In short

AI can be implemented in a company where the process lives across 1C, WhatsApp, and Excel. But the project must not pretend this is one clean system. The invoice may be in 1C, the approval in WhatsApp, the working status in Excel, the CRM stage may be stale, and the real agreement may be in a manager’s head.

The job of AI is not to replace every system. It is to collect context, show sources, draft the next action, and stop where it lacks permission or confidence.

Map sources first

Before choosing a model, map the sources: customer, invoice, payment status, order status, customer messages, exception approvals, real spreadsheets, and source owners.

This quickly shows that “integrate with 1C” is too broad. Sometimes you only need payment status. Sometimes an act, inventory, counterparties, or a temporary export is enough for a pilot.

Start with read-only 1C

1C often holds sensitive data: money, documents, counterparties, inventory, acts, and invoices. Start with reading and comparison.

AI can find an invoice, compare amount, BIN, contract, date, and counterparty, highlight mismatches, draft a note for accounting, show payment status to a manager, or warn that data is missing.

Avoid automatic posting, directory edits, payments, period closing, and customer promises at the start.

WhatsApp is an entry point

WhatsApp is a useful entry channel and a poor single source of truth. It has context but little structure.

AI can read the dialog, detect intent, extract details, draft a reply, create a task, and hand off. The final state should still be recorded in CRM, a task system, or another operational source.

Excel is often real

Excel may be called temporary, but many companies run on it for years: schedules, branches, shipment plans, statuses, prices, owners, and manual reconciliations.

For a pilot, Excel can be useful. It gives access to reality without a long integration. But it needs an owner, stable columns, update rules, access control, and conflict handling against 1C or CRM.

First pilot architecture

Do not connect everything to everything. For sales, connect WhatsApp lead intake, CRM, a service table, manager draft, and task creation. For finance, extract document fields and compare them with 1C or an export. For logistics, combine an Excel vehicle schedule, WhatsApp driver messages, and 1C documents into an exception summary.

Each case needs one owner, one workflow, one value metric, and clear action limits.

When custom work is needed

Low-code tools help when the process is linear: receive message, create task, send notification, fill row, call API.

Custom development is needed for conflicting sources, roles, mixed language, OCR, RAG, business rules, human approval, logs, error review, and non-standard 1C integration.

Quality control

Test real cases: messy Excel, stale CRM status, incomplete WhatsApp dialogs, 1C conflicts, document scans, mixed language, angry customers, and discount requests.

For every case, define expected behavior: answer, clarify, show source, draft, hand off, or refuse. That becomes the eval set.

For the rollout structure, see why AI projects need evals and AI pilot in 30 days.

FAQ

Can we implement AI if everything is in Excel?

Yes, if you choose one workflow and give the spreadsheet structure, ownership, access control, and update rules.

Should AI write to 1C automatically?

Not at the start. Begin with reading, comparison, and drafts. Add writing only after pilots, logs, and approvals.

What if WhatsApp and 1C disagree?

The agent should show the conflict and hand off to a person instead of silently choosing one version.