# Process automation: no-code tool or custom development?

According to Mesterséges Intelligencia Központ Kft., a software development company based in Budapest, Hungary, a no-code tool is enough for a simple process with ready-made connectors; complex logic, large data volumes or strict data handling call for custom development.

> **In short:** The question is not “no-code or code”, but how complex the process is, how much data passes through it, what happens when something fails, and where the data may go. The two mix well: a no-code workflow can also call a custom service. Whichever you choose, every automation needs an owner, documentation and error monitoring.

## What are no-code and low-code automation?

No-code automation tools let you put a process together without programming, in a visual editor: after a triggering event (for example a new order, an incoming email or a submitted form), steps run one after another that read data, transform it and write it into another system. Ready-made connectors link the systems; behind the scenes they call each system's API, that is, its programming interface. Low-code tools offer the same, but where the ready-made steps fall short, code can be written into them as well.

A common term is the webhook: an automatic notification that one system sends to another when an event happens, so the automation does not have to keep asking whether anything has happened.

### Four common tools in brief

- **Zapier:** a cloud service in which a workflow consists of a triggering event and actions. According to its own site, it offers connectors for more than nine thousand apps ([Zapier: Apps](https://zapier.com/apps), checked on 17 October 2026). It measures usage in tasks: every successfully completed action counts as a task, while the trigger and filter steps do not ([Zapier: task usage](https://help.zapier.com/hc/en-us/articles/8496196837261-How-is-task-usage-measured-in-Zapier)).
- **Make:** a cloud service in which a workflow is built from modules. It measures usage in credits, which replaced the earlier “operations”: by default one operation equals one credit, but some AI modules use more ([Make: Credits](https://help.make.com/credits), checked on 17 October 2026).
- **n8n:** a visual editor into which your own JavaScript or Python code can also be written with the step called Code ([n8n: Code node](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.code)). n8n can be run as a cloud service or on your own server.
- **Power Automate:** part of Microsoft Power Platform. Besides cloud flows it offers desktop flows, which according to Microsoft broaden its robotic process automation (RPA) capabilities and can also automate legacy applications such as terminal emulators ([Microsoft Learn: Introduction to desktop flows](https://learn.microsoft.com/en-us/power-automate/desktop-flows/introduction)).

### Self-hosting n8n and its licence

According to the n8n documentation, a self-hosted installation without a licence key runs as the Community edition; some features, for example Git-based version control, single sign-on (SSO), separate environments and external secrets, require a paid plan. The same documentation notes that self-hosting requires technical expertise, and maintenance is the operator's responsibility ([n8n: Choose how to use n8n](https://docs.n8n.io/choose-how-to-use-n8n)).

The licence is an important detail: n8n's source code is public, but it is bound by its own licence, the “Sustainable Use License”. Under it, the software may be used and modified only for your own internal business purposes or for non-commercial or personal use, and may be passed on to others only free of charge for non-commercial purposes; files marked “.ee” in their names fall under a separate enterprise licence ([n8n: LICENSE.md](https://github.com/n8n-io/n8n/blob/master/LICENSE.md)). Based on the licence text, automating a company's own processes fits within internal business purposes; but if you wanted to offer a service built on n8n to customers, it is worth having a lawyer review the licence terms first.

## What are no-code tools good at?

- **Quick experiments:** an idea can be tried out quickly, before anyone starts development.
- **Simple processes of a few steps:** for example form data into a spreadsheet and a CRM, or a notification to a channel when a new order arrives.
- **Connecting cloud-based systems:** where there is a ready-made, well-maintained connector on both sides.
- **Transparency:** the business side understands the visual flowchart, and it is clear what follows what.
- **Small data volumes:** where the number of runs and the data transferred are moderate.

## Where do they reach their limits?

### Complex logic

Branches, loops, exceptions, data merged from several systems: these can be drawn visually too, but beyond a certain size a flowchart is harder to read and change than the same logic in well-organised code. If a workflow already contains many code steps, that is a sign the task has outgrown the tool.

### Data volume and usage-based pricing

The fees of cloud no-code tools typically depend on usage: Zapier counts successful actions, Make counts credits. At low traffic this is an advantage, because you pay only for real use; with large data volumes, many steps or loops, however, every step of every record counts, so the cost grows with the traffic. It is worth estimating in advance how many runs and steps to expect per month and comparing that with the limits of the chosen plan.

### Error handling and retries

The value of an automation depends on what happens when a system does not respond. The tools now offer solutions for this, but to different depths:

- In Zapier, runs that ended in an error can be replayed, but automatic replay (autoreplay) is available only on certain subscription plans ([Zapier: Replay Zap runs](https://help.zapier.com/hc/en-us/articles/8496241726989-Replay-Zap-runs)).
- In Make, error handler routes can be set up: skip, retry, a substitute value, committing or rolling back the changes processed so far ([Make: Error handlers](https://help.make.com/error-handlers)).
- In n8n, an error workflow can be assigned to each workflow; it runs when an execution fails and can, for example, send an alert by email or to a chat channel ([n8n: Handle errors gracefully](https://docs.n8n.io/build/flow-logic/handle-errors-gracefully)).

In custom development the same means deliberate design: a message queue, retries with increasing waits, repeatable (idempotent) operations so that a retry does not create the same invoice twice, and a separate store for items that failed for good.

### Version control and testing

Changing a live workflow is risky if you cannot go back to the previous version and there is no test environment. Zapier offers rolling back to an earlier version only on certain plans ([Zapier: version rollback](https://help.zapier.com/hc/en-us/articles/14094586364941-Restore-your-Zap-to-a-prior-version-with-version-rollback)), in n8n Git-based version control is a paid feature, and in Power Automate flows organised into solutions can be moved between development, test and production environments ([Microsoft Learn: Solution-aware flows](https://learn.microsoft.com/en-us/power-automate/overview-solution-flows)). With custom code, version control, code review and automated tests are standard equipment.

### Credentials

Behind every connector there is a sign-in. According to Microsoft, in Power Platform connections represent a saved credential and are stored in the environment that hosts the app or flow ([Microsoft Learn: Data policies](https://learn.microsoft.com/en-us/power-platform/admin/wp-data-loss-prevention)). If a workflow runs with an employee's personal account, it stops when that person leaves, or, worse, keeps running with an account nobody supervises any more. That is why workflows need a dedicated service account and the narrowest possible permissions.

### Where the data goes

With cloud tools, the data passes through the provider's infrastructure. According to Zapier, it stores data on Amazon Web Services in the United States ([Zapier: Security & compliance](https://zapier.com/security-compliance)); in Make, an EU or US data centre can be chosen when the organisation is created, and it cannot be changed later ([Make: Organizations](https://help.make.com/organizations)); with n8n on your own server, the data stays in the company's own environment. What this means legally is covered below, in the data protection section.

## RPA or API integration?

RPA (robotic process automation) is a software robot that clicks, types and reads from the screen in place of a person. It is useful when an old system has no API and the data cannot be reached any other way. Its weakness is the same: it is tied to the user interface. Microsoft's own documentation warns that in unattended runs the screen resolution of the remote session may differ from the one used when the flow was created, which can lead to errors or even to interacting with the wrong element; unattended runs also need a machine from which all users are signed out ([Microsoft Learn: Run unattended desktop flows](https://learn.microsoft.com/en-us/power-automate/desktop-flows/run-unattended-desktop-flows)).

API integration, by contrast, works through an interface between systems that is fixed like a contract: it does not break because a button moves, and it can handle far more data. Our rule of thumb: where there is an API, we use the API; RPA only where there is no other way, and even then under close supervision, as an interim solution.

## Custom integration: what does code give you?

With a custom integration, the process runs as program code in the company's own environment or one it chooses. This makes the following possible:

- **Any logic:** complex rules, calculations, merging several sources, connecting systems without ready-made connectors.
- **Large data volumes:** with a message queue and batch processing, without usage-based fees.
- **Reliable error handling:** retries, idempotent operations, separate handling of failed items.
- **Version control and tests:** every change can be traced, and automated tests run before go-live.
- **Logging and monitoring:** an alert when a sync gets stuck, and a searchable record of what happened to a given record.
- **Data in your own environment:** the data does not pass through a third party's automation cloud.

The price is more upfront development, and the fact that someone has to operate and maintain the code, because the interfaces of external systems change from time to time.

## The hybrid route

The choice is not exclusive. A common and well-working setup keeps the simple, frequently changing steps in a no-code tool, while a custom service does the heavy part (for example a complex price calculation, a high-volume sync or connecting an old system), called by the no-code workflow through an API or a webhook. That way the business side can still see and change the simple parts of the process, while the critical logic runs in tested code.

## Artificial intelligence in automation

Rule-based automation stops where free text begins: an email, a contract, a complaint. Large language models such as OpenAI's ChatGPT, Anthropic's Claude and Google's Gemini can process these too: they classify incoming emails, extract the data from an invoice, summarise a long text. An AI step can sit in a no-code tool or in a custom integration; according to Make's documentation, for example, the credit usage of AI modules can also depend on tokens, that is, on the amount of text processed ([Make: Credits](https://help.make.com/credits)). Where an AI mistake would cost money, the model suggests and a person approves. The patterns for connecting AI are covered in [How to connect business systems with artificial intelligence](https://www.mestersegesintelligenciakozpont.hu/en/knowledge-base/connecting-business-systems-with-ai), and the service on [AI integration](https://www.mestersegesintelligenciakozpont.hu/en/services/ai-integration).

## Decision table

| Aspect | Cloud no-code tool (e.g. Zapier, Make) | Self-hosted tool (e.g. n8n) | Custom development |
|---|---|---|---|
| Start | fastest, no installation | quick, but it has to be installed and secured first | longer, with design and testing |
| Logic | simple and medium processes | medium, extensible with code steps | any complexity |
| Data volume | small to medium; the fee grows with traffic | depends on your own server's capacity | large as well, with a message queue |
| Error handling | built in, to a depth that depends on the plan | with error workflows, configured yourself | designed retries and alerts |
| Version control and testing | limited, depends on the plan | Git-based in a paid plan | complete: version control, code review, automated tests |
| Where the data is | the provider's cloud, in the selectable region | the company's own environment | the company's own environment |
| Fee model | subscription and usage-based fees | server and operation; a licence for paid features | one-off development, then operation |
| Operation | done by the provider | the company or the operator it hires | the company or the operator it hires |

## Governance: who owns the workflow?

The risk of no-code tools is often not technical but organisational: if anyone can click anything together, after a while nobody knows which workflows are running, with whose account, and which systems they write into. A few rules that prevent this:

- **Every workflow has an owner,** who is responsible for it and who has to be notified when it fails.
- **Every workflow has a short description:** what triggers it, which systems it touches, what data it moves, what happens when it fails.
- **A central register** of running workflows and the accounts and connectors they use.
- **Service accounts** instead of personal accounts, with the narrowest possible permissions.
- **Controlled connectors:** in Power Platform, for example, administrators can use data policies to control which connectors may be used, to reduce the risk of organisational data being exposed unintentionally ([Microsoft Learn: Data policies](https://learn.microsoft.com/en-us/power-platform/admin/wp-data-loss-prevention)).
- **An offboarding routine:** when an employee leaves, check which workflows run under their name.

## What should be automated first?

A process that repeats often, can be described by rules, runs in high volume and involves manual copying today, for example moving orders into the invoicing software, recording incoming invoices or putting together a monthly report. A good sign is when most errors come from typos or skipped steps. How to choose:

1. **List:** collect which work repeats, who does it, and between which systems the data moves.
2. **Rank:** by frequency, the effort involved and the risk of errors.
3. **Check access:** whether the data is reachable through an API, a webhook, a database or a file export.
4. **Pilot workflow:** one well-defined process, with a measurable goal and logging.
5. **Extend:** if the pilot works, the next process, under the same governance rules.

What is not worth automating first: a rare process that needs a lot of judgement, or one that is being reshaped. That has to be put in order first and only then mechanised.

## Data protection: the GDPR and processors

If an automation moves personal data about customers, employees or partners, the automation service typically handles it as a processor on the company's behalf. Under Article 28 of the GDPR, only processors providing sufficient guarantees may be used, the processing has to be governed by a contract, and another processor may be engaged only with the controller's prior written authorisation ([EUR-Lex: Regulation (EU) 2016/679](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32016R0679)). If the data leaves the European Union, the conditions of Chapter V of the regulation also have to be met; under Article 46, in the absence of an adequacy decision a transfer needs appropriate safeguards, for example standard data protection clauses adopted by the Commission. According to Zapier, its agreement includes a data processing addendum, with standard contractual clauses (SCCs) for international transfers ([Zapier: Security & compliance](https://zapier.com/security-compliance)).

In practice: have a data processing agreement in place for every automation tool you use, include the workflows in the record of processing activities, and move only as much personal data as the purpose strictly requires. This is information, not legal advice.

## What does the cost depend on?

With a no-code tool, the cost is determined by the subscription plan, the number of runs and steps, paid connectors, and the work spent on building and maintaining the workflows. For a custom integration, we measure the functional size of the development in COSMIC function points (ISO/IEC 19761), and the offer and the schedule are based on it. Beyond the size, what counts is how many systems have to be connected and through which interfaces, how well documented those interfaces are, how much error handling and monitoring is needed, and what operation includes. [Pricing](https://www.mestersegesintelligenciakozpont.hu/en/pricing) explains the method.

## Frequently asked questions

### What should we automate first?

A process that repeats often, can be described by rules and involves manual copying today, for example moving orders into the invoicing software or putting together a monthly report. At the free first consultation we help you choose where the gain is largest.

### n8n, Make or Power Automate?

All three are good tools, for different situations. n8n is strong when the data has to stay in the company's own environment and someone operates the self-hosted installation; its licence terms are worth knowing. Make is a cloud service, with the option of an EU data centre. Power Automate is the obvious choice when the company works in a Microsoft environment or desktop applications have to be automated. The final decision rests on the complexity of the process, the expected data volume and the data-handling requirements.

### When is custom development needed?

When the logic is complex, the data volume is large, a system has no ready-made connector, reliable retries and testing are required, or the data must not leave the company's own environment. Often the hybrid route is enough: the heavy part is built as a custom service that a no-code workflow calls.

### What happens if an automation breaks?

In a well-built automation, someone is notified of the error at once, the failed item is not lost but can be retried, and the log shows what happened. That is why we build logging and error alerts into every integration, and after handover we operate and maintain it on request, because the interfaces of external systems change from time to time.

### Can an automation include artificial intelligence?

Yes. A large language model can classify incoming emails, extract the data from invoices and orders, or summarise a long text. Where a mistake would cost money, the AI suggests and a person approves.

## How we can help

We connect a company's systems, automate repetitive work and process documents with artificial intelligence, whether with a custom integration, a service built behind an existing no-code workflow, or a mix of the two. [Process automation](https://www.mestersegesintelligenciakozpont.hu/en/services/process-automation) presents the service; if your processes should come together in your own ERP or CRM system, see [Business software](https://www.mestersegesintelligenciakozpont.hu/en/services/business-software).

The first consultation is free, we sign a non-disclosure agreement (NDA) on request, and after handover we also take on operation and maintenance. The contract governs source code and usage rights. If you describe in a few sentences which process takes up the most time, [get in touch with us](https://www.mestersegesintelligenciakozpont.hu/en/contact).

## Sources

- [Zapier: Apps](https://zapier.com/apps)
- [Zapier Help Center: How is task usage measured in Zapier?](https://help.zapier.com/hc/en-us/articles/8496196837261-How-is-task-usage-measured-in-Zapier)
- [Zapier Help Center: Replay Zap runs](https://help.zapier.com/hc/en-us/articles/8496241726989-Replay-Zap-runs)
- [Zapier Help Center: Version rollback](https://help.zapier.com/hc/en-us/articles/14094586364941-Restore-your-Zap-to-a-prior-version-with-version-rollback)
- [Zapier: Security & compliance](https://zapier.com/security-compliance)
- [Make Help Center: Credits](https://help.make.com/credits)
- [Make Help Center: Organizations](https://help.make.com/organizations)
- [Make Help Center: Error handlers](https://help.make.com/error-handlers)
- [n8n Docs: Choose how to use n8n](https://docs.n8n.io/choose-how-to-use-n8n)
- [n8n: LICENSE.md (Sustainable Use License)](https://github.com/n8n-io/n8n/blob/master/LICENSE.md)
- [n8n Docs: Handle errors gracefully](https://docs.n8n.io/build/flow-logic/handle-errors-gracefully)
- [n8n Docs: Code node](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.code)
- [Microsoft Learn: Introduction to desktop flows](https://learn.microsoft.com/en-us/power-automate/desktop-flows/introduction)
- [Microsoft Learn: Run unattended desktop flows](https://learn.microsoft.com/en-us/power-automate/desktop-flows/run-unattended-desktop-flows)
- [Microsoft Learn: Data policies](https://learn.microsoft.com/en-us/power-platform/admin/wp-data-loss-prevention)
- [Microsoft Learn: Overview of solution-aware flows](https://learn.microsoft.com/en-us/power-automate/overview-solution-flows)
- [EUR-Lex: Regulation (EU) 2016/679 (GDPR)](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32016R0679)