How to Choose an n8n Agency: Nine Questions to Ask
Settle the hosting question first
Before you shortlist anyone, decide whether you are running n8n Cloud or self-hosting it. The answer rules out a large share of the market either way, and asking it in week three wastes everybody’s time.
The decision is mostly about licensing and controls rather than preference. n8n is fair-code, not open source: its Sustainable Use License lets you self-host and run unlimited workflows for your own internal business purposes, but not embed n8n in a product other people pay to use. Self-hosting the community edition costs nothing in license fees. What it does not include is single sign-on, git-based version control and separate development and production environments, all of which sit behind the paid Business tier on n8n’s published plans. Cloud plans are billed by workflow execution rather than by seat, starting at 20 euro a month for 2,500 executions.
Three practical consequences. If your security team requires SSO, “self-hosted and free” is not an option you actually have. If you need a real staging environment, you are on Business or you are building your own promotion process by hand. And if your workflows are high-frequency, execution-based billing is the number to model before you model agency fees, because it can be the larger line.
The nine questions
1. Do you host workflows yourselves, or only work inside n8n Cloud? A self-hosting agency owns Docker, queue mode, worker scaling, backups and upgrades. A Cloud-only agency is a workflow builder. Both are real businesses; only one can help when the instance falls over.
2. How do you handle credentials and secrets across environments? Listen for n8n’s credential store, environment variables and, on Enterprise, external secret store integration. The wrong answer is credentials pasted into nodes, or one shared admin account for every client.
3. What does your error handling actually look like? The specific things to hear: a configured error workflow, retry with backoff on transient failures, idempotency so a re-run does not create duplicate records, and an alert that reaches a human. If the answer is that the workflow will show a red node in the editor, nobody is watching it.
4. Can you show me a workflow you built more than a year ago that is still running? This is the highest-signal question on the list and the hardest to fake. It tests durability and it tests whether the client relationship survived the handover.
5. How do you version and deploy changes? Git-based source control on Business and above, or an explicit export and review process below it. “We edit the live workflow” is an answer, and it tells you what an incident will look like.
6. What happens when n8n ships a breaking change or a node deprecates? Ask who monitors releases, who pays for the fix, and whether that is written into the contract. The platform moves quickly and this is the most common source of surprise invoices in the second year.
7. Where does the model actually sit in the workflow, and what happens when it is wrong? On most builds a model handles one or two steps out of fifteen. Ask which ones, what the accuracy threshold is, how it was tested, and what the fallback path is. An agency that cannot answer the fallback question has not run this in production.
8. What do you hand over, and who owns it? Exported workflow JSON, documentation, credentials in your accounts rather than theirs, and a named internal owner. Get this in the statement of work rather than as a verbal assurance. An agency that resists a clean handover has a business model that depends on you not developing the capability.
9. Who is actually doing the work, and what happens if they leave? In a category with this many one-person and two-person firms, this is not a rude question. It is the main risk.
Reading the answers
Two patterns are worth naming.
The first is the rebuild tell. An agency that can only build cannot debug. Ask about a workflow somebody else wrote that they were asked to fix, and listen for whether the answer is diagnosis or “we rebuilt it properly.” Rebuilding is sometimes correct and it is always more billable, so the reflex is worth testing before you have paid for it.
The second is the specialist tell. Several agencies in our n8n agency ranking do genuinely excellent work with n8n as a minority line: Goodspeed’s own Clutch service breakdown puts low-code and no-code work at 10% of its mix, and at cimt ag and SVA the n8n practice sits beside far larger SAP, Camunda and enterprise integration businesses. That is not a disqualification. It is a question about who you get on the account and how quickly they answer when your instance is down.
The scale of the market you are choosing from
The published entry prices in this category run from $1,000 to $10,000, and most of those figures come from the n8n partner directory rather than the agencies’ own sites. AiAdvisors, Agent Studio, Width.ai, DataFix, ipanemads and PulpSense all show $1,000 floors. 2V Automation publishes a $2,000 Discovery Week on its own site. Goodspeed, InfoGrab, Digitalcube.ai and SVA start at $5,000. cimt ag, TrueHorizon AI and AOE Group start at $10,000. Bitovi carries the highest published hourly rate in the category at $150 to $199 on Clutch.
Team size matters more here than in most categories, because the extremes are extreme. AiAdvisors, ranked second, is one named individual with no published backup or continuity plan. SVA is a specialist practice inside a 3,800-person system integrator. There is no correct answer, only the question of which failure mode you can live with: no bench, or being a minor account.
When not to hire an n8n agency at all
Three cases, and they cover more buyers than the category likes to admit.
You have one workflow across tools with native connectors. This is a genuine afternoon of work for a capable operations person, and n8n’s own templates and documentation are good. Paying an agency four figures for it buys documentation and a dependency.
Nobody internally will own it. An n8n estate with no owner degrades silently. Integrations change, credentials expire, a node deprecates, and nothing announces it. If you cannot name the person, buy managed operations explicitly or do not start.
The requirement is really a product. If what you want is a customer-facing application with a UI, accounts and a release cycle, n8n is the wrong shape and an n8n agency is the wrong vendor. Look at the AI agent development companies instead. Workflow orchestration is a superb backend for internal process and a poor foundation for a product other people log into.
Our n8n agency ranking notes which firms handle self-hosted deployments, what each publishes about price, and at least one honest drawback on every listing. The criteria behind the order are on the methodology page.