Choose the platform around the work and who will run it.
n8n, Zapier and Make can each be useful parts of a business system. The choice depends on your actual connectors, rules, data requirements and operating capacity. A mixed approach can be sensible when one platform does not own the entire responsibility.
Start here.
Use the platform that can complete and recover your workflow at an acceptable total cost. Check the exact triggers and actions, permissions, failure handling and handover. Compare self-hosting and licensing where relevant; do not choose only by the subscription price or the number of available apps.
A connector fit check
Verify the fields, events and write actions your process needs. An app appearing in an integration catalogue does not establish that every required operation is available.
A realistic usage model
Estimate records, branches, searches, retries and AI usage for a representative month. A business transaction does not necessarily equal one billable unit.
An owner and exit path
Decide who maintains credentials, handles failed work and can change or move the implementation. Platform support and responsibility for your workflow are different services.
From an idea to a checked result
Map the transaction
Write down the input, reads, decisions, writes and final check. Include what happens if a system becomes unavailable halfway through.
Test connector coverage
Try the specific required operations with representative data. Check pagination, custom fields, attachments and account-level permissions.
Compare operating models
Consider hosted services, customer-controlled infrastructure and any custom components. Identify the person or supplier responsible for each part.
Estimate total cost
Model vendor usage, hosting, support and human review. Include ordinary retries and exceptions, not just the cleanest demonstration run.
Choose and document
Record why the option fits, its known limitations, recovery procedures and how the business can take over or change provider.
What changes when the work gets complicated?
A straightforward handoff between existing apps
- What arrives
- An accepted enquiry needs a CRM record and an assigned follow-up task.
- What is checked
- Confirm the trigger, required fields, duplicate handling and assignment action exist.
- What happens next
- Use an appropriate hosted connector workflow and test a repeat event and failed destination.
- What finishes the work
- A simple maintainable integration, without building infrastructure the task does not need.
Stock decisions need custom business rules
- What arrives
- Local stock, supplier availability and online orders disagree.
- What is checked
- Check ownership of each quantity, event identity, source age and permitted corrections.
- What happens next
- Keep connector steps where useful; implement the critical reconciliation rules in a testable component.
- What finishes the work
- A deliberate combination of orchestration and custom logic, with explicit recovery.
The workflow must run in a controlled environment
- What arrives
- A business needs specific infrastructure or access arrangements.
- What is checked
- Review hosting capability, selected edition, licensing, network access and who operates updates and backups.
- What happens next
- Assess a customer-controlled deployment against a hosted alternative before committing.
- What finishes the work
- A deployment choice with operating responsibility and licence fit made explicit.
A practical comparison
These are starting points for selection, not claims that one platform cannot perform another platform’s tasks. Capabilities and plan entitlements should be checked for the proposed setup.
| Option | Consider it when | Check carefully |
|---|---|---|
| n8n | Self-hosting or custom integration work is useful | Edition, hosting responsibilities and the licence for the exact client arrangement |
| Zapier | The required app triggers and actions provide a practical path | Field coverage, task usage and the behaviour of replayed work |
| Make | A visual scenario helps the team understand a multi-step data flow | Module behaviour, credit usage and exception recovery |
| Custom code | State, rules or interfaces need a specifically engineered component | Build cost, tests, deployment, maintenance and handover |
Do not confuse a tool with the delivered service
A canvas or workflow file is only part of the result. The business still needs correct mappings, permissions, tests, exception ownership and a way to know whether work finished.
Compare the completed workload, not headline units
Vendor billing units are not interchangeable. Use current documentation and a representative run to estimate the setup you are actually considering.
Zapier task usage
Zapier generally counts successful action steps as tasks; triggers do not count. Replayed successful steps and some product-specific features can consume additional tasks, so one incoming order may use several units.
Make credits
Make uses credits for billing. Usage can depend on operations and, for some features, tokens or other factors. A one-operation assumption is not a reliable estimate for every AI-enabled scenario.
Include human and infrastructure costs
Compare the subscription with hosting, monitoring, backups, upgrades, review and failure recovery. Self-hosted software can reduce some dependencies while adding operating work. No option is automatically cheapest for every workload.
Agree ownership, licensing and support
Choose an arrangement the business can understand and continue operating after the initial build.
Check the exact n8n arrangement
n8n’s official guidance distinguishes consulting on a client’s own internal instance from hosting client workflows and credentials in the provider’s instance, or embedding n8n in a product. Confirm the applicable licence with n8n for the intended setup.
Define the operational owner
For self-hosting, someone must own patching, backups and recovery. For a hosted platform, someone still owns credentials, business-rule changes and failed transactions. Buying a platform does not automatically buy those responsibilities.
Keep a practical handover
Document account access, workflow configuration, custom components, required licences and recovery steps. Agree what can be exported or transferred and what depends on a vendor-specific feature.
Questions worth asking
Does Canvo only use n8n?
No. The system is scoped around the required result. A connector platform, custom code or a combination may fit. We should be able to explain the choice in terms of your workflow and operating needs.
Should we move an existing working workflow?
Only if a concrete limitation justifies the migration. Compare the benefit with rebuilding mappings, validating old behaviour and changing the team’s recovery procedures.
Can we use AI with these kinds of workflows?
Yes, where supported by the chosen setup. Model calls do not remove the need for deterministic checks, permissions or result verification. Confirm the specific integration and usage charges before selection.
Does self-hosting mean everything stays on our server?
Not necessarily. Connected apps, external APIs and model providers may still receive data. Map the actual data path and selected services; the location of the workflow engine alone does not establish the whole system’s data boundary.
What happens if usage grows or a service reaches its limit?
Agree a workload and spending budget, then check the selected plan’s quota and overage behaviour. Use available usage alerts or limits, and define what pauses and who receives the alert. API rate limits and temporary failures need controlled waiting, bounded retries and a record of work still pending; uncertain writes must be checked before retrying. Queue retention and recovery vary by platform and setup, so test the chosen path rather than assume every delayed transaction will resume automatically.
Sources and further reading
Official documentation for the platform facts discussed in this guide. Business scenarios and recommendations are Canvo’s analysis.
Bring one process. We’ll map the next step.
Start with the input, the result your team needs and an exception that keeps coming back. Estimate the value, then check the assumptions with George.
12 months of support from acceptance covers defects in the agreed implementation and minor compatibility changes to existing integrations. Optional proactive care from A$490/month ex GST and further development are scoped separately. Third-party software, hosting and API costs are separate.