n8n vs Make vs Custom Code: Which Automation Fits?

Use Make when non-technical teams need to connect popular SaaS tools quickly, use n8n when engineers want visual workflows with code, self-hosting and predictable per-run billing, and write custom code when the automation is part of your product, handles high volume, or carries business rules you must test and version like any other software. Most growing companies end up using two of the three. This guide explains how each option works, how they are billed, where each one breaks down, and a simple process for choosing.

What business process automation actually covers

Business process automation means moving repetitive work, such as copying data between systems, routing approvals, enriching leads, generating documents or triaging support tickets, from people to software. A typical workflow has three parts:

1. A trigger. A new form submission, a webhook from Stripe, a row in a spreadsheet, or a schedule.
2. Steps. Look up a record, transform data, call an API, ask an AI model to classify or summarise something, branch on a condition.
3. An outcome. Update the CRM, send an email, create a ticket, post to Slack.

The tool you pick decides who can build and change these workflows, how much each run costs, where your data lives, and how easily the workflow survives growth.

The three options at a glance

Make: fastest path for SaaS glue

Make is a visual builder where each workflow, called a scenario, is a chain of modules. It shines when a sales or operations lead needs to connect tools like a CRM, Google Sheets and an email platform without waiting for engineering.

How billing works. Make now bills in credits. According to Make's help centre, most non-AI module operations consume one credit, while some built-in AI features use a variable number of credits based on tokens or other usage. A module runs once per bundle of data it processes, so a scenario that loops over 500 rows can consume hundreds of credits in a single run. Make's pricing page also notes that scenarios stop running when credits run out until you add more.

Where it fits well:

• Lead routing, form-to-CRM syncs, simple notifications and report emails.
• Teams without in-house developers who still want to automate.
• Prototyping a process before deciding whether it deserves custom engineering.

Where it struggles: workflows that iterate over large datasets, complex branching that becomes hard to read on a canvas, and anything that needs code review, automated tests or a clear audit trail of changes.

n8n: visual workflows with an engineering escape hatch

n8n is also a visual workflow builder, but it is designed with technical users in mind. You can drop JavaScript or Python into any workflow, call any HTTP API, and build AI agent steps that use your own model keys.

How billing works. On n8n's paid plans, the unit is the execution. n8n's pricing page defines an execution as a single run of the whole workflow, however many steps it has or however much data it processes. For data-heavy workflows, that can make costs far more predictable than per-step billing.

Self-hosting and licensing. n8n can run on your own infrastructure. The free Community edition is distributed under the Sustainable Use License, and n8n's license FAQ says you can use it for your own internal business purposes, and an agency can build automations for clients on its instance as long as those clients cannot create or edit workflows themselves. You cannot host n8n as a service for others to build workflows, or let your product's end users configure their own workflows, without a commercial agreement. Features such as SSO, multiple environments and Git version control sit on paid plans, as described in n8n's plan comparison.

Where it fits well:

• Internal operations workflows with real logic: deduplication, enrichment, multi-step approvals.
• AI-assisted workflows, such as classifying inbound emails or drafting replies for human review.
• Companies with data residency or privacy requirements that prefer to keep workflow data on their own servers.

Where it struggles: self-hosting means you own uptime, upgrades, backups and security patches. And if customers of your product need to build their own automations, the licence terms above matter.

Custom code: when automation becomes product

Custom code means writing the workflow as a normal service: a queue consumer, a scheduled job, a serverless function or a set of background workers, written in your main language and deployed through your usual pipeline.

Choose custom code when:

• The workflow is a product feature. If customers see or depend on it, it deserves the same tests, monitoring and release process as the rest of your product.
• Volume is high or spiky. Thousands of events per minute are cheaper and more reliable on queues and workers you control than on a per-step billed platform.
• The logic is complex or regulated. Pricing rules, financial calculations or healthcare data flows benefit from unit tests, code review and clear audit history.
• You need fine-grained control over failure. Idempotency keys, retries with backoff, dead-letter queues and exactly-once style guarantees are easier to design in code.

The trade-off: custom code needs engineers to build and change it. A one-line change to an email template may need a deploy, unless you design for configuration from the start.

A five-step process to choose

1. Map the workflow on one page

Write down the trigger, each step, the systems involved and the final outcome. Note who owns the process and who will need to change it later. If the answer is "the ops team, weekly", that points towards Make or n8n.

2. Estimate volume and billing shape

Count how often the workflow runs and how many items each run handles. Per-step billing grows with every row you loop over, while per-execution billing grows with runs. Custom code grows with infrastructure and engineering time. Run a rough monthly estimate for all three using the vendors' current pricing pages, not old blog posts, because plans change often.

3. Check data sensitivity and residency

If the workflow touches personal data, payment data or health records, decide where that data is allowed to flow. Under GDPR, any tool that processes personal data on your behalf is a processor, and Article 28 requires a written processing agreement with it. Self-hosted n8n or custom code keeps data on infrastructure you control. Make and n8n Cloud may still be fine, but review their data processing terms first.

4. Decide how much logic lives in the workflow

A good rule of thumb: if a workflow needs more than a handful of code nodes, or you find yourself copying the same logic between workflows, move that logic into a small API or service you own. Let the visual tool orchestrate, and let code do the hard thinking.

5. Plan the exit before you start

Automations tend to grow quietly until a critical process depends on one person's canvas. Document every production workflow, name an owner, set up error alerts, and agree on the point (a volume threshold, a customer-facing requirement, or a compliance need) at which you will move it into custom code.

A practical hybrid that works for many teams

Many teams land on a layered setup:

• Make or n8n for internal glue: CRM updates, notifications, reporting and lead enrichment.
• n8n with AI nodes for human-in-the-loop AI steps, such as drafting replies or tagging tickets, where a person reviews the output. If you are designing AI steps that rely on your own documents, our guide to RAG vs fine-tuning explains how to ground model output in your data.
• Custom services for anything customer-facing, high-volume or regulated, exposed through clean APIs that the visual tools can call.

This keeps simple changes in the hands of the people who own the process, while core business logic stays tested, versioned and monitored.

Common mistakes to avoid

• Building a product on a no-code canvas. It works for a demo, then becomes the hardest part of the system to change safely.
• Ignoring per-step costs. A loop over a large list can multiply usage overnight. Test with realistic data volumes before going live.
• No error handling. Every production workflow should alert someone when it fails and should be safe to re-run without creating duplicates.
• Shared personal logins. Connect tools with service accounts and scoped API keys, not an employee's personal credentials.
• Unreviewed AI output going straight to customers. Keep a human approval step until you have measured quality on real cases.

How Technier helps

Technier has been building software since 2022, with a team of about 30 people, including around 20 senior engineers, working across any client time zone through Aspire Enterprises LLC. We help companies decide which workflows belong in a visual tool and which deserve proper engineering, then build both: AI and machine learning solutions such as AI-assisted triage and document processing, and SaaS development for automation that becomes part of your product. For an example of AI workflows in production, see our MyWorker AI project.

FAQ

Is n8n better than Make?

Neither is better for everyone. Make is usually quicker for non-technical teams connecting popular SaaS apps. n8n suits technical teams that want code inside workflows, self-hosting, and billing per workflow run rather than per step.

Can I use self-hosted n8n for client work?

According to n8n's license FAQ, you can build automations for clients on your own instance and charge for that work, as long as clients cannot create or edit the workflows themselves. Letting outside users build their own workflows requires a commercial agreement with n8n.

When should we replace a no-code workflow with custom code?

Move it when the workflow becomes customer-facing, when volume makes per-step billing expensive, when the logic needs automated tests and code review, or when compliance requires tighter control over where data flows.

Can AI agents replace these automation tools?

AI agents are good at judgement steps like classifying, summarising and drafting. They still need reliable triggers, integrations, permissions and error handling around them, which is exactly what automation platforms and well-built services provide. In practice, agents are a step inside a workflow, not a replacement for it.

Ready to automate the right way?

If you are deciding between n8n, Make and custom code, or want to move a fragile workflow into proper software, we can map your processes and recommend the simplest setup that will hold up as you grow. Book a call with Technier to talk it through.