n8n Build Cost: 5 Real Project Scopes, Itemised and Priced (2026)

TL;DR: A single-trigger lead intake workflow ships for around $750. A scheduled sync with pagination and a cursor is around $1,200. An ops dashboard feed pulling three systems is around $1,800. Document processing with AI extraction and a human review step is around $2,200. A multi-step AI agent with tool sub-workflows is around $2,900. Almost every n8n project lands between $500 and $3,000 as a fixed price. What moves you inside that range is not the number of nodes. It is the number of failure modes you have to handle, and whether a human has to approve anything.

Most articles about hiring an n8n developer promise pricing and then give you an hourly range. An hourly range tells you nothing, because it does not tell you how many hours. So here are five builds we actually scope, itemised.

What does an n8n build actually cost?

Between $500 and $3,000 as a fixed price for the overwhelming majority of projects. The number is set by scope, not by hours, so you know it before work starts. Below $500 the workflow is usually a template you should install yourself. Above $3,000 it stops being a project and becomes a retainer.

The hours are published so you can see what you are buying, not so you can back into a rate. We do not bill hourly. The price does not change if a build runs long, which is the entire point of quoting fixed.

One thing to notice before the detail: build hours roughly triple across the table, but testing and documentation hours also triple. That ratio holds because the thing that makes a workflow expensive is the number of ways it can quietly go wrong.

Scope 1: What does a lead intake webhook to CRM cost?

$750. A form or landing page posts to an n8n webhook, the payload is normalised and spam-checked, the contact is deduplicated against the CRM, created or updated, and the team gets a Slack ping. It runs unattended and it is the cheapest build worth paying for.

• Nodes: 11 in the main workflow, plus a 3-node error workflow.

• Integrations: an inbound webhook (Webflow, Typeform, a custom form), one CRM (HubSpot or Pipedrive), Slack.

• Build: 6 hours.

• Testing and documentation: 3 hours, including replaying real malformed payloads, not just the happy path.

• Maintenance: roughly an hour a quarter. Form fields change; CRM APIs rarely do.

The dedupe step is where these go wrong. A naive create-on-every-submission workflow will fill your CRM with duplicates within a month, and cleaning that up costs more than the build did. If you want the trigger mechanics, the webhook node reference covers auth and response modes.

What bumps it: routing to more than one CRM or brand, round-robin owner assignment, consent and lawful-basis fields that have to be mapped correctly, or an enrichment call that needs its own retry and budget logic. Any two of those and you are at Scope 2 pricing.

Scope 2: What does a scheduled data sync cost?

$1,200. A schedule trigger pulls records from a source API on a cursor, pages through them in batches, transforms them, and upserts into Postgres or Sheets, with a daily summary and a failure alert. The build hours are modest. The testing hours are not, because a sync that half-works is worse than one that fails.

• Nodes: 16, including a Split In Batches loop and a cursor read/write pair.

• Integrations: one source SaaS REST API, one destination (Postgres or Google Sheets), Slack for alerting.

• Build: 10 hours.

• Testing and documentation: 5 hours, including one full historical backfill run and a deliberate mid-run failure to prove the cursor resumes rather than restarts.

• Maintenance: 1-2 hours a month. Source APIs deprecate endpoints, and rate limits change without notice.

Pagination is the whole job. Most of the ten build hours go into handling the source API's specific paging contract, which is why the HTTP Request node patterns matter more here than any CRM node does.

What bumps it: a second source system, or two-way sync. Two-way is not twice the work, it is roughly three times, because you now need conflict resolution and a rule for which side wins. That build lands around $1,900.

Scope 3: What does an internal ops dashboard feed cost?

$1,800. Three source systems are pulled on a schedule, reconciled into a single metrics table, and pushed to a BI tool with a daily Slack digest. You are not paying for the dashboard. You are paying for the metric definitions to be right and for the numbers to survive an audit.

• Nodes: 19, split across one main workflow and two sub-workflows so the source pulls can be tested and rerun in isolation.

• Integrations: typically a billing system, a CRM, and the product database, writing to Postgres and Slack.

• Build: 13 hours.

• Testing and documentation: 6 hours, and this is the line that surprises people. Most of it is reconciling one full month of the automated numbers against whatever the finance or ops person currently produces by hand, and explaining every discrepancy.

• Maintenance: 1-2 hours a month.

These projects fail on definitions, not on code. If nobody has decided whether a refunded order counts as revenue in the month it was placed or the month it was refunded, no workflow can be correct, and the reconciliation step turns into a series of meetings. Agree the definitions before the build starts and this scope stays at $1,800.

What bumps it: a fourth and fifth source system, a backfill beyond twelve months, or arriving without agreed metric definitions. Any of those and it is a $2,500 build.

Scope 4: What does a document processing pipeline with AI extraction cost?

$2,200. Documents arrive by email or land in a Drive folder, text is extracted, an LLM returns structured fields against a fixed schema, low-confidence results are routed to a human for approval in Slack, and approved records are written into the accounting or ERP system. The human review step is not optional and it is a third of the cost.

• Nodes: 24, including the confidence branch and the approval loop.

• Integrations: Gmail or Drive as the trigger, an LLM provider, the target finance system, Slack for the review queue.

• Build: 16 hours.

• Testing and documentation: 8 hours. This includes hand-labelling a set of real documents and measuring extraction accuracy against them before go-live. Without a labelled set you have no idea whether the pipeline works, you only know that it runs.

• Maintenance: 2-3 hours a month, and it genuinely needs the attention. Model behaviour drifts, vendors change invoice layouts, and a silent extraction error writes a wrong number into your books.

The naive version of this build costs about half as much and skips the confidence check. Do not buy that version. An extraction pipeline with no human in the loop and no accuracy baseline is a machine for producing confidently wrong accounting entries.

What bumps it: more than two document layouts or languages, scanned documents that need OCR before extraction, or an approval interface that has to be something other than Slack. Those push it to the $3,000 ceiling.

Scope 5: What does a multi-step AI agent workflow cost?

$2,900. An agent that classifies an incoming request, retrieves context from a vector store, calls two or three tools to look things up, drafts an action, and escalates to a human when it is uncertain. This is the top of the fixed-price band and the only scope here where testing hours approach build hours.

• Nodes: around 30, counting the agent node itself plus four tool sub-workflows.

• Integrations: a helpdesk or inbox, a vector store, one or two systems of record, Slack for escalation.

• Build: 20 hours.

• Testing and documentation: 10 hours. The deliverable includes a golden set of roughly 40 real cases with expected outcomes, so that a prompt or model change can be regression-tested rather than eyeballed.

• Maintenance: 3-4 hours a month. Agents are the only n8n build that degrades without anyone touching them, because the model underneath changes.

Read the AI agent node reference and the agentic patterns before you specify one of these. A large share of "agent" requests turn out to be a deterministic workflow with an LLM classification step in the middle, which is Scope 1 or 2 pricing and a far more reliable system. Say so out loud and save yourself two thousand dollars.

What bumps it past $3,000: an agent that writes to customer-facing systems without human approval, multi-agent handoff, or an evaluation harness that runs in CI on every change. At that point it is not a project.

What pushes a build past $3,000?

Three things, and none of them are node count. Ongoing responsibility for a running system, an SLA on response time, and a scope that keeps growing because the business is still deciding what it wants. Those are retainer conditions, not project conditions, and quoting them fixed would be dishonest.

When a build hits any of those, it moves onto a retainer from $3,000 a month, or a fully managed arrangement from $8,000 a month, which is where on-call and continuous change belong. That is a different commercial shape and we say so rather than quietly inflating a project quote to cover risk we have not named.

The practical test: if the thing you want has a definition of done, it is a project and it fits in the band above. If what you actually want is somebody to own automation as a function, no fixed price can be honest, because you are not buying a deliverable.

Which inputs actually move the price?

Six inputs, in descending order of impact: whether a human has to approve anything, how many systems of record are written to, whether the data is personal data, whether an LLM is in the decision path, whether historical backfill is needed, and how many failure modes need explicit handling. Node count is not on the list.

Fill this in and you can place your own build in the table above without talking to anyone.

n8n BUILD SCOPING TEMPLATE
(copy, fill in, and the tier falls out of the answers)

1. TRIGGER
   [ ] Webhook (real-time, one source)
   [ ] Schedule (batch)
   [ ] App event / polling
   How many distinct triggers? ......

2. SYSTEMS
   Systems READ from:  ...................  (count: ....)
   Systems WRITTEN to: ...................  (count: ....)
   Any system without an existing n8n node? ...... (yes = custom code)

3. DECISION PATH
   [ ] Deterministic rules only
   [ ] LLM classification / extraction with a fixed output schema
   [ ] Agent that chooses which tools to call

4. HUMAN IN THE LOOP
   [ ] None. Fully unattended.
   [ ] Approval step before anything is written
   [ ] Exception queue only (approve on low confidence)

5. DATA
   [ ] No personal data
   [ ] Personal data (DPA required, retention rules apply)
   Historical backfill needed? ...... How far back? ......

6. FAILURE HANDLING (tick everything the build must survive)
   [ ] Source API rate limit / 429
   [ ] Partial run: must resume, not restart
   [ ] Duplicate records on retry
   [ ] Silent stop: nobody notices for two weeks
   [ ] Wrong-but-plausible output written downstream

7. DEFINITION OF DONE
   Write it in one sentence. If you cannot, the scope is not ready to price:
   ......................................................................

SCORING
  One trigger, one system written to, deterministic, no human step
      -> $500-$900
  Scheduled, pagination, cursor, one destination
      -> $1,000-$1,400
  Three or more systems read, reconciliation, agreed metric definitions
      -> $1,600-$2,000
  LLM in the decision path + a human approval or exception queue
      -> $2,000-$2,500
  Agent choosing tools + a regression set of real cases
      -> $2,500-$3,000
  Ongoing ownership, on-call, or an SLA
      -> not a project. Retainer from $3,000/month.

If more than three boxes are ticked in section 6, add a tier. That section is where the real money is, and it is the section clients skip.

What is not in these prices?

Platform cost, LLM tokens, and your team's time. The fixed price buys the build, the tests, and the documentation. It does not buy the server n8n runs on, the Cloud subscription, the model usage an extraction pipeline burns through, or the hours your ops person spends approving exceptions.

Platform and licence cost is a separate calculation and we have done the arithmetic elsewhere: n8n Cloud vs self-hosted has the real numbers, including the Postgres growth that catches people out. Budget the LLM spend separately too. On Scope 4 and Scope 5 it is a recurring operating cost, not a one-off, and it scales with volume in a way the build does not.

Also not included, and worth naming: nothing above assumes you already have error handling. If you have existing workflows with no error workflow, no alerting, and no execution pruning, that is a separate remediation job. The error handling guide will tell you whether you need one.

Getting a number for your own build

Take the scoping template above, fill it in, and you will already know which row of the table you are in. If it comes back somewhere between two tiers, it is almost always section 6 deciding it.

If you want the scope written down properly and priced, book a 20-minute call. We will tell you which tier it is on the call, and if what you are describing is really a template you should install yourself, we will tell you that instead. Who you buy it from is a separate question, and the delivery model comparison and the vetting checklist cover it better than a sales page would.

FAQ

How much does it cost to build an n8n workflow?

Most n8n projects land between $500 and $3,000 as a fixed price. A single-trigger lead intake workflow is around $750; a multi-step AI agent with tool sub-workflows is around $2,900. The variable is not node count, it is how many failure modes and approval steps the build has to handle.

Is n8n development priced hourly or fixed?

We price fixed, per scope, agreed before work starts. Published market hourly rates for n8n freelancers run roughly $30-$100/hour, but an hourly rate without an hour estimate is not a price. Fixed pricing puts the estimation risk on the builder, which is where it belongs.

Why is an AI agent workflow more expensive than a data sync with more integrations?

Because it can be wrong in ways that look right. A sync either moves the record or it does not. An agent can produce a plausible, confident, incorrect action, so the build needs a regression set of real cases and an escalation path, which is where the extra ten testing hours go.

What makes an n8n project cost more than $3,000?

Ongoing ownership rather than a deliverable: an SLA, on-call, or a scope that keeps expanding. Those are retainer conditions and move to a retainer from $3,000/month or a managed arrangement from $8,000/month. A one-off build with a clear definition of done stays inside the project band.

Does the price include hosting and running costs?

No. The fixed price covers the build, testing, and documentation. Your n8n instance (Cloud subscription or a self-hosted VPS plus Postgres) and any LLM token spend are separate recurring costs. See the Cloud vs self-hosted comparison for the platform arithmetic.

Can I just buy a template instead?

Often, yes. If your build is one trigger, one destination, no personal data, and no approval step, a free or paid template will get you most of the way and a paid build is poor value. The scoping template above is deliberately designed to tell you when that is the case.