n8n agency
n8n agency: what the work is, what it costs, and when n8n is the wrong platform
Short answer
An n8n agency designs, builds, and maintains automation workflows on n8n rather than selling you a platform. The work comes in four shapes: designing and building workflows, setting up and hardening a self-hosted instance, rescuing a half-built estate someone left behind, and a retainer that keeps live workflows running as your tools change. Hire one when your automations cross several systems, run at volume, or already exist and nobody owns them. If your automation is two steps and low volume, buy a Zapier subscription instead — an agency cannot make that cheaper, and a good one will tell you so.
- Four service shapes: build, self-host setup, rescue of inherited workflows, retainer maintenance
- We publish when Make or Zapier wins instead — we take no commission from any of the three
- Every workflow ships with an error path, a named owner, and a runbook someone else can read
- Your instance, your credentials, your workflow JSON — exported and handed over, always
Typical clients: 5–100 person teams already running n8n, or moving to it from Zapier or Make because per-step billing has stopped making sense.
How we decide whether n8n is the right platform — including when it is not
A page that says n8n is always the answer is an advertisement. We build on n8n most of the time because most of our work is multi-step and high-volume, but the platform decision follows the workflow. These are the tests we run before quoting, and the three cases where we hand you to a cheaper tool.
- 01Steps × runs: count the steps in the workflow and multiply by monthly runs. n8n bills per execution, so step count is free; Make bills per operation and Zapier bills per task, so both bill roughly that product. Above about 10,000 the arithmetic decides on its own.
- 02When we recommend Zapier instead: the automation is two or three linear steps, volume is low, nobody on the team is technical, and the person who will own it needs to read and edit it themselves. Zapier gets that working in an afternoon and hands over cleanly. Paying us to build it on n8n would be worse for you.
- 03When we recommend Make instead: the workflow branches and loops but still needs to stay inside a visual canvas that a marketing or ops hire can maintain without us. Make handles arrays, iterators, and conditional routing well, its per-operation pricing is usually cheaper than Zapier for the same work, and at small-business volume the per-step penalty is a rounding error.
- 04When we recommend no platform at all: one scheduled job moving data between two systems you already control is cheaper and less fragile as a small script on a cron schedule. Automation platforms earn their subscription by handling many connectors and retries, not by running one function an hour.
- 05Connector reality: every system in the chain has to have an n8n node, an API, an export, or a mailbox. Zapier still has the widest catalogue, and a missing connector outranks every preference above — including ours.
- 06Ownership after handover: someone has to be able to open the workflow in six months. If nobody on your side will ever read a workflow graph, n8n is the least forgiving of the three and we will say so before you commit.
- 07Self-host only for a reason: data residency, a private network, or execution volume high enough that server cost beats subscription cost. Otherwise n8n Cloud is the right default, because self-hosting costs a server plus somebody's attention for upgrades, backups, and the morning it stops.
- 08No commissions: we hold no partnership, reseller, or affiliate arrangement with n8n, Make, or Zapier. Revenue comes only from work delivered, which is why the audit is allowed to end in "keep your Zapier subscription".
Why companies come to an n8n agency
Almost never "we would like some automation". It is one of four specific situations, and the entry point changes what the first engagement looks like.
Forty workflows and nobody who can read them
The most common reason anyone hires an n8n agency. An ops person or a technical marketer built a working estate, then left. Nothing is documented, several workflows are near-duplicates, a few are silently failing, and one is doing something important that nobody can explain. This is a rescue job, and it starts with an inventory rather than a build — you cannot fix what you cannot list.
Marketing ops: where the tangle usually starts
Marketing accumulates automations faster than any other function, which is why "n8n marketing agency" is searched more than "n8n agency" itself. Lead enrichment loops over every record, campaign reporting fans out across five platforms, inbound replies get classified one at a time — the workflows with the most steps and the highest volume in the whole company. That is exactly where per-task billing hurts and where n8n's per-execution model pays for the migration. It is also the function most likely to have built it all on a growth hire's personal account.
Self-hosted once, then never touched
Someone put the Community edition on a small server because it was free, and it has been running unpatched for a year with no backup of the credentials store and no upgrade path. Free is a licence fee, not a total cost. The bill arrives as a weekend outage, or as a workflow quietly failing after a node changed behaviour three releases ago.
It works until the data is wrong
A workflow that handles the clean cases and breaks on the rest creates more work than it removes, because now somebody has to notice. Error branches, retries, deduplication, and a failure path that notifies a named person are the majority of the work in any real automation — and the part DIY builds skip, because on a per-step platform every one of those nodes costs money.
How an engagement runs
Audit first, whether you are building from scratch or inheriting a mess. You can stop after the audit and take the findings to your own team or to another agency.
- 01
Inventory or intake
If n8n is already running, we list every workflow: what triggers it, what it touches, how often it fires, whether it is failing, and whether another workflow already does the same thing. If you are starting fresh, the same session maps the process by hand instead — who touches the work, which systems it crosses, how often it repeats.
You get A written inventory of every workflow or candidate process, with run frequency and current state.
- 02
Cost it and pick the platform
Steps × runs for each workflow, priced against how each platform bills. This is the step where we tell you that three of your workflows belong on n8n, one should stay on Zapier because a non-technical colleague owns it, and one should be deleted because a report nobody reads depends on it.
You get Cost model per workflow, a platform recommendation for each, and a ranked list of what to fix first.
- 03
Build or rebuild
One workflow end to end: triggers, branching, the code steps where the visual builder runs out, retries with backoff, deduplication, and an error branch that notifies a person. It runs alongside whatever it replaces until the two agree on real data.
You get A live workflow, its exported JSON, and a measured before-and-after.
- 04
Harden the instance
For self-hosted: a reproducible deploy, queue mode if volume needs it, credentials out of the workflow bodies, encrypted backups that have actually been restored once, and a written upgrade procedure. For n8n Cloud: environments, access control, and the same credential hygiene. Skipped for nobody — an unhardened instance is where the next incident comes from.
You get Deploy configuration, backup and restore procedure, and an upgrade runbook.
- 05
Hand over, or keep us on
You get the workflow JSON, the credentials, the runbook, and a session walking your team through it. Some clients take it in-house from there — we will train the person who inherits it. Others keep us on a retainer to watch the live workflows and add the next one.
You get Runbook, admin access, a training session, and a decision on what happens next.
The arithmetic that decides n8n versus everything else
A model, not a case study — the numbers below are illustrative and the only inputs are yours. Substitute your step count and your monthly runs and re-run it in a minute. This is the same calculation our audit applies to every workflow, and it is why some of them stay on Zapier.
One workflow with 12 steps that do work — a webhook, a lookup, an enrichment call, a language-model classification, a branch, two writes, a notification, and the validation and formatting nodes between them. It fires 3,000 times a month. n8n bills per workflow execution; Make bills per operation and Zapier bills per task, and both of those mean per step.
- Steps in the workflow
- 12
- Times it runs per month
- 3,000
- n8n billable units — executions, step count free
- 3,000
- Zapier billable units — tasks = 12 × 3,000
- 36,000
- Make billable units — operations = 12 × 3,000
- 36,000
- Ratio, per-step billing vs per-execution
- 12×
- Error branch, retry, dedupe, and logging nodes we would add
- +5 steps
- n8n after those five nodes
- 3,000 — unchanged
- Zapier after those five nodes — 17 × 3,000
- 51,000 (+42%)
- Same workflow at 40 runs a month instead — Zapier tasks
- 480
- Vendor price per unit, to look up today
- all three change tiers — check before you commit
Two things fall out of this. First, the crossover: at 3,000 runs a per-step platform bills you for twelve times as many units, and every node you add to make the workflow debuggable makes that worse — which is precisely why DIY builds on Zapier have no error handling. At 40 runs a month the same workflow is 480 tasks, comfortably inside a cheap paid tier, and hiring anyone to move it to n8n would be a waste of your money. Second, the incentive: on n8n those five reliability nodes are free, so the workflow we build is the one we actually wanted. Multiply your own steps by your own runs. Above roughly 10,000, price all three properly. Below it, pick on who can maintain it — and that answer is often not n8n.
What we build on n8n, and what we will not sell you
Being explicit about the boundary is the fastest way to tell an implementation team from a reseller with a partner badge.
In scope
- Workflow design and build: triggers, branching, loops, retries with backoff, deduplication, and an error path that notifies a named person
- Marketing ops builds — lead enrichment and scoring, CRM hygiene and deduplication, form-to-CRM-to-sequence routing, campaign reporting rollups across ad and email platforms, inbound reply classification and triage, review and content pipelines
- Self-hosted setup and hardening: reproducible Docker or Kubernetes deploy, queue mode for volume, credential management, encrypted and tested backups, and a written upgrade procedure
- Rescue of an inherited estate: full inventory, deduplication, documentation, migration off personal accounts, and deletion of what turns out to be dead
- Migration of specific workflows off Zapier or Make where the steps × runs arithmetic justifies rebuilding them — workflow by workflow, never as a big-bang cutover
- Custom nodes and code steps for the logic no visual builder can express, plus API integrations where no node exists
- Language-model steps inside workflows: classification, extraction, drafting, summarising, with the prompt and its evaluation documented
- Monitoring, alerting, and a runbook written for whoever inherits the instance — including a training session for them
Not in scope
- Reselling n8n licences, Cloud seats, or anything else — we take no vendor commission and hold no partner agreement
- Building on n8n when the arithmetic says Make or Zapier. We quote the honest option or we quote nothing.
- Migrating your entire Zapier or Make account to n8n as a project in its own right — there is no export path, every workflow is a rebuild, and most estates contain workflows not worth rebuilding at all
- Running your production instance on our infrastructure as a black box you cannot leave
- Marketing strategy, campaign management, or creative work. We build the plumbing underneath a marketing team, not the campaigns.
- Automating a process nobody has written down — we document it or flag it first, because automating a moving target produces rework
- Model training or data science projects that need a team you do not have
What it costs — and what the same work costs elsewhere
Published in full, next to the market rate, because the first thing anyone shopping for an n8n consultant or an n8n developer for hire wants to know is whether this is a four-figure engagement or a six-figure one. We are deliberately at the bottom of the market, and the reasons are listed under the table rather than left for you to guess at.
Audit
$490
Typical market rate: $2,500–8,000
You inherited workflows nobody understands, or you are about to commit to n8n and want the arithmetic checked first.
- Inventory of every existing workflow, or a map of the processes you want built
- Steps × runs cost model per workflow, priced across n8n, Make, and Zapier
- A platform recommendation per workflow — including the ones that should not move to n8n
- Ranked fix list with build estimates, and instance health notes if you self-host
- Credited in full against your first build
First build
from $1,900
Typical market rate: $6,000–15,000
The workflow is picked and you want it running, instrumented, and documented.
- One workflow end to end, with error branches and retries
- Self-host deploy or Cloud setup hardened as part of the build
- Exported workflow JSON, runbook, and a handover session
- Two weeks of post-launch tuning on real data
Retainer
$690 / month
Typical market rate: $3,000–8,000 / month
Workflows are live and you want them watched, upgraded, and extended.
- Monitoring, alert triage, and incident response
- n8n version upgrades and node deprecation fixes
- One new or reworked workflow per month
- Cancel with 30 days notice — nothing is held hostage
The build price is a floor, not an average. A workflow crossing two systems with clean data and an existing node for each lands near $1,900; one crossing five systems with messy source documents, no node for the important one, and a self-host deploy to stand up first lands higher. Those variables move a quote more than anything else, and the rescue of an inherited estate is quoted from the audit findings rather than blind. You get a fixed number in writing after the intake call, before any work starts, and the $490 audit comes off it in full. There is no free pilot — the audit is the cheap way to find out whether this is worth doing.
Why this is a fraction of the agency price
Below-market pricing usually means junior people learning n8n on your instance. That is not what is happening here, so it is worth naming exactly where the difference comes from.
You talk to the builder
The person on the first call is the person who will open your instance. No account manager, no pre-sales engineer, nobody between you and the work whose salary has to come out of your invoice.
The audit is a product, not a project
Same inventory script, same steps × runs cost model, same deliverable every time. Agencies bill discovery as bespoke work because for them it is. The template already exists here, so you are not paying for its invention.
We reuse the node patterns
Error-branch-and-retry, idempotent writes, paginated fetch, rate-limit backoff, CRM upsert with dedupe — we have built these many times and keep them as patterns. Your process is specific; the plumbing under it almost never is, and we price the plumbing as the repeat work it is.
No licence margin, no commission
We take no commission from n8n and resell nothing — no partner programme, no affiliate link, no reseller margin on Cloud seats. Revenue comes only from work delivered, which is exactly why the audit is free to conclude that you should stay on Make.
n8n agency, in-house n8n developer, or a generalist automation agency
All three work, and they fail in different places. The honest answer depends on how much recurring n8n work you genuinely have twelve months from now.
| Criterion | n8n agency | In-house n8n developer | Generalist automation agency |
|---|---|---|---|
| Time to first working workflow | 2–5 weeks | 3–5 months including hiring | 4–8 weeks |
| Cost shape | Fixed audit, then per build, retainer optional | Salary, plus hosting and management overhead | Monthly minimum regardless of what ships |
| Platform judgement | Deep on n8n, and published about when Make or Zapier wins | Deep on n8n once they have learned your stack — on your time | Broad across platforms, sometimes shaped by reseller agreements |
| Best when | You have a burst of build or rescue work and want it owned afterwards | n8n is core to your product, not your back office | You need continuous multi-channel output more than platform depth |
| Main risk | Engagement ends before your team can maintain it — mitigated by the runbook and training session | One person becomes a single point of failure, and n8n knowledge leaves with them | Nobody goes deep enough on the instance to harden it |
| Who owns the instance | You — workflow JSON and credentials handed over | You | Sometimes the agency account |
What you actually need
Read down to the line that matches you. Three of the seven answers do not involve hiring us, and one of them is "buy a subscription and build it yourself this afternoon".
Your automation is two steps — a form fills, a row appears, someone gets notified
Buy Zapier and build it yourself this afternoon. It will be working in fifteen minutes, a non-technical colleague will be able to edit it, and no agency can make that cheaper. Hiring anyone here is the expensive mistake.
You do not know your monthly volume yet
Start on Make's free tier and instrument it for a month. Guessing volume is how companies end up on the wrong pricing model, and the measurement costs you nothing. Come back when you have a real number — or do not, if the number turns out to be small.
It is one scheduled job between two systems you already control
Write a small script on a cron schedule. Cheaper, faster, and less fragile than any of the three platforms. You do not have an automation-platform problem.
Someone built a pile of workflows and left, and nobody can explain what is running
Start with the audit. The inventory usually finds duplicates, silent failures, and at least one workflow safe to delete. Rebuilding before you have the list is how the tangle gets bigger.
Marketing ops volume has made the per-task bill conspicuous
Run the steps × runs arithmetic above on your three biggest workflows. If the product clears roughly 10,000 a month, an n8n rebuild pays for itself and the audit will show in which order. If it does not, stay where you are.
Data cannot leave your infrastructure
Self-hosted n8n, and it is genuinely the only one of the three with a serious self-hosting story. This is the build-plus-hardening engagement: deploy, credential management, tested backups, and an upgrade runbook, because the free licence is not a free instance.
n8n will be permanent, central, and busy — part of what you sell
Hire in-house. Consultants are the wrong structure for work that never stops. We can stand the instance up, write the patterns, and train your hire, then get out of the way.
A one-week n8n test you can run before hiring anyone
This is the first half of our audit, done by hand. It costs nothing and it answers the platform question honestly. If it tells you to stay on Zapier, you have saved yourself a fee, and that is a legitimate outcome.
- Day 1Write your most annoying process down step by step, including what happens when a field is missing or malformed. Count the steps that do work. That number is your billing multiplier on Make and Zapier, and free on n8n.
- Day 2Get the real monthly volume from the source system — search the inbox, count the CRM records, read the platform's own task counter. Do not estimate; estimates are wrong in the direction that flatters the decision you already made.
- Day 3Multiply steps × runs. Above roughly 10,000 a month, price all three platforms properly before choosing. Below it, choose on who can maintain it and stop reading about pricing models.
- Day 4Check every system in the chain against n8n's node list. A missing node means an HTTP request node and API docs, which is fine if someone will write it and a hard blocker if nobody will. Zapier's catalogue is still the widest — if that is where your long-tail tool lives, that fact outranks everything else here.
- Day 5Build the happy path. Run n8n locally with Docker — the Community edition is free under the Sustainable Use License for internal business use — or use whatever hosted trial n8n offers when you read this. An afternoon is enough to get the clean case working.
- Day 6Break it on purpose. Feed it a malformed record, a missing field, a duplicate, and an API that times out. How the platform surfaces that failure — and how much work the error branch turns out to be — is what you will live with daily. This is the day most DIY estimates double.
- Day 7Show the workflow to whoever will own it after you. If they cannot describe what it does from looking at the canvas, you have picked the wrong tool for your team, whatever the arithmetic said on day 3.
If day 3 clears 10,000 units and day 6 showed you more error handling than you want to build, an n8n build pays for itself and it is worth a conversation with someone — us or anyone else. If day 3 came in low, or day 7 found nobody who can read a workflow graph, stay on the tool you have. That is the recommendation we would give you for $490, so take it for free.
Looking for something more specific?
- AI automation consultantYou have not picked a platform yet and want the processes mapped and costed before anyone commits to n8n.
- AI automation servicesYou know what to build, the platform is not the question, and you want the implementation team.
- AI consultant for small businessUnder 50 people, no IT department, and this would be your first automation project of any kind.
Frequently asked questions
How much does it cost to hire an n8n developer?
Our audit is $490 and is credited in full against the first build; a workflow build starts at $1,900; ongoing maintenance is $690 a month, cancellable with 30 days notice. For comparison, agencies typically charge $2,500–8,000 for the audit alone and $6,000–15,000 per build. An independent n8n consultant or freelance n8n expert usually quotes hourly, which is fine for a well-specified single workflow and the arrangement most likely to go wrong when the scope is "clean up whatever is in there". We quote a fixed number in writing after the intake call.
Can you take over workflows someone else built?
Yes, and it is the most common way clients arrive. The first step is an inventory rather than a rebuild: every workflow listed with its trigger, the systems it touches, its run frequency, and whether it is currently failing. That exercise reliably turns up near-duplicates, workflows running on a former employee's personal credentials, and at least one that can simply be deleted. Rebuilding before you have the list is how an inherited tangle gets bigger instead of smaller.
Should I self-host n8n or use n8n Cloud?
Cloud unless you have a specific reason not to. Self-host for data residency, a private network, or execution volume high enough that server cost beats subscription cost. Self-hosting is not free even though the Community edition licence is: you pay for a server, backups you have actually tested, version upgrades, and somebody's attention the morning it stops. If nobody on your team owns servers, that cost lands on whoever is nearest, which is usually the person who least wants it.
Do you work as an n8n marketing agency?
We build the automation layer under a marketing team — lead enrichment and scoring, CRM deduplication, form-to-sequence routing, campaign reporting rollups across ad and email platforms, inbound reply classification. Those are the highest-step, highest-volume workflows in most companies, which is why n8n's per-execution billing matters most there. What we do not do is strategy, media buying, or creative. If you need campaigns run rather than plumbing built, you want a marketing agency, not us.
Is n8n free, and what does the licence actually allow?
The self-hosted Community edition is free to run under n8n's Sustainable Use License, which permits internal business use but restricts reselling it as a service. You pay for the server, the maintenance, and the upgrades instead of a subscription. n8n Cloud is the paid hosted version and starts in the low tens of euros per month. Check the current terms and tiers before committing — n8n changes both, and so do Make and Zapier.
What if Make or Zapier is actually the better fit for us?
Then we say so, in the audit, and we do not quote a build. Zapier wins for two- or three-step linear automations at low volume with a non-technical owner. Make wins where the workflow branches and loops but has to stay maintainable inside a visual canvas by someone who is not us. We hold no partnership or affiliate arrangement with any of the three, so there is no commission riding on the recommendation.
Should we migrate everything off Zapier to n8n?
Almost never all at once. There is no meaningful export path between the platforms — every workflow is a rebuild, so the migration cost is roughly the cost of building it again. We migrate workflow by workflow, in the order the steps × runs arithmetic says, and we leave behind the ones where Zapier is genuinely the right home. A big-bang cutover pays to rebuild workflows that should have been deleted.
Who owns the workflows and the instance when we stop working together?
You do. The instance lives in your account or on your infrastructure, credentials are yours, and you receive the exported workflow JSON plus a runbook written for whoever inherits it. We also run a handover session with that person. Nothing is held behind a retainer, and the retainer is cancellable with 30 days notice precisely so it does not have to be.
How long until the first n8n workflow is live?
The audit takes about two weeks. A first workflow typically goes live two to four weeks after that. Access delays, not build time, are the usual bottleneck — API keys, admin seats, and whoever has to approve them. Starting the credential requests during the audit is the single best thing a client can do to shorten the timeline. A self-host deploy adds roughly a week if the server does not exist yet.
Next step
Start with the n8n audit
Tell us what is running, or what you want built. We come back with the steps × runs arithmetic, which platform each workflow belongs on, and what a build would cost — including when the answer is to leave things where they are.