How to automate IT processes with n8n: visual editor, integrations, AI, and company case studies

n8n is a process automation platform: its visual editor connects events, data, and actions across different systems. For example, a form submission can trigger field validation, CRM record creation, and a notification to the person responsible. A separate AI step can classify the request or prepare a draft response.

Ready-made nodes reduce the amount of manual code. Custom APIs, complex transformations, and reliable operation under load require technical expertise. Value is determined by the outcome of a specific process, not by the number of workflows built.

A video about working with the platform is available on YouTube and Rutube; links are provided at the end of the article.

n8n source code is available, but the company classifies the product as fair-code rather than open source under the OSI definition. The core code is distributed under the Sustainable Use License; a separate license applies to enterprise components.

The license permits internal business use and workflow configuration services. Reselling access to hosted n8n or embedding it in a commercial product requires a separate review of the terms. The free Community Edition does not mean that servers, model calls, and maintenance are free. Edition terms and contents were checked on 28 September 2026; links are provided at the end of the article.

Sales: move the request to the CRM, check required fields, and assign an owner. Support: compile the request history and prepare a brief summary for the operator. Reporting: retrieve data from systems via API, check its completeness, and send a summary. These are workflow examples, not promises of a ready-made result.

Before choosing a node, check the required operation, API version, authentication method, and request limits. A service logo in the integrations catalog does not prove support for a specific action. For 1C, separately verify the published interface, permissions, and write rules in the configuration you use. More about system alignment — in integration decisions.

Community Edition does not include built-in Git version control, SSO, projects, shared access to workflows and credentials, integration with external secrets stores, or log streaming. Check feature availability in paid versions against the plan you choose. Standard execution logs and JSON exports are not equivalent to centralized auditing and a change-approval process.

For medium-sized and large businesses, selection starts with requirements: who uses the workflow, who can change it, where it runs, and who can see the data. Assess cost together with infrastructure, licensing, external APIs, models, and maintenance effort. When comparing Make, Zapier, or Airflow, evaluate the same process and these cost categories—there is no universally cheapest option.

1. Describe the outcome. For example, a valid request appeared in the CRM and the responsible person received a notification. Record manual time, volume, and frequent exceptions. 2. Choose a hosting option. For the cloud option, check the service terms; for your own server, follow n8n's current Docker Compose instructions. The Desktop App is no longer supported: its repository is archived, so it is not suitable for a new deployment. 3. Prepare test connections. Use the minimum necessary permissions and anonymized data. Start with a trigger, a check, and one action, then add the remaining steps. 4. Check exceptions. A duplicate event, empty field, unavailable API, exceeded limit, and approver rejection must each have an expected outcome. 5. Assign an owner. They review the pilot outcome, analyze failures, and approve changes. A catalog template can serve as a starting point, but its connections and logic must be checked.

The n8n graph defines the execution sequence and conditions; an AI agent can choose actions within the tools available to it. An agent can be part of the workflow itself. These approaches can be combined: a rule is sufficient for checking required fields, while a model may be needed for a brief summary of free text.

The surrounding graph does not make the model's response deterministic. Configure the tool list, permissions, result checks, and write authorization explicitly. For selected tools, n8n's AI Agent supports human approval; connect it to the required actions. An employee's access to the assistant should not automatically grant all permissions of the technical account.

For AI steps, set a process budget, account for model calls and retries, and define a stop condition when the limit is exceeded. It is useful to see costs for each workflow or agent, but this tracking must be designed and validated. Model selection and control of its actions are part of AI implementation in business processes.

Delivery Hero: restoring employee access. In the published n8n case study, a manager approves a request, after which the workflow works with Okta, Jira, and Google Workspace. The company used Enterprise. Employee lockout time fell from 35 to 20 minutes; with approximately 800 requests per month, this represents about 200 fewer hours spent waiting for access. This is employee time without access, not confirmed savings of 200 hours of IT staff work and not an AI result.

Oversee: preparing context for support. In the n8n case, an AI workflow collects technical and operational information, prepares a summary, and an employee chooses the next action. Fully autonomous customer communication is described as a direction for future development, not as an achieved result.

Both examples were published by the platform provider and are not KT.Team projects. They should not be applied to another process without measuring the baseline; the original sources are provided below.

Compare the full effort required for the accepted operation before and after the pilot: execution, result checking, corrections, exceptions, and maintenance. Use the same operation types and volume. Measure response waiting time and employee working time separately. Compare platform, server, and model costs with the actual labor released, not with the duration of an automated run.

For handover, prepare workflow exports, a list of dependencies and versions, connection details, access rules, and failure instructions. Check exports for sensitive data; transfer secrets through an approved secure store. A single JSON file does not replace environment configuration and permissions.

Acceptance testing is performed by the team that will support the process: it deploys the workflow according to the instructions, changes a connection, analyzes a test failure, and restores operation. Updates are first tested in a test environment, and the recovery method is documented. If you need help with this launch, at n8n page the approach to implementing the platform is described.