n8n vs Make vs Zapier for Client Deployments
Published October 3, 2026. Updated 2026-10-03. By StackScored.
Documentation review, checked October 3, 2026. Vendor capabilities below are attributed to their sources. Selection advice and suggested acceptance checks are StackScored's editorial judgment; we have not run a hands-on comparison.
For a client workflow, choose n8n, Make or Zapier by where it can run, who owns the connections and how the customer will recover a failure. Editor convenience matters after those requirements fit. The subscription you recommend must support the operating model you are selling.
Start with one representative workflow and list its required app actions, data destinations and human approvals. Name the customer's billing owner and operator. This comparison uses official documentation; the selection advice is our judgment. It includes no numeric prices because billing settings, usage and entitlements need a project-specific check.
Compare deployment boundaries first
| Tool | Observed source | Client selection question |
|---|---|---|
| n8n | Official plans list hosted and self-hosted options, including a self-hosted Community Edition. | Can the customer operate an instance, and does the applicable license permit the exact delivery model? |
| Make | On-premise agent documentation describes an Enterprise local-network connection feature. | Is connecting local systems to Make acceptable, or must the entire workflow runtime stay within customer infrastructure? |
| Zapier | Official plans describe the vendor's subscription service. | Does the purchased plan support the required collaboration and connections? This review establishes no self-hosted Zapier runtime option. |
The distinction matters if a requirement says “self-hosted.” A local connection agent and a customer-operated workflow service answer different questions. Draw the proposed data path and ask the customer to approve it before procurement. Do not infer data residency or a compliance guarantee from a product label.
Zapier's data privacy overview says it hosts data in the United States and does not support EU-only storage. If EU-only storage is mandatory, this documented boundary excludes that proposal. This is a residency constraint, not a conclusion that every European customer is prohibited from using Zapier. For every candidate, map workflow payloads, logs, backups and downstream services separately. Source checked October 3, 2026.
n8n: check the license before packaging a client service
The accessible n8n upstream LICENSE.md identifies Sustainable Use License restrictions for covered code and separate enterprise licensing for specified files. The covered license limits use to internal business purposes or non-commercial/personal use. This observation does not settle whether your consulting, embedding or multi-client hosting arrangement is permitted. Source checked October 3, 2026.
n8n's official use-case license guidance says consulting on clients' own internal instances does not require a commercial license on the consultant's part, while the client's use case still determines its licensing needs. Hosting clients' workflows and credentials in your own internal instance requires Enterprise; embedding workflows exposed to clients requires a white-labeled Embed license. Source checked October 3, 2026.
Those are the vendor's general categories, not approval of your contract. The help center explicitly cannot grant permission for a specific use case. Record instance ownership, who supplies credentials, who accesses the editor and what the customer buys. Resolve scope uncertainty through n8n's licensing or sales channel before committing to a service arrangement.
If you choose a self-hosted proposal, include maintenance and recovery in its scope. Identify the owner of the domain, instance, stored data and configuration. Ask how access survives your departure. An instance under your personal account leaves an unresolved operating dependency even when the workflow itself belongs to the client.
Make: local connectivity and recovery have conditions
Make's on-premise agent guide says the feature is available to Enterprise customers and currently supports the HTTP Agent application for connecting local systems. It does not document full self-hosting of Make. Confirm the required local API and Enterprise terms rather than assuming any local database can be connected in the same way. Source checked October 3, 2026.
Make's credential requests guide restricts that feature to official Make Partners and Enterprise customers with the required permissions and feature enabled. It supports requesting third-party connections through an authorization flow. An independent builder should not assume this workflow is included in every subscription. Source checked October 3, 2026.
Make's incomplete executions documentation says storage of unfinished runs is disabled by default. With it enabled, runs can be handled manually or retried for supported conditions; the stored allowance depends on usage. Buying Make does not by itself establish that your scenario's recovery path is configured. Source checked October 3, 2026.
Our recommendation is to make the recovery configuration part of customer acceptance. Choose an intentional failed test input, inspect what remains, and show the operator how to continue or abandon it. Agree which failures require investigation rather than automatic retry. Keep this exercise within a customer-approved test environment.
Zapier: sharing the workflow is only part of handover
Zapier's app connection guide documents sharing and transferring ownership on Team and Enterprise. Transferring a connection changes its owner, not the connected app account. Sharing a Zap without its connections does not let collaborators view or edit the affected step configuration. Source checked October 3, 2026.
In procurement, distinguish the Zapier owner from the identity authorized in the downstream application. Transferring one does not prove the other is customer-controlled. Ask the intended operator to inspect the connection and complete a test after access changes. Keep consultant removal separate from revoking a production connection the customer still needs.
Zapier's replay guide documents autoreplay for Professional, Team and Enterprise. It also distinguishes errored runs from safely halted runs: autoreplay does not replay the latter. Confirm the selected recovery setting and what the operator will see; do not equate an available retry feature with universal recovery. Source checked October 3, 2026.
Recover the business action before replaying the run
Suppose a test workflow creates a service ticket and then times out before recording the receipt. The platform's error status cannot establish whether the ticket exists. Look up the source event in the destination first. If the API supports an idempotency key, reuse the stable business-event key across retries and confirm its scope and expiration in that API's documentation. A new random key on every attempt defeats that protection.
If the destination cannot deduplicate, define a reconciliation procedure and an operator who can decide whether to resume, compensate or stop. Retrying a trigger and preventing repeated external actions are separate requirements. We make no exactly-once guarantee for any of these products. Capture both the platform run ID and the downstream receipt so recovery decisions can be reviewed.
Zapier's replay documentation warns that successful actions executed again can consume tasks again. Include the chosen replay boundary in both the recovery procedure and the cost model. Source checked October 3, 2026.
Deliver connections and recovery instructions with the workflow
Make's blueprint documentation says exported scenarios retain modules, settings and mapped values, but importing users must create their own account connections. A blueprint is useful configuration evidence; it is not a complete credential transfer or a tested restore. Source checked October 3, 2026.
Keep a customer-controlled inventory of platform accounts, downstream identities, OAuth scopes, scheduled triggers and webhook destinations. Store secret references rather than passwords in the handover record. Save the workflow version and environment settings with the agreed export. Reconnect in a test environment, check that a second scheduler cannot duplicate production work, then rehearse consultant access removal without revoking the client's needed identity.
For customer-operated infrastructure, name the backup owner, retention period, restore destination and upgrade procedure. For managed services, confirm which configuration and history can be exported, how long incident records remain available and who can contact support. Export coverage and successful restoration remain acceptance conditions until the customer operator demonstrates them.
Compare billing units with the same workflow
n8n's pricing FAQ defines an execution as a run of the entire workflow rather than each step. Make's pricing page uses credits and lists specific consumption rules, including a separate rule for its Code App. Neither definition makes the tools' usage counters interchangeable. Sources checked October 3, 2026.
Zapier's task usage guide counts successful action steps, including successful error-handler steps and previously successful steps executed again in a full replay. It also documents exceptions and search-action behavior. Use that guide for your workflow rather than assuming one incoming event always means one task. Source checked October 3, 2026.
Build a quote from the normal path, branches, repeated items and the expected recovery procedure. Use your own approved test observations to map those paths onto the vendor's unit. Record subscription charges separately from connected app fees, model calls, infrastructure and your support work. Compare the actual amount due and commitment period; do not label an annual monthly equivalent as monthly billing.
Leave unsupported costs unknown. This review does not establish the cheapest platform, a throughput benchmark or a savings forecast. It also does not reverify any stored StackScored pricing record.
A suggested acceptance exercise for the shortlist
Use a test destination and an input with a unique identifier. Observe a successful run, then deliberately exercise a failed connection and a failure after a downstream action. Ask the operator to establish whether the action occurred before deciding to replay. A retry control alone is insufficient evidence that a repeated business action will be prevented.
Repeat the exercise after the customer owns the relevant account and connection. Save the workflow version, source event, downstream receipt, failure state and recovery decision. These are recommended checks; StackScored has not executed them against these vendors. Actual API permissions, app-specific action coverage and your contractual operating rights remain requirements to verify.
Shortlist n8n when customer-controlled hosting is a requirement and licensing and operations are resolved. Shortlist Make when its scenario model fits and the required connectivity and recovery features are available. Shortlist Zapier when the required actions and the customer's connection-sharing plan fit. Those are conditional editorial recommendations, not a ranking or vendor endorsement.
Inspect the source-linked collection entries for n8n, Make and Zapier alongside these conditions.
Return to the client delivery toolkit guide for lifecycle selection. If your workflow uses an LLM, evaluate the separate quality and handover questions in Langfuse vs LangSmith.