n8n vs Make for a small business: compare the real cost of one workflow

Let's take a fictional example of preparing a quote. A request comes through a form, the system searches for the client, prepares a file, waits for approval, and records the result in the CRM. The final quote retains its usual validation path. To compare the tools, use the same fields, the same AI model, and the same acceptance criteria.

Keep a sample that includes complete files, ambiguous requests, and network errors. A platform that handles the simple case in few steps may require more interventions on exceptions. Also count the necessary modifications when a field changes in the CRM.

The guide for creating an agent with n8n provides ten starting cases. The expected results can be used for both versions of the process. The comparison then focuses on a defined job, with the same limits of action.

The n8n pricing page shows, as of October 12, 2026, Starter at €20 per month with annual billing for 2,500 executions, and Pro at €50 for 10,000 executions. These are amounts in euros and an annual commitment; the monthly cost without this commitment may differ. Options, taxes, and external services must be added to the calculation.

The Make pricing page defines credit as a consumption unit. An action generally consumes one credit, with exceptions and AI functions that can consume more. The rate of your plan must be noted with its volume, currency, and commitment. The number of modules displayed in the scenario does not directly give the total billed.

At n8n, an execution represents a workflow run. A new production execution triggered to resume a case can increase the count. Check the handling of sub-workflows and manual trials in your plan. In Make, repeated searches or writes on multiple items can consume credits for each processing. Review the consumption logs of both versions.

The following calculation is a sizing simulation. We assume one n8n execution per case, seven Make actions billed at one credit each, and 10% of cases fully resent. Waiting for approval remains in the retained workflow. No variable-consumption AI calls are included in the seven credits. These assumptions must be replaced by the consumption observed in your trial.

For 1,500 requests, the n8n calculation is 1,500 × 1.10. The Make calculation is 1,500 × 7 × 1.10. A partial resubmission, an additional AI call, or periodic reads changes these results. The numbers are used to select the package volumes to examine, not to announce a universal invoice.

Add the model expenses with its input and output tokens, then supervision. For an assumption of two hours of maintenance at CA$100 per hour, maintenance adds CA$200 per month. This fictional rate illustrates the formula and does not constitute a Daillac rate. Keep costs in euros and dollars separate until the application of a dated conversion rate.

The total monthly cost is the sum of the subscription, overages, the model, possible hosting, and operating time. Divide this amount by the accepted files to obtain a cost per successful file. The guide on budget of an AI agent details this measure and correction costs.

A hosted tool reduces server administration work. Connections, failures, and consumption still need to be monitored. With a self-hosted n8n instance, your team also handles updates, backups, and restoration. The Community edition documentation helps to check available functions before choosing this option.

Hosting responsibility must correspond to an available skill. Ask who will intervene if the workflow fails on a Friday afternoon and where to find recovery information. A low-cost server may require several hours of maintenance; a higher subscription may include features useful to your organization. Compare the functions that are actually necessary, such as access roles and history.

The platform location does not describe the entire path of the data. A workflow may call an external model, drop a file elsewhere, and notify messaging. Map these exchanges before choosing your architecture, especially when a request contains personal information.

Ask the person who will operate the process to correct a record in each tool. Have them find an error, change a rule, and check that a rejection prevents writing. Note the time required and the steps that remain difficult to understand.

Choose the solution that meets your company's constraints and that your team can maintain. The choice may favor n8n when self-hosting and technical integrations are necessary, or Make when the scenario and its operation are better suited to the team. The process test should support this decision. Our CRM and ERP automation service allows scoping this type of comparison on your test data.