format differs every time: scans, free-form emails, documents from different suppliers.

Search, employee prompts and CRM exchange through an API

Search in the accounting database lets you ask a question in plain words: "which contracts with this counterparty have no signed acts" or "which invoices on the project are waiting for payment." AI turns the question into a query to 1C, assembles the answer and gives links to the documents so you can check it. It is critical that access rights are preserved: an employee gets only what they already see in 1C under their own account.

Prompts work at the moment of entry. The solution notices that a field in a document is empty, an amount differs from the contract, or a counterparty looks like one already in the reference list, and it suggests what to check. This does not replace an accountant's review. It is a safeguard against inattention. The same prompts help new employees learn standard operations faster.

Exchange with CRM and other systems is built through an API, an agreed way for one program to call another without a person involved. When a deal is closed in the CRM, a draft invoice appears in 1C. When a payment is posted in 1C, the status changes on the deal card. AI is needed at the start of such chains: to parse a request from an email or messenger, determine the type of inquiry, fill in the card and assign a responsible person. Then managers and accountants don't have to move data between systems by hand.

Integrating AI with 1C: how to build it in without breaking your accounting

The main principle: AI should not be the only one deciding what goes into your accounting. It works as a separate layer around 1C and exchanges data with it through an API and standard exchange mechanisms, rather than rewriting it from the inside. So your current database and existing customizations stay in place, and updating the configuration is easier than after deep changes to the code. In practice, this comes down to a few rules.

Data is a separate question. Documents often contain personal data, so before you start, decide which fields the solution really needs and which tasks can be done with anonymized data. We account for the requirements of Federal Law No. 152-FZ (the Russian personal data law) at the design stage, not after the solution is built. The environment is chosen at the start: your own infrastructure or a Russian cloud.

• AI creates drafts and suggestions. Posting documents stays with an employee or with 1C's own rules until quality is confirmed on a real flow.
• The connection uses a separate account with the minimum necessary rights: only the reference lists and documents the solution works with.
• Test environment first, then production environment. Failures and rare cases are found before they reach your real accounting.
• Every action is logged: what the solution received, what it suggested, who confirmed it and when.
• A rollback procedure is written in advance: how to find and undo everything the solution created if something goes wrong.
• Monitoring and failure alerts are set up, so a stopped exchange isn't discovered through employee complaints.

Who reviews disputed cases

Any AI solution has doubts from time to time: a poor scan, an unfamiliar counterparty, an amount in a document that doesn't match the contract. So different routes are designed from the start. Clear cases with a full match move on automatically and land in 1C as a draft. Doubtful ones go to a review queue. Cases where a contradiction is found are stopped and passed straight to the responsible person.

The person who already owns that area does the checking: an accountant for primary documents, a procurement specialist for supplier documents, a manager for requests. Who handles which type of case is set in advance. In the queue, the reviewer sees the source document, the extracted data and the reason for doubt, so they don't start from scratch. Every correction is saved and used to refine the rules.

Before launch, the solution runs in parallel with normal work. People work as before, the solution runs alongside them, and the results are compared. Where they differ, we add rules and limits. Only then is the solution trusted with the live flow. Responsibility for posted documents stays with your employees, not with AI.

What to prepare before the start and what to ask a contractor

Preparing before the start makes the estimate more accurate and shortens discovery. The materials in the list at the end of this section are usually enough. If something is missing, we can discuss it at the first meeting.

It helps to ask a contractor these questions. How does the solution write data to 1C (a Russian ERP platform), and how is that different from a standard import? What happens when something fails, and how do you roll back what was created? Who checks disputed cases, and what does the queue look like? What criterion do you use to accept the work? Who owns the code, rules and settings? Will the solution still work after your configuration is updated?

Typical mistakes: trying to automate all accounting at once instead of one process; testing only on tidy examples; not agreeing on an acceptance criterion in advance; letting AI post documents from day one; not appointing someone to handle disputed cases. All of these can be fixed during discovery, not after launch.

• A short description of the process: what comes in, from where, who handles it and where the result goes.
• Examples of real documents, including the awkward ones: poor scans, non-standard forms, free-form emails. Please anonymize personal data where possible.
• Exports of 1C reference lists: counterparties, contracts, items, departments. The result is matched against them.
• Access to a test copy of the database and details of the customizations already in your configuration.
• A person on the business side who knows the process and can approve the acceptance criterion.
• An idea of how many errors and delays the process has now. Without a starting point, there is nothing to compare the quality of the solution against.

How we approach AI tasks in 1C

We start with a free discovery of one of your processes. We look at how it works now, whether there is data for it and which part makes sense to automate first. The result is a written document with a process map, the boundaries of the solution and an acceptance criterion. It stays with you even if you decide not to continue. If the task can be solved with a regulation or ordinary automation without AI, we will say so.

Next comes a prototype built on your real data, not on demo data. We build the first prototype in 3 weeks or more, and 1C and CRM integrations in 4 weeks or more. Cost guidelines: 1C, CRM and API integrations start at $3,600, and AI document processing starts at $2,100. We give the exact price after discovery. If the solution falls short of the agreed acceptance criterion, we improve it at our own expense. In this setup, disputed cases always go to a person.

After launch, we set up monitoring and analyze incidents. Support is available in monthly hour packages starting at $400, and the mode is fixed in the contract. Your own team can also run the solution. Choose the mode before launch, not after the first failure.